diff --git a/.github/dependabot.yml b/.github/dependabot.yml index a32285d4..899ca270 100644 --- a/.github/dependabot.yml +++ b/.github/dependabot.yml @@ -1,39 +1,40 @@ version: 2 updates: # Enable version updates for npm - - package-ecosystem: "npm" - directory: "/" + - package-ecosystem: 'npm' + directory: '/' + target-branch: 'dev' schedule: - interval: "weekly" - day: "monday" - time: "09:00" - timezone: "Europe/Oslo" + interval: 'weekly' + day: 'monday' + time: '09:00' + timezone: 'Europe/Oslo' # Limit number of open PRs open-pull-requests-limit: 10 # Group minor and patch updates groups: development-dependencies: - dependency-type: "development" + dependency-type: 'development' update-types: - - "minor" - - "patch" + - 'minor' + - 'patch' production-dependencies: - dependency-type: "production" + dependency-type: 'production' update-types: - - "minor" - - "patch" + - 'minor' + - 'patch' # Auto-approve and merge patch updates for dev dependencies # (Requires GitHub Actions workflow with dependabot-auto-merge) labels: - - "dependencies" - - "automated" + - 'dependencies' + - 'automated' # Ignore major version bumps for these (manual review required) ignore: - - dependency-name: "*" - update-types: ["version-update:semver-major"] + - dependency-name: '*' + update-types: ['version-update:semver-major'] # Rebase strategy (cleaner history) - rebase-strategy: "auto" + rebase-strategy: 'auto' # Commit message prefix for semantic versioning commit-message: - prefix: "chore" - include: "scope" + prefix: 'chore' + include: 'scope' diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml index ac5fcbed..90d6ac26 100644 --- a/.github/workflows/ci.yml +++ b/.github/workflows/ci.yml @@ -60,6 +60,8 @@ jobs: cache: 'npm' - name: Install dependencies + env: + ZEROSHOT_TUI_BINARY_SKIP: '1' run: npm ci - name: Security audit @@ -75,6 +77,12 @@ jobs: - name: Type check run: npm run typecheck + - name: Type check (tui-backend) + run: npm run typecheck:tui-backend + + - name: Build tui-backend + run: npm run build:tui-backend + - name: Unit tests with coverage (fast) run: npm run test:coverage timeout-minutes: 3 @@ -83,6 +91,36 @@ jobs: run: npm run test:slow timeout-minutes: 15 + - name: Setup Rust toolchain + uses: dtolnay/rust-toolchain@stable + with: + components: clippy, rustfmt + + - name: Cache cargo registry + build + uses: actions/cache@v4 + with: + path: | + ~/.cargo/registry + ~/.cargo/git + tui-rs/target + key: rust-${{ runner.os }}-${{ hashFiles('tui-rs/Cargo.lock') }} + restore-keys: rust-${{ runner.os }}- + + - name: Rust fmt + run: cargo fmt --manifest-path tui-rs/Cargo.toml --all -- --check + + - name: Rust build + run: cargo build --release --manifest-path tui-rs/Cargo.toml + + - name: Rust test + run: cargo test --manifest-path tui-rs/Cargo.toml + + - name: Rust clippy + run: cargo clippy --manifest-path tui-rs/Cargo.toml -- -D warnings + + - name: Smoke test Rust TUI binary + run: ./tui-rs/target/release/zeroshot-tui --version + - name: Upload coverage to Codecov uses: codecov/codecov-action@v5 with: @@ -100,7 +138,7 @@ jobs: if: | github.ref == 'refs/heads/main' || (github.event_name == 'pull_request' && github.base_ref == 'main') || - (github.event_name == 'merge_group' && github.event.merge_group.base_ref == 'refs/heads/main') + github.event_name == 'merge_group' strategy: fail-fast: false matrix: @@ -122,6 +160,8 @@ jobs: cache: 'npm' - name: Install dependencies + env: + ZEROSHOT_TUI_BINARY_SKIP: '1' run: npm ci - name: Link CLI diff --git a/.github/workflows/codeql.yml b/.github/workflows/codeql.yml index d7e3a20a..92ca8290 100644 --- a/.github/workflows/codeql.yml +++ b/.github/workflows/codeql.yml @@ -1,10 +1,10 @@ -name: "CodeQL Security Analysis" +name: 'CodeQL Security Analysis' on: push: - branches: [ main ] + branches: [main] pull_request: - branches: [ main ] + branches: [main] schedule: # Run at 3 AM UTC every Monday - cron: '0 3 * * 1' @@ -56,4 +56,4 @@ jobs: - name: Perform CodeQL Analysis uses: github/codeql-action/analyze@v3 with: - category: "/language:javascript" + category: '/language:javascript' diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml index 54af8efa..34f14aa3 100644 --- a/.github/workflows/release.yml +++ b/.github/workflows/release.yml @@ -3,7 +3,7 @@ name: Release on: # Auto-release after CI passes on main workflow_run: - workflows: ["CI"] + workflows: ['CI'] types: [completed] branches: [main] # Manual trigger for DRY-RUN ONLY (testing release process) @@ -50,6 +50,8 @@ jobs: cache: 'npm' - name: Install dependencies + env: + ZEROSHOT_TUI_BINARY_SKIP: '1' run: npm ci - name: Link CLI @@ -107,8 +109,72 @@ jobs: console.log('āœ“ StatusFooter test passed'); " + tui-binary-matrix: + if: | + (github.event_name == 'workflow_dispatch' && inputs.dry_run == true) || + (github.event_name == 'workflow_run' && github.event.workflow_run.conclusion == 'success') + strategy: + fail-fast: false + matrix: + include: + - os: ubuntu-latest + platform: linux + arch: x64 + target: x86_64-unknown-linux-gnu + - os: ubuntu-24.04-arm64 + platform: linux + arch: arm64 + target: aarch64-unknown-linux-gnu + - os: macos-13 + platform: darwin + arch: x64 + target: x86_64-apple-darwin + - os: macos-14 + platform: darwin + arch: arm64 + target: aarch64-apple-darwin + + runs-on: ${{ matrix.os }} + + steps: + - uses: actions/checkout@v4 + + - name: Setup Rust toolchain + uses: dtolnay/rust-toolchain@stable + with: + targets: ${{ matrix.target }} + + - name: Build zeroshot-tui + run: | + cargo build --release -p zeroshot-tui --target ${{ matrix.target }} --manifest-path tui-rs/Cargo.toml + + - name: Smoke test binary + run: | + BIN_PATH="tui-rs/target/${{ matrix.target }}/release/zeroshot-tui" + "$BIN_PATH" --version + "$BIN_PATH" --smoke-test + + - name: Package release asset + run: | + BIN_DIR="tui-rs/target/${{ matrix.target }}/release" + ASSET="zeroshot-tui-${{ matrix.platform }}-${{ matrix.arch }}.tar.gz" + tar -czf "$ASSET" -C "$BIN_DIR" zeroshot-tui + if command -v sha256sum >/dev/null 2>&1; then + sha256sum "$ASSET" > "$ASSET.sha256" + else + shasum -a 256 "$ASSET" > "$ASSET.sha256" + fi + + - name: Upload TUI binary artifact + uses: actions/upload-artifact@v4 + with: + name: tui-binary-${{ matrix.platform }}-${{ matrix.arch }} + path: | + zeroshot-tui-${{ matrix.platform }}-${{ matrix.arch }}.tar.gz + zeroshot-tui-${{ matrix.platform }}-${{ matrix.arch }}.tar.gz.sha256 + release: - needs: [install-matrix] + needs: [install-matrix, tui-binary-matrix] # SAFETY: Only run if CI passed OR manual dry-run # Real releases MUST go through CI (workflow_run with success) # Manual trigger MUST be dry-run only (cannot publish untested code) @@ -137,8 +203,23 @@ jobs: cache: 'npm' - name: Install dependencies + env: + ZEROSHOT_TUI_BINARY_SKIP: '1' run: npm ci + - name: Download TUI binaries + uses: actions/download-artifact@v4 + with: + pattern: tui-binary-* + path: dist/tui + merge-multiple: true + + - name: Prepare TUI binary for install validation + run: | + ls -la dist/tui + tar -xzf dist/tui/zeroshot-tui-linux-x64.tar.gz -C dist/tui + chmod +x dist/tui/zeroshot-tui + - name: Verify package before release run: | # Create tarball (respects 'files' array in package.json) @@ -147,13 +228,16 @@ jobs: echo "Created: $TARBALL" # Install from tarball (simulates npm install -g @covibes/zeroshot) - npm install -g "./$TARBALL" + ZEROSHOT_TUI_BINARY_PATH="$PWD/dist/tui/zeroshot-tui" npm install -g "./$TARBALL" # Smoke tests - must all pass echo "=== Testing installed package ===" zeroshot --version zeroshot --help zeroshot list + INSTALLED_ROOT="$(npm root -g)/@covibes/zeroshot" + test -x "$INSTALLED_ROOT/libexec/zeroshot-tui" + "$INSTALLED_ROOT/libexec/zeroshot-tui" --version # Cleanup npm uninstall -g @covibes/zeroshot diff --git a/.gitignore b/.gitignore index 2a205d66..5d2f06d4 100644 --- a/.gitignore +++ b/.gitignore @@ -10,6 +10,9 @@ dist/ build/ *.tsbuildinfo +# Rust build output +tui-rs/target/ + # Environment .env .env.local @@ -40,7 +43,8 @@ tmp/ *.tmp # Zeroshot runtime -.zeroshot/ +.zeroshot/* +!.zeroshot/settings.json .claude-zeroshots/ zeroshot-isolated/ zeroshot-cluster-configs/ @@ -56,3 +60,4 @@ report/ # Generated/temp files test-metadata-manual.sh test-isolated-fix.js +lib/tui-backend/ diff --git a/.husky/pre-commit b/.husky/pre-commit index 9fe39a25..ada93a93 100644 --- a/.husky/pre-commit +++ b/.husky/pre-commit @@ -7,6 +7,15 @@ if [ -n "$BAD_FILES" ]; then exit 1 fi +# Enforce Rust formatting when Rust files are staged +if git diff --cached --name-only -- '*.rs' | grep -q .; then + if ! command -v rustfmt >/dev/null 2>&1; then + echo "ERROR: rustfmt is required to format Rust (install Rust toolchain)." + exit 1 + fi + git diff --cached --name-only -z -- '*.rs' | xargs -0 rustfmt --edition 2021 --check +fi + # Lint and format only staged files npx lint-staged diff --git a/.mocharc.cjs b/.mocharc.cjs new file mode 100644 index 00000000..a9b624d6 --- /dev/null +++ b/.mocharc.cjs @@ -0,0 +1,16 @@ +const hasCliTestFile = process.argv + .slice(2) + .some((arg) => typeof arg === 'string' && /\.test\.[jt]s$/.test(arg)); + +const config = { + parallel: true, + jobs: 4, + timeout: 10000, + slow: 1000, +}; + +if (!hasCliTestFile) { + config.spec = 'tests/**/*.test.js'; +} + +module.exports = config; diff --git a/.mocharc.json b/.mocharc.json deleted file mode 100644 index 651732a1..00000000 --- a/.mocharc.json +++ /dev/null @@ -1,7 +0,0 @@ -{ - "parallel": true, - "jobs": 4, - "timeout": 10000, - "slow": 1000, - "spec": "tests/**/*.test.js" -} diff --git a/.releaserc.json b/.releaserc.json index 90723e6d..0b914af6 100644 --- a/.releaserc.json +++ b/.releaserc.json @@ -9,6 +9,11 @@ "npmPublish": true } ], - "@semantic-release/github" + [ + "@semantic-release/github", + { + "assets": ["dist/tui/zeroshot-tui-*.tar.gz", "dist/tui/zeroshot-tui-*.tar.gz.sha256"] + } + ] ] } diff --git a/.zeroshot/settings.json b/.zeroshot/settings.json new file mode 100644 index 00000000..838c097f --- /dev/null +++ b/.zeroshot/settings.json @@ -0,0 +1,8 @@ +{ + "github": { + "prBase": "dev" + }, + "worktree": { + "baseRef": "origin/dev" + } +} diff --git a/AGENTS.md b/AGENTS.md index 803bbfe1..8a522258 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -11,6 +11,7 @@ Operational rules and references for automated agents working on this repo. Inst - Never use git in validator prompts. Validate files directly. - Never ask questions. Agents run non-interactively; make autonomous decisions. - Never edit `CLAUDE.md` unless explicitly asked to update docs. +- Detached (`-d`) runs must forward all `zeroshot run` options via `ZEROSHOT_RUN_OPTIONS` (see `buildDaemonEnv` + `buildStartOptions`) so PR/worktree config cannot be dropped. Worker git operations are allowed only with isolation (`--worktree`, `--docker`, `--pr`, `--ship`). They are forbidden without isolation. @@ -20,22 +21,70 @@ Destructive commands (need permission): `zeroshot kill`, `zeroshot clear`, `zero ## Where to Look -| Concept | File | -| ------------------------ | ----------------------------------- | -| Conductor classification | `src/conductor-bootstrap.js` | -| Base templates | `cluster-templates/base-templates/` | -| Message bus | `src/message-bus.js` | -| Ledger (SQLite) | `src/ledger.js` | -| Trigger evaluation | `src/logic-engine.js` | -| Agent wrapper | `src/agent-wrapper.js` | -| Providers registry | `src/providers/index.js` | -| Provider implementations | `src/providers/` | -| Provider detection | `lib/provider-detection.js` | -| Provider capabilities | `src/providers/capabilities.js` | -| TUI dashboard | `src/tui/` | -| Docker mounts/env | `lib/docker-config.js` | -| Container lifecycle | `src/isolation-manager.js` | -| Settings | `lib/settings.js` | +| Concept | File | +| ------------------------------ | ---------------------------------------------------------- | +| Conductor classification | `src/conductor-bootstrap.js` | +| Base templates | `cluster-templates/base-templates/` | +| Message bus | `src/message-bus.js` | +| Ledger (SQLite) | `src/ledger.js` | +| Guidance topics | `src/guidance-topics.js` | +| Guidance mailbox helper | `src/ledger.js` | +| Guidance live injection | `src/orchestrator.js` | +| Trigger evaluation | `src/logic-engine.js` | +| Agent wrapper | `src/agent-wrapper.js` | +| Providers registry | `src/providers/index.js` | +| Provider implementations | `src/providers/` | +| Provider detection | `lib/provider-detection.js` | +| Provider capabilities | `src/providers/capabilities.js` | +| TUI backend entrypoint | `src/tui-backend/index.ts` | +| TUI backend server | `src/tui-backend/server.ts` | +| TUI backend services | `src/tui-backend/services/` | +| TUI backend subscriptions | `src/tui-backend/subscriptions/` | +| TUI backend build output | `lib/tui-backend/` | +| TUI launcher (Node) | `lib/tui-launcher.js` | +| TUI binary mapping | `lib/tui-binary.js` | +| TUI start-cluster helper | `lib/start-cluster.js` | +| TUI binary installer | `scripts/install-tui-binary.js` | +| TUI v2 protocol spec | `docs/tui-v2/protocol.md` | +| TUI v2 protocol types (TS) | `src/tui-backend/protocol/` | +| TUI v2 protocol types (Rust) | `tui-rs/crates/zeroshot-tui/src/protocol/` | +| Rust TUI backend client | `tui-rs/crates/zeroshot-tui/src/backend/` | +| Rust TUI entrypoint | `tui-rs/crates/zeroshot-tui/src/main.rs` | +| Rust TUI core loop (MVU) | `tui-rs/crates/zeroshot-tui/src/app/mod.rs` | +| Rust TUI spine completion | `tui-rs/crates/zeroshot-tui/src/app/spine_completion.rs` | +| Rust TUI input routing | `tui-rs/crates/zeroshot-tui/src/input.rs` | +| Rust TUI commands | `tui-rs/crates/zeroshot-tui/src/commands/` | +| Rust TUI command parser | `tui-rs/crates/zeroshot-tui/src/commands/parser.rs` | +| Rust TUI command dispatch | `tui-rs/crates/zeroshot-tui/src/commands/dispatcher.rs` | +| Rust TUI command types | `tui-rs/crates/zeroshot-tui/src/commands/types.rs` | +| Rust TUI screens | `tui-rs/crates/zeroshot-tui/src/screens/` | +| Rust TUI Fleet Radar screen | `tui-rs/crates/zeroshot-tui/src/screens/radar.rs` | +| Rust TUI Cluster Canvas screen | `tui-rs/crates/zeroshot-tui/src/screens/cluster_canvas.rs` | +| Rust TUI render entrypoint | `tui-rs/crates/zeroshot-tui/src/ui/mod.rs` | +| Rust TUI widgets | `tui-rs/crates/zeroshot-tui/src/ui/widgets/` | +| Rust TUI toast widget | `tui-rs/crates/zeroshot-tui/src/ui/widgets/toast.rs` | +| Rust TUI command bar widget | `tui-rs/crates/zeroshot-tui/src/ui/widgets/command_bar.rs` | +| Rust TUI terminal guard | `tui-rs/crates/zeroshot-tui/src/terminal.rs` | +| Docker mounts/env | `lib/docker-config.js` | +| Container lifecycle | `src/isolation-manager.js` | +| Settings | `lib/settings.js` | + +TUI v2 (Rust) convention: +Ratatui is the only supported TUI; legacy UI removed. centralized key routing in `src/input.rs`; `app::update()` is pure and returns effects; `ui::render()` is pure and performs no IO. Adding a screen requires a `ScreenId` variant plus a screen reducer and render entry. +TUI v2 (Rust) command flow: `Effect::Command(CommandRequest)` is emitted by `app::update()` and executed in `src/main.rs` via `commands::dispatch()`, with failures surfaced through `BackendAction::Error`. +TUI v2 (Rust) provider override lives in `AppState.provider_override` and is forwarded when launching clusters (e.g. `StartClusterFromText`). +TUI v2 (Rust) command bar: `AppState.command_bar` captures input; `/` opens it outside Launcher; Esc closes; Submit dispatches. Toast output lives in `AppState.toast` and renders via `ui/widgets/toast.rs`. +TUI v2 (Rust) Agent Microscope renders phase markers derived from cluster timeline events (deduped, capped) in a left margin when space allows. +TUI v2 (Rust) Disruptive zoom stack: `ScreenId::IntentConsole` (root), `FleetRadar`, `ClusterCanvas { id }`, `AgentMicroscope { cluster_id, agent_id }`; zoom stack context drives spine whisper targets. +TUI v2 (Rust) Cluster Canvas overlays: use `ui/widgets/stream.rs` StreamOverlay + placement helper in `screens/cluster_canvas.rs` to render bounded log/timeline slices near focus, clamped to canvas bounds and never intersecting the spine; render after the canvas draw. +TUI v2 (Rust) calm empty states: use `ui/shared.rs::calm_empty_state` for centered headline/detail/footer cards in Disruptive screens. +TUI v2 (Rust) Disruptive stream windowing: `TimeCursor` (mode, `t_ms`, `window_ms`) plus `TimeIndexedBuffer` in `ui/shared.rs` back logs/timeline window queries; cluster canvas overlay renders windowed slices from time-indexed buffers. +TUI v2 (Rust) motion/smoothing: `app/animation.rs` defines `AnimClock` + smoothing helpers; `AppState.anim_clock` + `last_tick_ms` advance on `Tick`; Fleet Radar smooths orb radius/intensity in `FleetRadarState.orb_states` and uses `pulse_factor` for error pulses; Cluster Canvas uses camera target/velocity smoothing via `State.tick_camera()` (render consumes smoothed camera). +TUI v2 (Rust) spine intent submit detects issue refs (`123`, `owner/repo#123`, GitHub issue URL) ā†’ `StartClusterFromIssue`; otherwise `StartClusterFromText`. +TUI v2 (Rust) Disruptive pre-M3 decisions live in `docs/ZEROSHOT-DISRUPTIVE-TUI-DECISIONS.md` (focus, labels, topology, scrub, spine height). +TUI backend test envs: `ZEROSHOT_TUI_BACKEND_MOCK_LAUNCH`, `ZEROSHOT_TUI_BACKEND_MOCK_GUIDANCE`, `ZEROSHOT_TUI_BACKEND_METRICS_PLATFORM` (override platform for metrics; unsupported values force `supported=false`). +TUI backend path override: `ZEROSHOT_TUI_BACKEND_PATH`. +TUI launcher env: `ZEROSHOT_TUI_BINARY_PATH` overrides the installed Rust binary, `ZEROSHOT_TUI_PATH`/`ZEROSHOT_TUI_BIN` override Rust binary path, `ZEROSHOT_TUI_BINARY_URL` overrides release asset URL, `ZEROSHOT_TUI_BINARY_SKIP` skips download, `ZEROSHOT_TUI_INITIAL_SCREEN` + `ZEROSHOT_TUI_PROVIDER_OVERRIDE` + `ZEROSHOT_TUI_UI` feed Rust startup defaults (UI variants: classic, disruptive; CLI: `zeroshot tui --ui `). ## CLI Quick Reference @@ -44,6 +93,7 @@ Destructive commands (need permission): `zeroshot kill`, `zeroshot clear`, `zero zeroshot run 123 # Local, no isolation zeroshot run 123 --worktree # Git worktree isolation zeroshot run 123 --pr # Worktree + create PR +zeroshot run 123 --pr --pr-base dev # PR base: dev, worktree base: origin/dev (incl. -d) zeroshot run 123 --ship # Worktree + PR + auto-merge zeroshot run 123 --docker # Docker container isolation zeroshot run 123 -d # Background (daemon) mode @@ -57,7 +107,9 @@ zeroshot stop # Graceful stop zeroshot kill # Force kill # Utilities -zeroshot watch # TUI dashboard +zeroshot # TUI (TTY only; Rust default) +zeroshot tui # TUI explicit entry +zeroshot watch # TUI Monitor view zeroshot export # Export conversation zeroshot agents list # Available agents zeroshot settings # View/modify settings @@ -70,7 +122,7 @@ UX modes: - Daemon (`-d`): background, Ctrl+C detaches. - Attach (`zeroshot attach`): connect to daemon, Ctrl+C detaches only. -Settings: `defaultProvider`, `providerSettings` (claude/codex/gemini), legacy `maxModel`, `defaultConfig`, `logLevel`. +Settings: `defaultProvider`, `providerSettings` (claude/codex/gemini), legacy `maxModel`, `defaultConfig`, `logLevel`, robustness (`maxRetries`, `backoffBaseMs`, `backoffMaxMs`, `jitterFactor`, `maxRestartAttempts`, `maxTotalRestarts`, `staleWarningsBeforeKill`). ## Architecture @@ -89,6 +141,15 @@ Agent A -> publish() -> SQLite Ledger -> LogicEngine -> trigger match -> Agent B | Logic Script | JS predicate for complex conditions | | Hook | Post-task action (publish message, execute command) | +Restart persistence: orchestrator publishes `AGENT_RESTART_ATTEMPT` to the ledger so restart limits survive orchestrator restarts. + +### Guidance Messaging + +- Topics: `USER_GUIDANCE_CLUSTER`, `USER_GUIDANCE_AGENT` (see `src/guidance-topics.js`). +- Mailbox helper: `ledger.queryGuidanceMailbox()` with `messageBus.queryGuidanceMailbox()` passthrough. +- Live injection: `Orchestrator.sendGuidanceToAgent()` uses `agent.injectInput()` to attempt PTY stdin; always persists `USER_GUIDANCE_AGENT` with `metadata.delivery` (`status: injected|unsupported`, `method: pty`, `taskId`, `reason`). +- Safe-point queue fallback: `AgentWrapper._buildContext()` pulls queued guidance via `collectQueuedGuidance()` and injects a delimited block in `agent-context-builder` between Instructions and Output Schema. Cursor: `agent.lastGuidanceAppliedAt`. + ### Agent Configuration (Minimal) ```json @@ -375,6 +436,8 @@ npm run lint npm run test ``` +Mocha config: `.mocharc.cjs` applies defaults; passing explicit `*.test.js` files on the CLI skips the default `tests/**/*.test.js` spec. + Workers are now explicitly ordered to treat every `VALIDATION_RESULT` line as non-negotiable law before typing again. Failing to read and address each validator complaint before claiming completion will be rejected automatically. ## CI Failure Diagnosis @@ -418,3 +481,4 @@ Do NOT assume single root cause. | Multiple impl files (-v2) | Pre-commit hook | | Spawn without permission | Runtime check | | Git stash usage | Pre-commit hook | +| Rust formatting drift | Pre-commit hook | diff --git a/CHANGELOG.md b/CHANGELOG.md index 394c5c34..3ac46ad2 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -461,7 +461,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0 - `zeroshot kill` - Force stop running cluster - `zeroshot clear` - Remove all stopped clusters - `zeroshot export` - Export conversation as JSON or Markdown -- `zeroshot watch` - Interactive TUI dashboard (htop-style) +- `zeroshot` - Interactive Ink TUI (TTY only) +- `zeroshot tui` - Open Ink TUI explicitly +- `zeroshot watch` - Open Ink TUI Monitor view (htop-style) - `zeroshot agents` - View available agent definitions - `zeroshot settings` - Manage global settings - Shell completion support via omelette @@ -493,9 +495,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0 - Auto-merge support via git-pusher agent - Token authentication with hosts.yml fallback -#### TUI Dashboard +#### Ink TUI -- Real-time cluster monitoring with blessed/blessed-contrib +- Real-time cluster monitoring in Ink TUI - Cluster list with state, agent count, and message count - Message viewer with topic filtering - Agent status display with iteration tracking diff --git a/CLAUDE.md b/CLAUDE.md index 97be5516..bbfbff7d 100644 --- a/CLAUDE.md +++ b/CLAUDE.md @@ -6,12 +6,37 @@ Message-passing primitives for multi-agent workflows. **Install:** `npm i -g @co ## šŸ”“ CRITICAL RULES -| Rule | Why | Forbidden | Required | -| ---------------------------------- | ---------------------------- | ------------------------------------------- | ------------------------------------------------ | -| **Never spawn without permission** | Consumes API credits | "I'll run zeroshot on 123" | User says "run zeroshot" | -| **Never use git in validators** | Git state unreliable | `git diff`, `git status` in prompts | Validate files directly | -| **Never ask questions** | Agents run non-interactively | `AskUserQuestion`, waiting for confirmation | Make autonomous decisions | -| **Never edit CLAUDE.md** | Context file for Claude Code | Editing this file | Read-only unless explicitly asked to update docs | +| Rule | Why | Forbidden | Required | +| ---------------------------------- | ---------------------------- | -------------------------------------------- | ------------------------------------------------ | +| **GENERAL PURPOSE ONLY** | Zeroshot runs on ANY repo | Hardcoded paths, scripts, languages, domains | Discover from target repo's CLAUDE.md/README | +| **Never spawn without permission** | Consumes API credits | "I'll run zeroshot on 123" | User says "run zeroshot" | +| **Never use git in validators** | Git state unreliable | `git diff`, `git status` in prompts | Validate files directly | +| **Never ask questions** | Agents run non-interactively | `AskUserQuestion`, waiting for confirmation | Make autonomous decisions | +| **Never edit CLAUDE.md** | Context file for Claude Code | Editing this file | Read-only unless explicitly asked to update docs | + +### šŸ”“ GENERAL PURPOSE REQUIREMENT (CRITICAL) + +**Zeroshot is a GENERAL-PURPOSE multi-agent orchestrator. It MUST work on ANY repository, ANY programming language, ANY domain.** + +**FORBIDDEN in templates/prompts:** + +- Hardcoded script names (`check-all.sh`, `validate.sh`) +- Hardcoded test commands (`npm test`, `pytest`, `cargo test`) +- Hardcoded file paths (`server/`, `src/`, `tests/`) +- Hardcoded context file names (`CLAUDE.md` - other providers use different files) +- Language-specific assumptions (TypeScript, Python, Rust) +- Domain-specific assumptions (web, CLI, mobile) +- Provider-specific assumptions (Claude, Codex, Gemini) +- Covibes-specific patterns + +**REQUIRED:** + +- Discover validation commands from target repo's context files (README, Makefile, package.json, pyproject.toml, Cargo.toml, etc.) +- Discover test runners from target repo's build system +- Use generic examples in prompts (e.g., "the repo's validation script" NOT "./scripts/check-all.sh") +- Use generic terms for context files ("repo context files" NOT "CLAUDE.md") +- Work correctly on: Python projects, Rust crates, Go modules, Ruby gems, Java/Kotlin, C/C++, etc. +- Work correctly with: Claude, Codex, Gemini, OpenAI, and any future providers **Worker git operations:** Allowed with isolation (`--worktree`, `--docker`, `--pr`, `--ship`). Forbidden without isolation (auto-injected restriction). @@ -19,6 +44,8 @@ Message-passing primitives for multi-agent workflows. **Install:** `npm i -g @co **Destructive (needs permission):** `zeroshot kill`, `zeroshot clear`, `zeroshot purge` +**Detached runs:** Always forward `zeroshot run` options via `ZEROSHOT_RUN_OPTIONS` (see `buildDaemonEnv` + `buildStartOptions`) so PR/worktree config survives daemon mode. + ## šŸ”“ BEHAVIORAL STANDARDS ``` @@ -45,7 +72,7 @@ IS THIS HOW A SENIOR STAFF ARCHITECT WOULD DO IT? ACT LIKE ONE. | Ledger (SQLite) | `src/ledger.js` | | Trigger evaluation | `src/logic-engine.js` | | Agent wrapper | `src/agent-wrapper.js` | -| TUI dashboard | `src/tui/` | +| Rust TUI (Ratatui) | `tui-rs/crates/zeroshot-tui/` | | Docker mounts/env | `lib/docker-config.js` | | Container lifecycle | `src/isolation-manager.js` | | Issue providers | `src/issue-providers/` | @@ -60,6 +87,7 @@ IS THIS HOW A SENIOR STAFF ARCHITECT WOULD DO IT? ACT LIKE ONE. zeroshot run 123 # Local, no isolation zeroshot run 123 --worktree # Git worktree isolation zeroshot run 123 --pr # Worktree + create PR +zeroshot run 123 --pr --pr-base dev # PR base: dev, worktree base: origin/dev (incl. -d) zeroshot run 123 --ship # Worktree + PR + auto-merge zeroshot run 123 --docker # Docker container isolation zeroshot run 123 -d # Background (daemon) mode @@ -73,7 +101,9 @@ zeroshot stop # Graceful stop zeroshot kill # Force kill # Utilities -zeroshot watch # TUI dashboard +zeroshot # Rust TUI (TTY only) +zeroshot tui # Rust TUI explicit entry +zeroshot watch # Rust TUI Monitor view zeroshot export # Export conversation zeroshot agents list # Available agents zeroshot settings # View/modify settings diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md index a1b81ec7..50d5a505 100644 --- a/CONTRIBUTING.md +++ b/CONTRIBUTING.md @@ -22,7 +22,7 @@ Thank you for your interest in contributing to Zeroshot! This guide covers every - **Node.js 18+** (check: `node --version`) - **npm** (bundled with Node) - **Docker** (optional, for isolation mode tests) -- **Claude Code CLI** - `npm i -g @anthropic-ai/claude-code && claude auth login` +- **AI Provider CLI** - At least one: Claude Code (`npm i -g @anthropic-ai/claude-code`), Codex, Gemini CLI, or OpenCode - **GitHub CLI** - Required for PR creation features ([install guide](https://cli.github.com/)) ### Installation @@ -351,29 +351,41 @@ evaluate(script, agent, message) { ## Debugging -### Debug the TUI (zeroshot watch) +### Debug the Rust TUI (zeroshot, zeroshot tui, zeroshot watch) + +Use `zeroshot` (TTY only) or `zeroshot tui` for a normal session. Use `zeroshot watch` to open Monitor view directly. 1. **Run in development mode** ```bash - zeroshot watch + zeroshot tui ``` -2. **Common TUI issues** +2. **CI note** + + The Rust TUI backend integration tests require the Node TUI backend to be built. + CI enforces this (fails if missing). Locally, run: + + ```bash + npm ci + npm run build:tui-backend + ``` + +3. **Common TUI issues** | Issue | Fix | | ------------------ | ----------------------------------------------------- | | Garbled output | Terminal too small - resize to 80x24+ | | Missing agents | Cluster not running - start with `zeroshot run` first | | Stats not updating | File polling delay - wait 2-5 seconds | - | Crash on resize | Known blessed bug - restart TUI | + | Resize glitches | Restart TUI | -3. **Debug TUI rendering** +4. **Debug TUI rendering** - Edit `src/tui/index.js` and add: + Edit `tui-rs/crates/zeroshot-tui/src/` (screen or widget) and add: - ```javascript - screen.log(`Debug: ${JSON.stringify(data)}`); + ```rust + eprintln!("Debug: {data:?}"); ``` ### Debug Agent Execution @@ -574,21 +586,33 @@ See `src/logic-engine.js` for sandbox implementation. ### Context Building -Agents build context from ledger messages before executing: +Agents build context from ledger messages before executing. Use explicit selection semantics: -```javascript +```json { "contextStrategy": { "sources": [ - { "topic": "ISSUE_OPENED", "limit": 1 }, - { "topic": "VALIDATION_RESULT", "since": "last_task_end", "limit": 10 } + { "topic": "ISSUE_OPENED", "priority": "required", "strategy": "latest", "amount": 1 }, + { "topic": "STATE_SNAPSHOT", "priority": "required", "strategy": "latest", "amount": 1 }, + { + "topic": "VALIDATION_RESULT", + "priority": "high", + "since": "last_task_end", + "strategy": "latest", + "amount": 10, + "compactAmount": 3 + } ], "maxTokens": 100000 } } ``` -See `src/agent/agent-context-builder.js` for implementation. +Notes: `limit` is a deprecated alias for `amount`. If `amount` is set and `strategy` is omitted, +the default is `latest`; otherwise defaults to `all`. + +See `docs/context-management.md` for the full reference and diagrams. Implementation lives in +`src/agent/agent-context-builder.js`. ### Template Resolution diff --git a/PUBLISHING.md b/PUBLISHING.md index 64d73c18..3c987bf2 100644 --- a/PUBLISHING.md +++ b/PUBLISHING.md @@ -25,6 +25,7 @@ If you get an error or "organization not found", you need to create it: ### Create the organization: 1. Log in to npm: + ```bash npm login ``` @@ -65,6 +66,7 @@ For CI/CD publishing, you need an **automation token** (not a classic token). ### Token Permissions: Automation tokens have these capabilities: + - āœ… Publish packages - āœ… Update package metadata - āœ… Read private packages (if your org has any) @@ -144,6 +146,7 @@ Once the `NPM_TOKEN` secret is configured, publishing happens automatically: 1. **Make changes** to the codebase 2. **Commit with conventional commit messages**: + ```bash git commit -m "feat: add new feature" # Minor version bump (0.1.0 ā†’ 0.2.0) git commit -m "fix: fix bug" # Patch version bump (0.1.0 ā†’ 0.1.1) @@ -151,6 +154,7 @@ Once the `NPM_TOKEN` secret is configured, publishing happens automatically: ``` 3. **Push to main branch**: + ```bash git push origin main ``` @@ -171,12 +175,12 @@ Once the `NPM_TOKEN` secret is configured, publishing happens automatically: semantic-release uses conventional commits to determine version bumps: -| Commit Type | Version Bump | Example | -|-------------|-------------|---------| -| `fix:` | Patch (0.1.0 ā†’ 0.1.1) | `fix: resolve memory leak` | -| `feat:` | Minor (0.1.0 ā†’ 0.2.0) | `feat: add cluster resume` | -| `feat!:` or `BREAKING CHANGE:` | Major (0.1.0 ā†’ 1.0.0) | `feat!: change API signature` | -| `docs:`, `chore:`, `style:`, etc. | No release | `docs: update README` | +| Commit Type | Version Bump | Example | +| --------------------------------- | --------------------- | ----------------------------- | +| `fix:` | Patch (0.1.0 ā†’ 0.1.1) | `fix: resolve memory leak` | +| `feat:` | Minor (0.1.0 ā†’ 0.2.0) | `feat: add cluster resume` | +| `feat!:` or `BREAKING CHANGE:` | Major (0.1.0 ā†’ 1.0.0) | `feat!: change API signature` | +| `docs:`, `chore:`, `style:`, etc. | No release | `docs: update README` | ### Breaking changes: @@ -203,6 +207,7 @@ git commit -m "feat: new API" -m "BREAKING CHANGE: removes old API" **Cause:** The NPM_TOKEN doesn't have permission to publish to @covibes. **Fix:** + 1. Verify you're a member of the @covibes npm organization 2. Regenerate the automation token 3. Update the GitHub secret diff --git a/README.md b/README.md index 04a46141..a6a8b359 100644 --- a/README.md +++ b/README.md @@ -79,7 +79,7 @@ and surface conflicts with details. Handle the ABA problem where version goes A- ## Install and Requirements -**Platforms**: Linux, macOS (Windows WSL not yet supported) +**Platforms**: Linux, macOS. Windows (native/WSL) is deferred while we harden reliability and multi-provider correctness. ```bash npm install -g @covibes/zeroshot @@ -95,7 +95,7 @@ npm i -g @google/gemini-cli # Opencode: see https://opencode.ai # Authenticate with the provider CLI -claude login # Claude +claude auth login # Claude codex login # Codex gemini auth login # Gemini opencode auth login # Opencode @@ -439,6 +439,7 @@ zeroshot settings set dockerEnvPassthrough '["MY_API_KEY", "TF_VAR_*"]' - [CLAUDE.md](./CLAUDE.md) - Architecture, cluster config schema, agent primitives - `docs/providers.md` - Provider setup, model levels, and Docker mounts +- `docs/context-management.md` - Context selection, context packs, and state snapshots - [Discord](https://discord.gg/PdZ3UEXB) - Support and community - `zeroshot export ` - Export conversation to markdown - `sqlite3 ~/.zeroshot/*.db` - Direct ledger access for debugging @@ -468,6 +469,77 @@ Please read [CODE_OF_CONDUCT.md](CODE_OF_CONDUCT.md) before participating. For security issues, see [SECURITY.md](SECURITY.md). +## TUI + +Ratatui (Rust) is the only supported TUI. Entry points: + +- `zeroshot` (TTY + no args) +- `zeroshot tui` +- `zeroshot watch` + +### TUI Development + +The Rust TUI spawns a Node backend over stdio. Run both while iterating. + +Single command dev loop (auto-rebuild + restart): + +```bash +cargo install cargo-watch +npm run dev:tui +``` + +1. Install deps + + ```bash + npm ci + ``` + +2. Build the TUI backend in watch mode + + ```bash + npm run build:tui-backend -- --watch + # or + npx tsc -p tsconfig.tui-backend.json -w + ``` + +3. Run the Rust TUI (second terminal) + + ```bash + cd tui-rs + cargo run -p zeroshot-tui -- --ui disruptive + ``` + +### Local CLI From This Repo + +If you want `zeroshot` to run from the dev branch globally: + +```bash +npm run dev:link +``` + +One command to link + run the TUI dev loop: + +```bash +npm run dev:bootstrap +``` + +### CLI Integration Loop + +Use the Node CLI to launch your local Rust binary: + +```bash +cd tui-rs +cargo build -p zeroshot-tui +cd .. +ZEROSHOT_TUI_BINARY_PATH="$PWD/tui-rs/target/debug/zeroshot-tui" node cli/index.js tui +``` + +### TUI Overrides + +- `ZEROSHOT_TUI_BACKEND_PATH` points to a specific `lib/tui-backend/server.js` +- `ZEROSHOT_TUI_BINARY_PATH` points to a local Rust binary +- `ZEROSHOT_TUI_UI=classic|disruptive` forces UI variant + --- MIT - [Covibes](https://github.com/covibes) diff --git a/cli/index.js b/cli/index.js index 821976a3..bb02265f 100755 --- a/cli/index.js +++ b/cli/index.js @@ -49,6 +49,17 @@ const { const { normalizeProviderName } = require('../lib/provider-names'); const { getProvider, parseProviderChunk } = require('../src/providers'); const { MOUNT_PRESETS, resolveEnvs } = require('../lib/docker-config'); +const { launchTuiSession } = require('../lib/tui-launcher'); +const { + detectGitRepoRoot, + detectRunInput, + loadClusterConfig, + resolveConfigPath, + resolveProviderOverride, + startClusterFromFile, + startClusterFromIssue, + startClusterFromText, +} = require('../lib/start-cluster'); const { requirePreflight } = require('../src/preflight'); const { providersCommand, setDefaultCommand, setupCommand } = require('./commands/providers'); // Setup wizard removed - use: zeroshot settings set @@ -152,43 +163,6 @@ process.on('unhandledRejection', (reason) => { // Package root directory (for resolving default config paths) const PACKAGE_ROOT = path.resolve(__dirname, '..'); -/** - * Detect git repository root from current directory - * Critical for CWD propagation - agents must work in the target repo, not where CLI was invoked - * @returns {string} Git repo root, or process.cwd() if not in a git repo - */ -function detectGitRepoRoot() { - const { execSync } = require('child_process'); - try { - const root = execSync('git rev-parse --show-toplevel', { - encoding: 'utf8', - stdio: ['pipe', 'pipe', 'pipe'], - }).trim(); - return root; - } catch { - // Not in a git repo - use current directory - return process.cwd(); - } -} - -/** - * Parse CLI mount specs (host:container[:ro]) into mount config objects - * @param {string[]} specs - Array of mount specs from CLI - * @returns {Array<{host: string, container: string, readonly: boolean}>} - */ -function parseMountSpecs(specs) { - return specs.map((spec) => { - const parts = spec.split(':'); - if (parts.length < 2) { - throw new Error(`Invalid mount spec: "${spec}". Format: host:container[:ro]`); - } - const host = parts[0]; - const container = parts[1]; - const readonly = parts[2] === 'ro'; - return { host, container, readonly }; - }); -} - function normalizeRunOptions(options) { if (options.ship) { options.pr = true; @@ -204,45 +178,6 @@ function normalizeRunOptions(options) { } } -function detectRunInput(inputArg) { - const input = {}; - - // Check if it looks like an issue URL or key - const isGitHubUrl = /^https?:\/\/github\.com\/[\w-]+\/[\w-]+\/issues\/\d+/.test(inputArg); - const isGitLabUrl = /gitlab\.(com|[\w.-]+)\/[\w-]+\/[\w-]+\/-\/issues\/\d+/.test(inputArg); - const isJiraUrl = /(atlassian\.net|jira\.[\w.-]+)\/browse\/[A-Z][A-Z0-9]+-\d+/.test(inputArg); - const isAzureUrl = - /dev\.azure\.com\/.*\/_workitems\/edit\/\d+/.test(inputArg) || - /visualstudio\.com\/.*\/_workitems\/edit\/\d+/.test(inputArg); - const isJiraKey = /^[A-Z][A-Z0-9]+-\d+$/.test(inputArg); - const isIssueNumber = /^\d+$/.test(inputArg); - const isRepoIssue = /^[\w-]+\/[\w-]+#\d+$/.test(inputArg); - const isMarkdownFile = /\.(md|markdown)$/i.test(inputArg); - - if ( - isGitHubUrl || - isGitLabUrl || - isJiraUrl || - isAzureUrl || - isJiraKey || - isIssueNumber || - isRepoIssue - ) { - input.issue = inputArg; - } else if (isMarkdownFile) { - input.file = inputArg; - } else { - input.text = inputArg; - } - return input; -} - -function resolveProviderOverride(options, settings) { - return normalizeProviderName( - options.provider || process.env.ZEROSHOT_PROVIDER || settings.defaultProvider - ); -} - function runClusterPreflight({ input, options, providerOverride, settings, forceProvider }) { // Detect which issue provider tool is needed let issueProvider = null; @@ -301,7 +236,23 @@ function createDaemonLogFile(clusterId) { return fs.openSync(logPath, 'w'); } +function serializeRunOptions(options) { + try { + return JSON.stringify(options); + } catch { + return ''; + } +} + +function resolveMergeQueueEnv(value) { + if (value === true) return '1'; + if (value === false) return '0'; + return ''; +} + function buildDaemonEnv(options, clusterId, targetCwd) { + const mergeQueueEnv = resolveMergeQueueEnv(options.mergeQueue); + const runOptionsEnv = serializeRunOptions(options); return { ...process.env, ZEROSHOT_DAEMON: '1', @@ -313,6 +264,10 @@ function buildDaemonEnv(options, clusterId, targetCwd) { ZEROSHOT_WORKERS: options.workers?.toString() || '', ZEROSHOT_MODEL: options.model || '', ZEROSHOT_PROVIDER: options.provider || '', + ZEROSHOT_RUN_OPTIONS: runOptionsEnv, + ZEROSHOT_PR_BASE: options.prBase || '', + ZEROSHOT_MERGE_QUEUE: mergeQueueEnv, + ZEROSHOT_CLOSE_ISSUE: options.closeIssue || '', ZEROSHOT_CWD: targetCwd, }; } @@ -342,42 +297,6 @@ function resolveConfigName(options, settings) { return options.config || settings.defaultConfig; } -function resolveConfigPath(configName) { - if (path.isAbsolute(configName) || configName.startsWith('./') || configName.startsWith('../')) { - return path.resolve(process.cwd(), configName); - } - if (configName.endsWith('.json')) { - return path.join(PACKAGE_ROOT, 'cluster-templates', configName); - } - return path.join(PACKAGE_ROOT, 'cluster-templates', `${configName}.json`); -} - -function ensureConfigProviderDefaults(config, settings) { - if (!config.defaultProvider) { - config.defaultProvider = settings.defaultProvider || 'claude'; - } - config.defaultProvider = normalizeProviderName(config.defaultProvider) || 'claude'; -} - -function applyProviderOverrideToConfig(config, providerOverride, settings) { - const provider = getProvider(providerOverride); - const providerSettings = settings.providerSettings?.[providerOverride] || {}; - config.forceProvider = providerOverride; - config.defaultProvider = providerOverride; - config.forceLevel = providerSettings.defaultLevel || provider.getDefaultLevel(); - config.defaultLevel = config.forceLevel; - console.log(chalk.dim(`Provider override: ${providerOverride} (all agents)`)); -} - -function loadClusterConfig(orchestrator, configPath, settings, providerOverride) { - const config = orchestrator.loadConfig(configPath); - ensureConfigProviderDefaults(config, settings); - if (providerOverride) { - applyProviderOverrideToConfig(config, providerOverride, settings); - } - return config; -} - function trackActiveCluster(clusterId, orchestrator) { activeClusterId = clusterId; orchestratorInstance = orchestrator; @@ -436,8 +355,11 @@ function applyModelOverrideToConfig(config, modelOverride, providerOverride, set ); } - if (providerName === 'claude' && ['opus', 'sonnet', 'haiku'].includes(modelOverride)) { - const { validateModelAgainstMax } = require('../lib/settings'); + if (providerName === 'claude') { + const { validateModelAgainstMax, VALID_MODELS } = require('../lib/settings'); + if (!VALID_MODELS.includes(modelOverride)) { + return; + } try { validateModelAgainstMax(modelOverride, settings.maxModel); } catch (err) { @@ -455,34 +377,6 @@ function applyModelOverrideToConfig(config, modelOverride, providerOverride, set console.log(chalk.dim(`Model override: ${modelOverride} (all agents)`)); } -function buildStartOptions({ - clusterId, - options, - settings, - providerOverride, - modelOverride, - forceProvider, -}) { - const targetCwd = process.env.ZEROSHOT_CWD || detectGitRepoRoot(); - return { - clusterId, - cwd: targetCwd, - isolation: options.docker || process.env.ZEROSHOT_DOCKER === '1' || settings.defaultDocker, - isolationImage: options.dockerImage || process.env.ZEROSHOT_DOCKER_IMAGE || undefined, - worktree: options.worktree || process.env.ZEROSHOT_WORKTREE === '1', - autoPr: options.pr || process.env.ZEROSHOT_PR === '1', - autoMerge: process.env.ZEROSHOT_MERGE === '1', - autoPush: process.env.ZEROSHOT_PUSH === '1', - modelOverride: modelOverride || undefined, - providerOverride: providerOverride || undefined, - noMounts: options.noMounts || false, - mounts: options.mount ? parseMountSpecs(options.mount) : undefined, - containerHome: options.containerHome || undefined, - forceProvider: forceProvider || undefined, - settings, // Pass settings for provider detection - }; -} - function createStatusFooter(clusterId, messageBus) { const statusFooter = new StatusFooter({ refreshInterval: 1000, @@ -2325,6 +2219,9 @@ Examples: ${chalk.cyan('zeroshot run "Implement feature X"')} Run cluster from plain text ${chalk.cyan('zeroshot run 123 -d')} Run in background (detached) ${chalk.cyan('zeroshot run 123 --docker')} Run in Docker container (safe for e2e tests) + ${chalk.cyan('zeroshot')} Open TUI (TTY only) + ${chalk.cyan('zeroshot tui')} Open TUI explicitly + ${chalk.cyan('zeroshot watch')} Open TUI Monitor view ${chalk.cyan('zeroshot task run "Fix the bug"')} Run single-agent background task ${chalk.cyan('zeroshot list')} List all tasks and clusters ${chalk.cyan('zeroshot task list')} List tasks only @@ -2381,12 +2278,23 @@ program '--ship', 'Full automation: worktree isolation + PR + auto-merge (use --docker for Docker)' ) + .option('--pr-base ', 'Target branch for PRs (default: repo default branch)') + .option('--merge-queue', 'Use GitHub merge queue instead of direct merge') + .option( + '--close-issue ', + 'When to close issue after merge: auto|always|never (default: from .zeroshot/settings.json or never)' + ) .option('--workers ', 'Max sub-agents for worker to spawn in parallel', parseInt) .option( '--provider ', 'Override all agents to use a provider (claude, codex, gemini, opencode)' ) .option('--model ', 'Override all agent models (provider-specific model id)') + .option( + '--sim ', + 'Token-free simulation gate for templates (off|fast|deep). Default: fast', + 'fast' + ) .option('-G, --github', 'Force GitHub as issue source') .option('-L, --gitlab', 'Force GitLab as issue source') .option('-J, --jira', 'Force Jira as issue source') @@ -2440,11 +2348,35 @@ Force provider flags: -G (GitHub), -L (GitLab), -J (Jira), -D (DevOps) // Auto-detect input type const input = detectRunInput(inputArg); const settings = loadSettings(); - const providerOverride = resolveProviderOverride(options, settings); + const providerOverride = resolveProviderOverride(options); // Preflight checks runClusterPreflight({ input, options, providerOverride, settings, forceProvider }); + // Secondary preflight: token-free template simulation/validation + const simMode = String(options.sim || 'fast').toLowerCase(); + if (simMode !== 'off') { + const { validateTemplates } = require('../src/template-validation'); + const templatesDir = path.join(PACKAGE_ROOT, 'cluster-templates'); + const deep = simMode === 'deep'; + const report = await validateTemplates({ templatesDir, deep }); + if (!report.valid) { + console.error('\n' + '='.repeat(60)); + console.error(`TEMPLATE VALIDATION FAILED (sim=${simMode})`); + console.error('='.repeat(60)); + for (const { filePath, result } of report.results) { + if (result.valid) continue; + const rel = path.relative(process.cwd(), filePath); + console.error(`\nāŒ ${rel}`); + for (const err of result.errors) { + console.error(` ERROR: ${err}`); + } + } + console.error('\nFix template errors before running to avoid token burn.\n'); + process.exit(1); + } + } + const { generateName } = require('../src/name-generator'); if (shouldRunDetached(options)) { @@ -2470,17 +2402,46 @@ Force provider flags: -G (GitHub), -L (GitLab), -J (Jira), -D (DevOps) const modelOverride = resolveModelOverride(options); applyModelOverrideToConfig(config, modelOverride, providerOverride, settings); - const startOptions = buildStartOptions({ - clusterId, - options, - settings, - providerOverride, - modelOverride, - forceProvider, - }); - - // Start cluster - const cluster = await orchestrator.start(config, input, startOptions); + let cluster = null; + if (input.text) { + cluster = await startClusterFromText({ + orchestrator, + text: input.text, + config, + settings, + providerOverride, + modelOverride, + forceProvider, + clusterId, + options, + }); + } else if (input.issue) { + cluster = await startClusterFromIssue({ + orchestrator, + issue: input.issue, + config, + settings, + providerOverride, + modelOverride, + forceProvider, + clusterId, + options, + }); + } else if (input.file) { + cluster = await startClusterFromFile({ + orchestrator, + file: input.file, + config, + settings, + providerOverride, + modelOverride, + forceProvider, + clusterId, + options, + }); + } else { + throw new Error('Invalid run input: expected text, issue, or file'); + } if (!process.env.ZEROSHOT_DAEMON) { await streamClusterInForeground(cluster, orchestrator, clusterId); @@ -3006,7 +2967,7 @@ program try { // Try cluster first, then task (both use same ID format: "adjective-noun-number") const OrchestratorModule = require('../src/orchestrator'); - const orchestrator = new OrchestratorModule(); + const orchestrator = await OrchestratorModule.create(); // Check if cluster exists const cluster = orchestrator.getCluster(id); @@ -3336,25 +3297,61 @@ program } }); -// Watch command - interactive TUI dashboard +// Watch command - TUI Monitor view program .command('watch') - .description('Interactive TUI to monitor clusters') + .description('Open TUI in Monitor view') .option('--refresh-rate ', 'Refresh interval in milliseconds', '1000') - .action(async (options) => { + .action((_options) => { try { - const TUI = require('../src/tui'); - const tui = new TUI({ - orchestrator: await getOrchestrator(), - refreshRate: parseInt(options.refreshRate, 10), - }); - await tui.start(); + launchTuiSession({ initialView: 'monitor' }); + } catch (error) { + console.error('Error starting TUI:', error.message); + process.exit(1); + } + }); + +// TUI command - TUI session +program + .command('tui') + .description('Open TUI') + .option( + '--provider ', + 'Override provider for this TUI session (claude, codex, gemini, opencode)' + ) + .option('--ui ', 'Select UI variant (classic, disruptive)') + .allowExcessArguments(true) + .allowUnknownOption(true) + .action((options) => { + try { + launchTuiSession(options); } catch (error) { console.error('Error starting TUI:', error.message); process.exit(1); } }); +function registerTuiEntrypoint(commandName, providerName) { + program + .command(commandName) + .description(`Interactive TUI to monitor clusters (provider: ${providerName})`) + .allowExcessArguments(true) + .allowUnknownOption(true) + .action(() => { + try { + launchTuiSession({ provider: providerName }); + } catch (error) { + console.error('Error starting TUI:', error.message); + process.exit(1); + } + }); +} + +registerTuiEntrypoint('codex', 'codex'); +registerTuiEntrypoint('claude', 'claude'); +registerTuiEntrypoint('gemini', 'gemini'); +registerTuiEntrypoint('opencode', 'opencode'); + // Settings management const settingsCmd = program.command('settings').description('Manage zeroshot settings'); @@ -4418,6 +4415,8 @@ function formatTokenUsage(tokensByRole) { // Total line const inputTokens = total.inputTokens || 0; const outputTokens = total.outputTokens || 0; + const cacheReadTokens = total.cacheReadInputTokens || 0; + const uncachedInputTokens = inputTokens - cacheReadTokens; const totalTokens = inputTokens + outputTokens; const cost = total.totalCostUsd || 0; @@ -4431,6 +4430,17 @@ function formatTokenUsage(tokensByRole) { chalk.dim(' out)') ); + // Cache breakdown (if cache data available) + if (cacheReadTokens > 0) { + const cachePercent = Math.round((cacheReadTokens / inputTokens) * 100); + lines.push( + chalk.dim('Cache: ') + + chalk.green(fmt(cacheReadTokens) + ' cached') + + chalk.dim(' (' + cachePercent + '%) + ') + + chalk.yellow(fmt(uncachedInputTokens) + ' new') + ); + } + // Cost line (if available) if (cost > 0) { lines.push(chalk.dim('Cost: ') + chalk.green('$' + cost.toFixed(4))); @@ -5271,9 +5281,21 @@ async function main() { // Check for updates (non-blocking if offline) await checkForUpdates({ quiet: isQuiet }); + let args = process.argv.slice(2); + + if (args.length === 0) { + const isInteractiveTty = Boolean(process.stdin.isTTY && process.stdout.isTTY); + if (isInteractiveTty) { + process.argv.splice(2, 0, 'tui'); + } else { + program.outputHelp(); + return; + } + } + // Default command handling: if first arg doesn't match a known command, treat it as 'run' // This allows `zeroshot "task"` to work the same as `zeroshot run "task"` - const args = process.argv.slice(2); + args = process.argv.slice(2); if (args.length > 0) { const firstArg = args[0]; diff --git a/cli/message-formatters-watch.js b/cli/message-formatters-watch.js index 288561c2..b09613b0 100644 --- a/cli/message-formatters-watch.js +++ b/cli/message-formatters-watch.js @@ -4,7 +4,11 @@ */ const chalk = require('chalk'); -const { buildClusterPrefix, getColorForSender, parseDataField } = require('./message-formatter-utils'); +const { + buildClusterPrefix, + getColorForSender, + parseDataField, +} = require('./message-formatter-utils'); /** * Format AGENT_ERROR for watch mode diff --git a/cluster-templates/base-templates/debug-workflow.json b/cluster-templates/base-templates/debug-workflow.json index c95130d3..7925fb51 100644 --- a/cluster-templates/base-templates/debug-workflow.json +++ b/cluster-templates/base-templates/debug-workflow.json @@ -120,7 +120,15 @@ "sources": [ { "topic": "ISSUE_OPENED", - "limit": 1 + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "STATE_SNAPSHOT", + "priority": "required", + "strategy": "latest", + "amount": 1 } ], "format": "chronological", @@ -164,16 +172,28 @@ "sources": [ { "topic": "ISSUE_OPENED", - "limit": 1 + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "STATE_SNAPSHOT", + "priority": "required", + "strategy": "latest", + "amount": 1 }, { "topic": "INVESTIGATION_COMPLETE", - "limit": 1 + "priority": "high", + "strategy": "latest", + "amount": 1 }, { "topic": "VALIDATION_RESULT", + "priority": "high", "since": "last_task_end", - "limit": 5 + "strategy": "latest", + "amount": 5 } ], "format": "chronological", @@ -333,16 +353,28 @@ "sources": [ { "topic": "ISSUE_OPENED", - "limit": 1 + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "STATE_SNAPSHOT", + "priority": "required", + "strategy": "latest", + "amount": 1 }, { "topic": "INVESTIGATION_COMPLETE", - "limit": 1 + "priority": "high", + "strategy": "latest", + "amount": 1 }, { "topic": "FIX_APPLIED", + "priority": "medium", "since": "last_agent_start", - "limit": 1 + "strategy": "latest", + "amount": 1 } ], "format": "chronological", diff --git a/cluster-templates/base-templates/full-workflow.json b/cluster-templates/base-templates/full-workflow.json index 80b25602..d7a7da3f 100644 --- a/cluster-templates/base-templates/full-workflow.json +++ b/cluster-templates/base-templates/full-workflow.json @@ -108,13 +108,21 @@ "required": ["plan", "summary", "filesAffected", "acceptanceCriteria"] }, "prompt": { - "system": "## šŸš« YOU CANNOT ASK QUESTIONS\n\nYou are running non-interactively. There is NO USER to answer.\n- NEVER use AskUserQuestion tool\n- NEVER say \"Should I...\" or \"Would you like...\"\n- When unsure: Make the SAFER choice and proceed.\n\nYou are a planning agent for a {{complexity}} {{task_type}} task.\n\n## Your Job\nCreate a FLAT LIST of executable steps. The worker will execute them IN ORDER.\n\n## Plan Scope: Single-Session Execution\n\nEvery step must be completable in ONE autonomous session.\n\n**Allowed:**\n- Code/test/doc changes\n- Immediate verification (run tests, check files exist)\n\n**Forbidden:**\n- Waiting periods (hours/days/weeks)\n- Deployment/operations tasks\n- Monitoring over time\n\nFinal step: \"Ready to deploy\" (NOT \"deploy it\").\n\n## šŸ”“ PLAN FORMAT (CRITICAL)\n\nOutput a flat list of numbered steps in the `plan` field. Each step is ONE concrete action.\n\n**EXAMPLE - CORRECT:**\n```\n1. Create server/services/rate-limiter.ts with RateLimiter class\n2. Add middleware registration in server/src/server.ts:45\n3. Add config constant in server/config/limits.ts\n4. Write test in tests/unit/rate-limiter.test.ts\n5. Run npm test to verify\n```\n\n**FORBIDDEN:**\n- \"Phase 1\", \"Phase 2\" ā†’ NO PHASES. Just steps.\n- \"Future work\" ā†’ NO. Everything NOW.\n- \"We could do X or Y\" ā†’ NO OPTIONS. Pick one.\n- Delegation to sub-agents ā†’ NO. Worker does it all.\n- Deferring anything ā†’ FORBIDDEN.\n\nJust numbered steps. Execute in order. Done.\n\n## šŸ”“ ONE PLAN. THE BEST PLAN.\n\nāŒ ABSOLUTELY FORBIDDEN:\n- 'Option 1... Option 2...'\n- 'Alternative approaches include...'\n- 'We could either X or Y'\n- Hedging with 'alternatively'\n\nāœ… REQUIRED:\n- ONE decisive implementation approach\n- The approach a FAANG Staff/Principal Engineer would choose\n- Clean architecture, no hacks\n\nYou are a STAFF LEVEL PRINCIPAL ENGINEER. Make THE decision. Present THE plan.\n\n## Planning Process\n1. Analyze requirements thoroughly\n2. Explore codebase to understand architecture\n3. Identify ALL files that need changes\n4. Break down into concrete, actionable steps\n5. Consider cross-component dependencies\n\n{{#if complexity == 'CRITICAL'}}\n## CRITICAL TASK - EXTRA SCRUTINY\n- This is HIGH RISK (auth, payments, security, production)\n- Plan must include rollback strategy\n- Consider blast radius of changes\n- Identify all possible failure modes\n{{/if}}\n\n## šŸ”“ ACCEPTANCE CRITERIA (REQUIRED - minItems: 3)\n\nYou MUST output explicit, testable acceptance criteria.\n\n### BAD vs GOOD Criteria:\n\nāŒ BAD: \"Dark mode works correctly\"\nāœ… GOOD: \"Toggle dark mode ā†’ all text readable (contrast ratio >4.5:1), background #1a1a1a\"\n\nāŒ BAD: \"API handles errors\"\nāœ… GOOD: \"POST /api/users with invalid email ā†’ returns 400 + {error: 'Invalid email format'}\"\n\nāŒ BAD: \"Tests pass\"\nāœ… GOOD: \"Test suite passes with 100% success, coverage >80% on new files\"\n\n### Criteria Format:\nEach criterion MUST have:\n- **id**: AC1, AC2, AC3, etc.\n- **criterion**: TESTABLE statement\n- **verification**: EXACT steps to verify\n- **priority**: MUST (blocks completion), SHOULD (important), NICE (bonus)\n\nMinimum 3 criteria. At least 1 MUST be priority=MUST.\n\n## šŸ”“ OUTPUT CONCISENESS (CRITICAL)\n\nYour plan will be consumed by other agents. Be CONCISE.\n\n**FORBIDDEN:**\n- āŒ Paragraphs explaining WHY\n- āŒ Background context\n- āŒ Explaining obvious steps\n- āŒ Code examples for trivial changes\n\n**REQUIRED:**\n- āœ… Steps as imperative commands (\"Add X to file.ts:123\")\n- āœ… File paths without explanations\n- āœ… Target: <2000 words total\n\n**EXAMPLE - BAD:**\n\"First, we need to update the health monitor service located at server/services/preview/health-monitor.ts. This file is responsible for monitoring container health...\"\n\n**EXAMPLE - GOOD:**\n\"Refactor health-monitor.ts:validateContainerHealth() - delegate to executor.checkContainerHealth()\"\n\nDO NOT implement - planning only." + "system": "## šŸš« YOU CANNOT ASK QUESTIONS\n\nYou are running non-interactively. There is NO USER to answer.\n- NEVER use AskUserQuestion tool\n- NEVER say \"Should I...\" or \"Would you like...\"\n- When unsure: Make the SAFER choice and proceed.\n\nYou are a planning agent for a {{complexity}} {{task_type}} task.\n\n## Your Job\nCreate a FLAT LIST of executable steps. The worker will execute them IN ORDER.\n\n## Plan Scope: Single-Session Execution\n\nEvery step must be completable in ONE autonomous session.\n\n**Allowed:**\n- Code/test/doc changes\n- Immediate verification (run tests, check files exist)\n\n**Forbidden:**\n- Waiting periods (hours/days/weeks)\n- Deployment/operations tasks\n- Monitoring over time\n\nFinal step: \"Ready to deploy\" (NOT \"deploy it\").\n\n## šŸ”“ PLAN FORMAT (CRITICAL)\n\nOutput a flat list of numbered steps in the `plan` field. Each step is ONE concrete action.\n\n**EXAMPLE - CORRECT:**\n```\n1. Create server/services/rate-limiter.ts with RateLimiter class\n2. Add middleware registration in server/src/server.ts:45\n3. Add config constant in server/config/limits.ts\n4. Write test in tests/unit/rate-limiter.test.ts\n5. Run npm test to verify\n```\n\n## šŸ”“ ACTIONABLE STEPS - NO RE-EXPLORATION\n\nYour plan must be complete enough that ANY developer could implement it without re-reading the codebase.\n\nYou already analyzed the code. The worker should NOT re-analyze it.\n\n**INCLUDE IN EACH STEP:**\n- What to create/change\n- Key signatures, structures, patterns needed\n- Reference file:lines for quick lookup (not deep reading)\n\n**NEVER write:** \"Read file X to understand Y\" - you already understand it. Write what you learned.\n**ALWAYS write:** Concrete details the worker needs to execute immediately.\n**BE CONCISE:** Patterns in 1-2 lines, not tutorials. Target <2500 chars total.\n\n**FORBIDDEN:**\n- \"Phase 1\", \"Phase 2\" ā†’ NO PHASES. Just steps.\n- \"Future work\" ā†’ NO. Everything NOW.\n- \"We could do X or Y\" ā†’ NO OPTIONS. Pick one.\n- Delegation to sub-agents ā†’ NO. Worker does it all.\n- Deferring anything ā†’ FORBIDDEN.\n\nJust numbered steps. Execute in order. Done.\n\n## šŸ”“ ONE PLAN. THE BEST PLAN.\n\nāŒ ABSOLUTELY FORBIDDEN:\n- 'Option 1... Option 2...'\n- 'Alternative approaches include...'\n- 'We could either X or Y'\n- Hedging with 'alternatively'\n\nāœ… REQUIRED:\n- ONE decisive implementation approach\n- The approach a FAANG Staff/Principal Engineer would choose\n- Clean architecture, no hacks\n\nYou are a STAFF LEVEL PRINCIPAL ENGINEER. Make THE decision. Present THE plan.\n\n## Planning Process\n1. Analyze requirements thoroughly\n2. Explore codebase to understand architecture\n3. Identify ALL files that need changes\n4. Break down into concrete, actionable steps\n5. Consider cross-component dependencies\n\n{{#if complexity == 'CRITICAL'}}\n## CRITICAL TASK - EXTRA SCRUTINY\n- This is HIGH RISK (auth, payments, security, production)\n- Plan must include rollback strategy\n- Consider blast radius of changes\n- Identify all possible failure modes\n{{/if}}\n\n## šŸ”“ ACCEPTANCE CRITERIA (REQUIRED - minItems: 3)\n\nYou MUST output explicit, testable acceptance criteria.\n\n### BAD vs GOOD Criteria:\n\nāŒ BAD: \"Dark mode works correctly\"\nāœ… GOOD: \"Toggle dark mode ā†’ all text readable (contrast ratio >4.5:1), background #1a1a1a\"\n\nāŒ BAD: \"API handles errors\"\nāœ… GOOD: \"POST /api/users with invalid email ā†’ returns 400 + {error: 'Invalid email format'}\"\n\nāŒ BAD: \"Tests pass\"\nāœ… GOOD: \"Test suite passes with 100% success, coverage >80% on new files\"\n\n### Criteria Format:\nEach criterion MUST have:\n- **id**: AC1, AC2, AC3, etc.\n- **criterion**: TESTABLE statement\n- **verification**: EXACT steps to verify\n- **priority**: MUST (blocks completion), SHOULD (important), NICE (bonus)\n\nMinimum 3 criteria. At least 1 MUST be priority=MUST.\n\n## šŸ”“ OUTPUT CONCISENESS (CRITICAL)\n\nYour plan will be consumed by other agents. Be CONCISE.\n\n**FORBIDDEN:**\n- āŒ Paragraphs explaining WHY\n- āŒ Background context\n- āŒ Explaining obvious steps\n- āŒ Code examples for trivial changes\n\n**REQUIRED:**\n- āœ… Steps as imperative commands (\"Add X to file.ts:123\")\n- āœ… File paths without explanations\n- āœ… Target: <2000 words total\n\n**EXAMPLE - BAD:**\n\"First, we need to update the health monitor service located at server/services/preview/health-monitor.ts. This file is responsible for monitoring container health...\"\n\n**EXAMPLE - GOOD:**\n\"Refactor health-monitor.ts:validateContainerHealth() - delegate to executor.checkContainerHealth()\"\n\nDO NOT implement - planning only." }, "contextStrategy": { "sources": [ { "topic": "ISSUE_OPENED", - "limit": 1 + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "STATE_SNAPSHOT", + "priority": "medium", + "strategy": "latest", + "amount": 1 } ], "format": "chronological", @@ -193,21 +201,35 @@ "sources": [ { "topic": "ISSUE_OPENED", - "limit": 1 + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "STATE_SNAPSHOT", + "priority": "required", + "strategy": "latest", + "amount": 1 }, { "topic": "PLAN_READY", - "limit": 1 + "priority": "required", + "strategy": "latest", + "amount": 1 }, { "topic": "WORKER_PROGRESS", + "priority": "medium", "since": "last_task_end", - "limit": 3 + "strategy": "latest", + "amount": 3 }, { "topic": "VALIDATION_RESULT", + "priority": "high", "since": "last_task_end", - "limit": 10 + "strategy": "latest", + "amount": 10 } ], "format": "chronological", @@ -230,7 +252,7 @@ "topic": "VALIDATION_RESULT", "logic": { "engine": "javascript", - "script": "const validators = cluster.getAgentsByRole('validator');\nconst lastPush = ledger.findLast({ topic: 'IMPLEMENTATION_READY' });\nif (!lastPush) return false;\nconst responses = ledger.query({ topic: 'VALIDATION_RESULT', since: lastPush.timestamp });\nif (responses.length < validators.length) return false;\nreturn responses.some(r => r.content?.data?.approved === false || r.content?.data?.approved === 'false');" + "script": "const validators = cluster.getAgentsByRole('validator');\nconst lastPush = ledger.findLast({ topic: 'IMPLEMENTATION_READY' });\nif (!lastPush) return false;\n\nconst responses = ledger.query({ topic: 'VALIDATION_RESULT', since: lastPush.timestamp });\nif (responses.length === 0) return false;\n\nconst validatorIds = new Set(validators.map((v) => v.id));\nconst validatorResponses = responses.filter((r) => validatorIds.has(r.sender));\n\n// Support both patterns:\n// - Per-validator VALIDATION_RESULT (STANDARD) ā†’ wait for all validators.\n// - Consensus-only VALIDATION_RESULT (staged quick/heavy) ā†’ use consensus messages.\nif (validators.length > 0 && validatorResponses.length > 0) {\n const latestByValidator = new Map();\n for (const msg of validatorResponses) {\n latestByValidator.set(msg.sender, msg);\n }\n if (latestByValidator.size < validators.length) return false;\n return Array.from(latestByValidator.values()).some(\n (r) => r.content?.data?.approved === false || r.content?.data?.approved === 'false'\n );\n}\n\nreturn responses.some((r) => r.content?.data?.approved === false || r.content?.data?.approved === 'false');" }, "action": "execute_task" } @@ -255,12 +277,92 @@ }, "maxIterations": "{{max_iterations}}" }, + { + "id": "meta-coordinator", + "role": "coordinator", + "modelLevel": "level1", + "timeout": "{{timeout}}", + "condition": "{{complexity}} == 'CRITICAL' && {{validator_count}} == 0", + "outputFormat": "json", + "jsonSchema": { + "type": "object", + "properties": { + "action": { + "type": "string", + "enum": ["load_quick", "load_heavy", "skip"] + } + }, + "required": ["action"] + }, + "prompt": { + "system": "Coordinate two-stage validation. On IMPLEMENTATION_READY: output {\"action\":\"load_quick\"}. On QUICK_VALIDATION_PASSED: output {\"action\":\"load_heavy\"}. On VALIDATION_RESULT from Stage 1 (rejection): output {\"action\":\"skip\"}" + }, + "contextStrategy": { + "sources": [ + { + "topic": "IMPLEMENTATION_READY", + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "QUICK_VALIDATION_PASSED", + "priority": "medium", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "VALIDATION_RESULT", + "priority": "medium", + "strategy": "latest", + "amount": 1 + } + ], + "format": "chronological", + "maxTokens": "{{max_tokens}}" + }, + "triggers": [ + { + "topic": "IMPLEMENTATION_READY", + "logic": { + "engine": "javascript", + "script": "return !(message && message.metadata && message.metadata._republished);" + }, + "action": "execute_task" + }, + { + "topic": "QUICK_VALIDATION_PASSED", + "logic": { + "engine": "javascript", + "script": "return !(message && message.metadata && message.metadata._republished);" + }, + "action": "execute_task" + } + ], + "hooks": { + "onComplete": { + "action": "publish_message", + "config": { + "topic": "CLUSTER_OPERATIONS" + }, + "transform": { + "engine": "javascript", + "script": "const template = result.action === 'load_quick' ? 'quick-validation' : 'heavy-validation'; const republishTopic = result.action === 'load_quick' ? 'IMPLEMENTATION_READY' : 'QUICK_VALIDATION_PASSED'; const republishContent = triggeringMessage && triggeringMessage.content ? triggeringMessage.content : { text: '', data: {} }; return { topic: 'CLUSTER_OPERATIONS', content: { text: result.action === 'load_quick' ? 'Stage 1 (quick)' : 'Stage 2 (heavy)', data: { operations: [{ action: 'load_config', config: { base: template, params: { validator_level: '{{validator_level}}', max_tokens: {{max_tokens}}, timeout: {{timeout}} } } }, { action: 'publish', topic: republishTopic, content: republishContent, metadata: { _republished: true } }] } } };" + }, + "logic": { + "engine": "javascript", + "script": "return result.action !== 'skip';" + } + } + } + }, { "id": "validator-requirements", "role": "validator", "modelLevel": "{{validator_level}}", "timeout": "{{timeout}}", "maxRetries": 3, + "condition": "{{validator_count}} >= 1 && {{validator_count}} < 4", "outputFormat": "json", "jsonSchema": { "type": "object", @@ -316,25 +418,37 @@ } } }, - "required": ["approved", "summary", "criteriaResults"] + "required": ["approved", "summary"] }, "prompt": { - "system": "# REQUIREMENTS VALIDATOR\n\nVerify implementation meets ALL requirements from issue. Hold a HIGH BAR.\n\n## WORKFLOW\n1. Read context files (CLAUDE.md, AGENTS.md, README) for repo-specific validation\n2. Parse acceptanceCriteria from PLAN_READY\n3. For EACH criterion: run verification, record evidence\n4. If repo has validation script (e.g. `./scripts/check-all.sh`), RUN IT\n\n## VERIFICATION\n- SEARCH before claiming 'missing' (Glob, Grep, Read)\n- RUN commands, capture output as evidence\n- CANNOT_VALIDATE only for: tool not installed, no network, permission denied\n\n## INSTANT REJECT\n- TODO/FIXME/placeholder = REJECT\n- Silent error swallowing = REJECT\n- 'Phase 2 deferred' = REJECT\n- 'Will add tests later' = REJECT\n- ANY priority=MUST criterion fails = REJECT\n\n## APPROVAL\n- approved:true = ALL MUST criteria pass + no blocking issues\n- approved:false = any MUST fails OR incomplete implementation\n\nšŸš« NO questions. Make safe choice and proceed.\n\n## šŸ”“ OUTPUT FORMAT (CRITICAL)\n\nYou MUST return valid JSON with these REQUIRED fields:\n```json\n{\n \"approved\": boolean,\n \"summary\": \"<100 chars max>\",\n \"errors\": [\"blocking issue 1\", \"blocking issue 2\"],\n \"criteriaResults\": [{\"id\": \"AC1\", \"status\": \"PASS|FAIL|CANNOT_VALIDATE\", \"evidence\": {\"command\": \"...\", \"exitCode\": 0, \"output\": \"<200 chars>\"}, \"reason\": \"for CANNOT_VALIDATE only\"}]\n}\n```\nNo preamble. JSON only." + "system": "# REQUIREMENTS VALIDATOR\n\nVerify implementation meets ALL requirements from issue. Hold a HIGH BAR.\n\n## WORKFLOW\n1. Read context files (README, CONTRIBUTING, Makefile, package.json, etc.) for repo-specific validation\n2. Parse acceptanceCriteria from PLAN_READY\n3. For EACH criterion: run verification, record evidence\n4. If repo has validation script (validation script, make check, etc.), RUN IT\n\n## VERIFICATION\n- SEARCH before claiming 'missing' (Glob, Grep, Read)\n- RUN commands, capture output as evidence\n- CANNOT_VALIDATE only for: tool not installed, no network, permission denied\n\n## INSTANT REJECT\n- TODO/FIXME/placeholder = REJECT\n- Silent error swallowing = REJECT\n- 'Phase 2 deferred' = REJECT\n- 'Will add tests later' = REJECT\n- ANY priority=MUST criterion fails = REJECT\n\n## APPROVAL\n- approved:true = ALL MUST criteria pass + no blocking issues\n- approved:false = any MUST fails OR incomplete implementation\n\nšŸš« NO questions. Make safe choice and proceed.\n\n## šŸ”“ OUTPUT FORMAT (CRITICAL)\n\nYou MUST return valid JSON with these REQUIRED fields:\n```json\n{\n \"approved\": boolean,\n \"summary\": \"<100 chars max>\",\n \"errors\": [\"blocking issue 1\", \"blocking issue 2\"],\n \"criteriaResults\": [{\"id\": \"AC1\", \"status\": \"PASS|FAIL|CANNOT_VALIDATE\", \"evidence\": {\"command\": \"...\", \"exitCode\": 0, \"output\": \"<200 chars>\"}, \"reason\": \"for CANNOT_VALIDATE only\"}]\n}\n```\nNo preamble. JSON only." }, "contextStrategy": { "sources": [ { "topic": "ISSUE_OPENED", - "limit": 1 + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "STATE_SNAPSHOT", + "priority": "required", + "strategy": "latest", + "amount": 1 }, { "topic": "PLAN_READY", - "limit": 1 + "priority": "required", + "strategy": "latest", + "amount": 1 }, { "topic": "IMPLEMENTATION_READY", + "priority": "high", "since": "last_agent_start", - "limit": 1 + "strategy": "latest", + "amount": 1 } ], "format": "chronological", @@ -369,7 +483,7 @@ "modelLevel": "{{validator_level}}", "timeout": "{{timeout}}", "maxRetries": 3, - "condition": "{{validator_count}} >= 2", + "condition": "{{validator_count}} >= 2 && {{validator_count}} < 4", "outputFormat": "json", "jsonSchema": { "type": "object", @@ -390,22 +504,34 @@ "required": ["approved", "summary"] }, "prompt": { - "system": "# CODE VALIDATOR\n\nSenior engineer code review. Catch REAL bugs, not style preferences.\n\n## WORKFLOW\n1. Read context files (CLAUDE.md, AGENTS.md, README) for repo-specific validation\n2. SEARCH before claiming 'missing' (Glob, Grep, Read)\n3. RUN validation scripts if specified\n\n## INSTANT REJECT\n- TODO/FIXME/placeholder = REJECT\n- Silent error swallowing = REJECT\n- Dangerous fallbacks hiding failures = REJECT\n\n## šŸ”“ GENERALIZATION CHECK (CRITICAL)\nWorker fixed a bug? Verify they fixed ALL instances:\n1. Identify the PATTERN (not just the line)\n2. `grep -rn \"pattern\" .` - search codebase\n3. If N > 1 exists ā†’ Did worker fix ALL? If NO ā†’ REJECT\n\nExamples: null check in one handler? Check ALL. SQL injection in one query? Check ALL. A fix that leaves identical bugs elsewhere is NOT a fix.\n\n## BLOCKING (reject with WHAT/HOW/WHY)\n- Logic/off-by-one bugs\n- Race conditions\n- Security holes (injection, auth bypass)\n- Resource leaks (timers, connections)\n- God functions (>50 lines) - SPLIT\n- DRY violation (same logic 2+ places)\n- Missing error handling\n- Hardcoded values that should be config\n\n## NOT BLOCKING (summary only)\n- Style/naming preferences\n- 'Could theoretically...' without proof\n\nšŸš« NO questions. Make safe choice and proceed.\n\n## šŸ”“ OUTPUT FORMAT (CRITICAL)\n\nYou MUST return valid JSON:\n```json\n{\n \"approved\": boolean,\n \"summary\": \"<100 chars max>\",\n \"errors\": [\"WHAT: X. HOW: Y. WHY: Z\"]\n}\n```\nNo preamble. JSON only." + "system": "# CODE VALIDATOR\n\nSenior engineer code review. Catch REAL bugs, not style preferences.\n\n## WORKFLOW\n1. Read context files (README, CONTRIBUTING, Makefile, package.json, etc.) for repo-specific validation\n2. SEARCH before claiming 'missing' (Glob, Grep, Read)\n3. RUN validation scripts if specified\n\n## INSTANT REJECT\n- TODO/FIXME/placeholder = REJECT\n- Silent error swallowing = REJECT\n- Dangerous fallbacks hiding failures = REJECT\n\n## šŸ”“ GENERALIZATION CHECK (CRITICAL)\nWorker fixed a bug? Verify they fixed ALL instances:\n1. Identify the PATTERN (not just the line)\n2. `grep -rn \"pattern\" .` - search codebase\n3. If N > 1 exists ā†’ Did worker fix ALL? If NO ā†’ REJECT\n\nExamples: null check in one handler? Check ALL. SQL injection in one query? Check ALL. A fix that leaves identical bugs elsewhere is NOT a fix.\n\n## šŸ”“ CLEAN DESIGN CHECK (CRITICAL)\nUnless the issue EXPLICITLY requests backwards compatibility, REJECT any of these:\n- Old function/class kept alongside new one (\"deprecated but still works\") ā†’ REJECT. Delete old, update ALL callers.\n- Re-exports or wrappers forwarding to new implementation ā†’ REJECT. Update imports at call sites.\n- `_unused` parameter renames to preserve old signatures ā†’ REJECT. Change the signature, update callers.\n- Fallback paths handling \"old format\" or \"legacy data\" ā†’ REJECT. Migrate the data, remove the fallback.\n- Comments like \"kept for backwards compatibility\" ā†’ REJECT. Remove the old code.\n- Feature flags toggling between old and new behavior ā†’ REJECT. Ship the new behavior. Delete the old.\n\nThe CLEAN solution: delete the old, update all callers, ship only the current implementation. A senior architect demolishes old scaffolding - they don't leave it standing \"just in case\".\n\n## BLOCKING (reject with WHAT/HOW/WHY)\n- Logic/off-by-one bugs\n- Race conditions\n- Security holes (injection, auth bypass)\n- Resource leaks (timers, connections)\n- God functions (>50 lines) - SPLIT\n- DRY violation (same logic 2+ places)\n- Missing error handling\n- Hardcoded values that should be config\n- Backwards compatibility shims or legacy wrappers (see CLEAN DESIGN CHECK)\n\n## NOT BLOCKING (summary only)\n- Style/naming preferences\n- 'Could theoretically...' without proof\n\nšŸš« NO questions. Make safe choice and proceed.\n\n## šŸ”“ OUTPUT FORMAT (CRITICAL)\n\nYou MUST return valid JSON:\n```json\n{\n \"approved\": boolean,\n \"summary\": \"<100 chars max>\",\n \"errors\": [\"WHAT: X. HOW: Y. WHY: Z\"]\n}\n```\nNo preamble. JSON only." }, "contextStrategy": { "sources": [ { "topic": "ISSUE_OPENED", - "limit": 1 + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "STATE_SNAPSHOT", + "priority": "required", + "strategy": "latest", + "amount": 1 }, { "topic": "PLAN_READY", - "limit": 1 + "priority": "required", + "strategy": "latest", + "amount": 1 }, { "topic": "IMPLEMENTATION_READY", + "priority": "high", "since": "last_agent_start", - "limit": 1 + "strategy": "latest", + "amount": 1 } ], "format": "chronological", @@ -439,7 +565,7 @@ "modelLevel": "{{validator_level}}", "timeout": "{{timeout}}", "maxRetries": 3, - "condition": "{{validator_count}} >= 3", + "condition": "{{validator_count}} == 3", "outputFormat": "json", "jsonSchema": { "type": "object", @@ -460,22 +586,34 @@ "required": ["approved", "summary"] }, "prompt": { - "system": "## šŸ”“ OUTPUT FORMAT (CRITICAL - READ FIRST)\n\nYour output MUST be MINIMAL and STRUCTURED:\n- Output ONLY the required JSON schema fields\n- NO preambles (\"Here is my analysis...\", \"Let me explain...\")\n- NO verbose summaries - be CONCISE (max 100 chars per string field)\n- NO redundant information\n- NO explanations before or after the JSON\n\n## šŸš« YOU CANNOT ASK QUESTIONS\n\nYou are running non-interactively. There is NO USER to answer.\n- NEVER use AskUserQuestion tool\n- NEVER say \"Should I...\" or \"Would you like...\"\n- When unsure: Make the SAFER choice and proceed.\n\n## šŸ”“ READ CONTEXT FILES FOR REPO-SPECIFIC VALIDATION\n\n**BEFORE approving any implementation:**\n1. Read the repo's context files (CLAUDE.md, AGENTS.md, README if they exist)\n2. Look for validation instructions, scripts, or commands the repo specifies\n3. If context files say to run a validation script (e.g., `./scripts/check-all.sh`), RUN IT\n4. If the validation script fails, the implementation is NOT complete - REJECT\n\nThis ensures you validate according to THIS repo's standards, not generic rules.\n\n## šŸ”“ VERIFICATION PROTOCOL (REQUIRED - PREVENTS FALSE CLAIMS)\n\nBefore making ANY claim about security vulnerabilities or missing protections:\n\n1. **SEARCH FIRST** - Use Glob to find ALL relevant files\n2. **READ THE CODE** - Use Read to inspect actual implementation\n3. **GREP FOR PATTERNS** - Use Grep to search for specific code (auth checks, validation, etc.)\n\n**NEVER claim a vulnerability exists without FIRST searching for the relevant code.**\n\nThe worker may have implemented security features in different files than originally planned. If you claim 'missing input validation' without searching, you may miss that validation exists in 'server/middleware/validator.ts' instead of the controller.\n\n### Example Verification Flow:\n```\n1. Claim: 'Missing SQL injection protection'\n2. BEFORE claiming ā†’ Grep for 'parameterized', 'prepared', 'escape' in relevant files\n3. BEFORE claiming ā†’ Read the actual database query code\n4. ONLY IF NOT FOUND ā†’ Add to errors array\n```\n\nYou are a security auditor for a {{complexity}} task.\n\n## Security Review Checklist\n1. Input validation (injection attacks)\n2. Authentication/authorization checks\n3. Sensitive data handling\n4. OWASP Top 10 vulnerabilities\n5. Secrets management\n6. Error messages don't leak info\n\n## Output\n- approved: true if no security issues\n- summary: Security assessment\n- errors: Security vulnerabilities found\n\n## šŸ”“ DEBUGGING METHODOLOGY CHECK\n\nBefore approving, verify the worker didn't take shortcuts:\n\n### Ad Hoc Fix Detection\n- Did worker fix ONE instance? ā†’ Grep for similar patterns. If N > 1 exists, REJECT.\n- Example: Fixed null check in `auth.ts:42`? ā†’ `grep -r \"similar pattern\" .` - are there others?\n\n### Root Cause vs Symptom\n- Did worker add a workaround? ā†’ Find the ACTUAL bug. If workaround hides real issue, REJECT.\n- Example: Added `|| []` fallback? ā†’ WHY is it undefined? Fix THAT.\n\n### Lazy Debugging Red Flags (INSTANT REJECT)\n- Worker suggests \"restart the service\" ā†’ REJECT (hides the bug)\n- Worker suggests \"clear the cache\" ā†’ REJECT (hides the bug)\n- Worker says \"works on my machine\" ā†’ REJECT (not a fix)\n- Worker blames the test ā†’ REJECT unless they PROVE test is wrong with evidence" + "system": "## šŸ”“ OUTPUT FORMAT (CRITICAL - READ FIRST)\n\nYour output MUST be MINIMAL and STRUCTURED:\n- Output ONLY the required JSON schema fields\n- NO preambles (\"Here is my analysis...\", \"Let me explain...\")\n- NO verbose summaries - be CONCISE (max 100 chars per string field)\n- NO redundant information\n- NO explanations before or after the JSON\n\n## šŸš« YOU CANNOT ASK QUESTIONS\n\nYou are running non-interactively. There is NO USER to answer.\n- NEVER use AskUserQuestion tool\n- NEVER say \"Should I...\" or \"Would you like...\"\n- When unsure: Make the SAFER choice and proceed.\n\n## šŸ”“ READ CONTEXT FILES FOR REPO-SPECIFIC VALIDATION\n\n**BEFORE approving any implementation:**\n1. Read the repo's context files (README, CONTRIBUTING, Makefile, package.json, etc.)\n2. Look for validation instructions, scripts, or commands the repo specifies\n3. If context files say to run a validation script (validation script, make check, etc.), RUN IT\n4. If the validation script fails, the implementation is NOT complete - REJECT\n\nThis ensures you validate according to THIS repo's standards, not generic rules.\n\n## šŸ”“ VERIFICATION PROTOCOL (REQUIRED - PREVENTS FALSE CLAIMS)\n\nBefore making ANY claim about security vulnerabilities or missing protections:\n\n1. **SEARCH FIRST** - Use Glob to find ALL relevant files\n2. **READ THE CODE** - Use Read to inspect actual implementation\n3. **GREP FOR PATTERNS** - Use Grep to search for specific code (auth checks, validation, etc.)\n\n**NEVER claim a vulnerability exists without FIRST searching for the relevant code.**\n\nThe worker may have implemented security features in different files than originally planned. If you claim 'missing input validation' without searching, you may miss that validation exists in 'server/middleware/validator.ts' instead of the controller.\n\n### Example Verification Flow:\n```\n1. Claim: 'Missing SQL injection protection'\n2. BEFORE claiming ā†’ Grep for 'parameterized', 'prepared', 'escape' in relevant files\n3. BEFORE claiming ā†’ Read the actual database query code\n4. ONLY IF NOT FOUND ā†’ Add to errors array\n```\n\nYou are a security auditor for a {{complexity}} task.\n\n## Security Review Checklist\n1. Input validation (injection attacks)\n2. Authentication/authorization checks\n3. Sensitive data handling\n4. OWASP Top 10 vulnerabilities\n5. Secrets management\n6. Error messages don't leak info\n\n## Output\n- approved: true if no security issues\n- summary: Security assessment\n- errors: Security vulnerabilities found\n\n## šŸ”“ DEBUGGING METHODOLOGY CHECK\n\nBefore approving, verify the worker didn't take shortcuts:\n\n### Ad Hoc Fix Detection\n- Did worker fix ONE instance? ā†’ Grep for similar patterns. If N > 1 exists, REJECT.\n- Example: Fixed null check in `auth.ts:42`? ā†’ `grep -r \"similar pattern\" .` - are there others?\n\n### Root Cause vs Symptom\n- Did worker add a workaround? ā†’ Find the ACTUAL bug. If workaround hides real issue, REJECT.\n- Example: Added `|| []` fallback? ā†’ WHY is it undefined? Fix THAT.\n\n### Lazy Debugging Red Flags (INSTANT REJECT)\n- Worker suggests \"restart the service\" ā†’ REJECT (hides the bug)\n- Worker suggests \"clear the cache\" ā†’ REJECT (hides the bug)\n- Worker says \"works on my machine\" ā†’ REJECT (not a fix)\n- Worker blames the test ā†’ REJECT unless they PROVE test is wrong with evidence" }, "contextStrategy": { "sources": [ { "topic": "ISSUE_OPENED", - "limit": 1 + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "STATE_SNAPSHOT", + "priority": "required", + "strategy": "latest", + "amount": 1 }, { "topic": "PLAN_READY", - "limit": 1 + "priority": "required", + "strategy": "latest", + "amount": 1 }, { "topic": "IMPLEMENTATION_READY", + "priority": "high", "since": "last_agent_start", - "limit": 1 + "strategy": "latest", + "amount": 1 } ], "format": "chronological", @@ -509,7 +647,7 @@ "modelLevel": "{{validator_level}}", "timeout": "{{timeout}}", "maxRetries": 3, - "condition": "{{validator_count}} >= 4", + "condition": "{{validator_count}} == 3", "outputFormat": "json", "jsonSchema": { "type": "object", @@ -533,22 +671,34 @@ "required": ["approved", "summary"] }, "prompt": { - "system": "## šŸ”“ OUTPUT FORMAT (CRITICAL - READ FIRST)\n\nYour output MUST be MINIMAL and STRUCTURED:\n- Output ONLY the required JSON schema fields\n- NO preambles (\"Here is my analysis...\", \"Let me explain...\")\n- NO verbose summaries - be CONCISE (max 100 chars per string field)\n- NO redundant information\n- NO explanations before or after the JSON\n- testResults field: ONLY include pass/fail counts and key errors, NOT full test output\n\n## šŸš« YOU CANNOT ASK QUESTIONS\n\nYou are running non-interactively. There is NO USER to answer.\n- NEVER use AskUserQuestion tool\n- NEVER say \"Should I...\" or \"Would you like...\"\n- When unsure: Make the SAFER choice and proceed.\n\nYou are a TEST EXECUTOR. Your job is to RUN TESTS, not read them.\n\n## šŸ”“ CORE PRINCIPLE: RUN THE TESTS, DON'T JUST READ THEM\n\n**Reading test code is NOT verification. You must EXECUTE tests.**\n\n- 'Tests look correct' = NOT ACCEPTABLE\n- 'Test output shows 15/15 passing' = ACTUAL VERIFICATION\n\n## šŸ”“ STEP 1: FIND AND RUN THE TEST SUITE (MANDATORY)\n\n1. Read context files (CLAUDE.md, AGENTS.md, README) for repo-specific test commands\n2. Find the test runner: `npm test`, `pytest`, `go test`, `cargo test`, etc.\n3. **RUN THE TESTS** using Bash tool\n4. Record FULL output in testResults field\n5. If ANY tests fail ā†’ REJECT immediately\n\n**This is not optional. You MUST run tests, not just search for them.**\n\n## šŸ”“ STEP 2: RUN REPO-SPECIFIC VALIDATION\n\nIf context files specify validation commands (e.g., `./scripts/check-all.sh`):\n1. RUN THEM\n2. Record output\n3. If they fail ā†’ REJECT\n\n## šŸ”“ STEP 3: VERIFY TEST QUALITY BY RUNNING\n\n**DO NOT assess quality by reading code. Assess by execution:**\n\n1. Run tests with verbose output: `npm test -- --verbose`\n2. Check coverage: `npm test -- --coverage`\n3. Record actual numbers in testResults\n\n**Quality indicators from EXECUTION:**\n- Coverage percentage (from actual run)\n- Number of test cases (from actual output)\n- Test duration (from actual output)\n\n## FORBIDDEN PATTERNS\n\n- āŒ 'Tests appear to have good coverage' without running them\n- āŒ 'Test assertions look correct' without executing them\n- āŒ 'The test file exists' as evidence of testing\n- āŒ Approving without testResults containing actual test output\n\n## APPROVAL CRITERIA\n\nONLY approve if:\n1. You RAN the test suite (actual output in testResults)\n2. All tests pass (verified by execution)\n3. Repo-specific validation commands pass (if specified)\n4. Coverage is acceptable for the repo (from actual coverage report)\n\n## Output\n- **approved**: true if tests RAN and PASSED\n- **summary**: Assessment based on ACTUAL test execution results\n- **errors**: Issues found (from running tests, not reading code)\n- **testResults**: ACTUAL OUTPUT from running test commands (REQUIRED)\n\n## šŸ”“ DEBUGGING METHODOLOGY CHECK\n\nBefore approving, verify the worker didn't take shortcuts:\n\n### Ad Hoc Fix Detection\n- Did worker fix ONE instance? ā†’ Grep for similar patterns. If N > 1 exists, REJECT.\n- Example: Fixed null check in `auth.ts:42`? ā†’ `grep -r \"similar pattern\" .` - are there others?\n\n### Root Cause vs Symptom\n- Did worker add a workaround? ā†’ Find the ACTUAL bug. If workaround hides real issue, REJECT.\n- Example: Added `|| []` fallback? ā†’ WHY is it undefined? Fix THAT.\n\n### Lazy Debugging Red Flags (INSTANT REJECT)\n- Worker suggests \"restart the service\" ā†’ REJECT (hides the bug)\n- Worker suggests \"clear the cache\" ā†’ REJECT (hides the bug)\n- Worker says \"works on my machine\" ā†’ REJECT (not a fix)\n- Worker blames the test ā†’ REJECT unless they PROVE test is wrong with evidence" + "system": "## šŸ”“ OUTPUT FORMAT (CRITICAL - READ FIRST)\n\nYour output MUST be MINIMAL and STRUCTURED:\n- Output ONLY the required JSON schema fields\n- NO preambles (\"Here is my analysis...\", \"Let me explain...\")\n- NO verbose summaries - be CONCISE (max 100 chars per string field)\n- NO redundant information\n- NO explanations before or after the JSON\n- testResults field: ONLY include pass/fail counts and key errors, NOT full test output\n\n## šŸš« YOU CANNOT ASK QUESTIONS\n\nYou are running non-interactively. There is NO USER to answer.\n- NEVER use AskUserQuestion tool\n- NEVER say \"Should I...\" or \"Would you like...\"\n- When unsure: Make the SAFER choice and proceed.\n\nYou are a TEST EXECUTOR. Your job is to RUN TESTS, not read them.\n\n## šŸ”“ CORE PRINCIPLE: RUN THE TESTS, DON'T JUST READ THEM\n\n**Reading test code is NOT verification. You must EXECUTE tests.**\n\n- 'Tests look correct' = NOT ACCEPTABLE\n- 'Test output shows 15/15 passing' = ACTUAL VERIFICATION\n\n## šŸ”“ STEP 1: FIND AND RUN THE TEST SUITE (MANDATORY)\n\n1. Read context files (README, CONTRIBUTING, Makefile, package.json, etc.) for repo-specific test commands\n2. Find the test runner: `npm test`, `pytest`, `go test`, `cargo test`, etc.\n3. **RUN THE TESTS** using Bash tool\n4. Record FULL output in testResults field\n5. If ANY tests fail ā†’ REJECT immediately\n\n**This is not optional. You MUST run tests, not just search for them.**\n\n## šŸ”“ STEP 2: RUN REPO-SPECIFIC VALIDATION\n\nIf context files specify validation commands (validation script, make check, etc.):\n1. RUN THEM\n2. Record output\n3. If they fail ā†’ REJECT\n\n## šŸ”“ STEP 3: VERIFY TEST QUALITY BY RUNNING\n\n**DO NOT assess quality by reading code. Assess by execution:**\n\n1. Run tests with verbose output: `npm test -- --verbose`\n2. Check coverage: `npm test -- --coverage`\n3. Record actual numbers in testResults\n\n**Quality indicators from EXECUTION:**\n- Coverage percentage (from actual run)\n- Number of test cases (from actual output)\n- Test duration (from actual output)\n\n## FORBIDDEN PATTERNS\n\n- āŒ 'Tests appear to have good coverage' without running them\n- āŒ 'Test assertions look correct' without executing them\n- āŒ 'The test file exists' as evidence of testing\n- āŒ Approving without testResults containing actual test output\n\n## APPROVAL CRITERIA\n\nONLY approve if:\n1. You RAN the test suite (actual output in testResults)\n2. All tests pass (verified by execution)\n3. Repo-specific validation commands pass (if specified)\n4. Coverage is acceptable for the repo (from actual coverage report)\n\n## Output\n- **approved**: true if tests RAN and PASSED\n- **summary**: Assessment based on ACTUAL test execution results\n- **errors**: Issues found (from running tests, not reading code)\n- **testResults**: ACTUAL OUTPUT from running test commands (REQUIRED)\n\n## šŸ”“ DEBUGGING METHODOLOGY CHECK\n\nBefore approving, verify the worker didn't take shortcuts:\n\n### Ad Hoc Fix Detection\n- Did worker fix ONE instance? ā†’ Grep for similar patterns. If N > 1 exists, REJECT.\n- Example: Fixed null check in `auth.ts:42`? ā†’ `grep -r \"similar pattern\" .` - are there others?\n\n### Root Cause vs Symptom\n- Did worker add a workaround? ā†’ Find the ACTUAL bug. If workaround hides real issue, REJECT.\n- Example: Added `|| []` fallback? ā†’ WHY is it undefined? Fix THAT.\n\n### Lazy Debugging Red Flags (INSTANT REJECT)\n- Worker suggests \"restart the service\" ā†’ REJECT (hides the bug)\n- Worker suggests \"clear the cache\" ā†’ REJECT (hides the bug)\n- Worker says \"works on my machine\" ā†’ REJECT (not a fix)\n- Worker blames the test ā†’ REJECT unless they PROVE test is wrong with evidence" }, "contextStrategy": { "sources": [ { "topic": "ISSUE_OPENED", - "limit": 1 + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "STATE_SNAPSHOT", + "priority": "required", + "strategy": "latest", + "amount": 1 }, { "topic": "PLAN_READY", - "limit": 1 + "priority": "required", + "strategy": "latest", + "amount": 1 }, { "topic": "IMPLEMENTATION_READY", + "priority": "high", "since": "last_agent_start", - "limit": 1 + "strategy": "latest", + "amount": 1 } ], "format": "chronological", diff --git a/cluster-templates/base-templates/heavy-validation.json b/cluster-templates/base-templates/heavy-validation.json new file mode 100644 index 00000000..50e76024 --- /dev/null +++ b/cluster-templates/base-templates/heavy-validation.json @@ -0,0 +1,272 @@ +{ + "name": "Heavy Validation", + "description": "Stage 2: Security + Adversarial testing. Expensive (120-180s). Only runs after Stage 1 passes.", + "params": { + "validator_level": { + "type": "string", + "enum": ["level1", "level2", "level3"], + "default": "level2" + }, + "max_tokens": { + "type": "number", + "default": 100000 + }, + "timeout": { + "type": "number", + "default": 0, + "description": "Task timeout in milliseconds (0 = no timeout)" + } + }, + "agents": [ + { + "id": "validator-security", + "role": "validator", + "modelLevel": "{{validator_level}}", + "timeout": "{{timeout}}", + "maxRetries": 3, + "outputFormat": "json", + "jsonSchema": { + "type": "object", + "properties": { + "approved": { + "type": "boolean" + }, + "summary": { + "type": "string" + }, + "errors": { + "type": "array", + "items": { + "type": "string" + } + } + }, + "required": ["approved", "summary"] + }, + "prompt": { + "system": "## šŸ”“ OUTPUT FORMAT (CRITICAL - READ FIRST)\n\nYour output MUST be MINIMAL and STRUCTURED:\n- Output ONLY the required JSON schema fields\n- NO preambles (\"Here is my analysis...\", \"Let me explain...\")\n- NO verbose summaries - be CONCISE (max 100 chars per string field)\n- NO redundant information\n- NO explanations before or after the JSON\n\n## šŸš« YOU ARE A VALIDATOR - READ-ONLY MODE\n\nYour ONLY job is to VALIDATE and OUTPUT JSON.\n- NEVER use Edit, Write, or any file modification tools\n- NEVER fix code - REJECT with errors array instead\n- NEVER create files, modify files, or make commits\n\n## šŸš« YOU CANNOT ASK QUESTIONS\n\nYou are running non-interactively. There is NO USER to answer.\n- NEVER use AskUserQuestion tool\n- NEVER say \"Should I...\" or \"Would you like...\"\n- When unsure: Make the SAFER choice and proceed.\n\n## šŸ”“ READ CONTEXT FILES FOR REPO-SPECIFIC VALIDATION\n\n**BEFORE approving any implementation:**\n1. Read the repo's context files (README, CONTRIBUTING, Makefile, package.json, etc.)\n2. Look for validation instructions, scripts, or commands the repo specifies\n3. If context files say to run a validation script (validation script, make check, etc.), RUN IT\n4. If the validation script fails, the implementation is NOT complete - REJECT\n\nThis ensures you validate according to THIS repo's standards, not generic rules.\n\n## šŸ”“ VERIFICATION PROTOCOL (REQUIRED - PREVENTS FALSE CLAIMS)\n\nBefore making ANY claim about security vulnerabilities or missing protections:\n\n1. **SEARCH FIRST** - Use Glob to find ALL relevant files\n2. **READ THE CODE** - Use Read to inspect actual implementation\n3. **GREP FOR PATTERNS** - Use Grep to search for specific code (auth checks, validation, etc.)\n\n**NEVER claim a vulnerability exists without FIRST searching for the relevant code.**\n\nThe worker may have implemented security features in different files than originally planned. If you claim 'missing input validation' without searching, you may miss that validation exists in 'server/middleware/validator.ts' instead of the controller.\n\n### Example Verification Flow:\n```\n1. Claim: 'Missing SQL injection protection'\n2. BEFORE claiming ā†’ Grep for 'parameterized', 'prepared', 'escape' in relevant files\n3. BEFORE claiming ā†’ Read the actual database query code\n4. ONLY IF NOT FOUND ā†’ Add to errors array\n```\n\nYou are a security auditor for a CRITICAL task.\n\n## Security Review Checklist\n1. Input validation (injection attacks)\n2. Authentication/authorization checks\n3. Sensitive data handling\n4. OWASP Top 10 vulnerabilities\n5. Secrets management\n6. Error messages don't leak info\n\n## Output\n- approved: true if no security issues\n- summary: Security assessment\n- errors: Security vulnerabilities found\n\n## šŸ”“ DEBUGGING METHODOLOGY CHECK\n\nBefore approving, verify the worker didn't take shortcuts:\n\n### Ad Hoc Fix Detection\n- Did worker fix ONE instance? ā†’ Grep for similar patterns. If N > 1 exists, REJECT.\n- Example: Fixed null check in `auth.ts:42`? ā†’ `grep -r \"similar pattern\" .` - are there others?\n\n### Root Cause vs Symptom\n- Did worker add a workaround? ā†’ Find the ACTUAL bug. If workaround hides real issue, REJECT.\n- Example: Added `|| []` fallback? ā†’ WHY is it undefined? Fix THAT.\n\n### Lazy Debugging Red Flags (INSTANT REJECT)\n- Worker suggests \"restart the service\" ā†’ REJECT (hides the bug)\n- Worker suggests \"clear the cache\" ā†’ REJECT (hides the bug)\n- Worker says \"works on my machine\" ā†’ REJECT (not a fix)\n- Worker blames the test ā†’ REJECT unless they PROVE test is wrong with evidence" + }, + "contextStrategy": { + "sources": [ + { + "topic": "ISSUE_OPENED", + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "STATE_SNAPSHOT", + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "PLAN_READY", + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "IMPLEMENTATION_READY", + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "QUICK_VALIDATION_PASSED", + "priority": "required", + "strategy": "latest", + "amount": 1 + } + ], + "format": "chronological", + "maxTokens": "{{max_tokens}}" + }, + "triggers": [ + { + "topic": "QUICK_VALIDATION_PASSED", + "action": "execute_task" + } + ], + "hooks": { + "onComplete": { + "action": "publish_message", + "config": { + "topic": "HEAVY_VALIDATION_RESULT", + "content": { + "text": "{{result.summary}}", + "data": { + "approved": "{{result.approved}}", + "errors": "{{result.errors}}", + "validatorId": "validator-security" + } + } + } + } + } + }, + { + "id": "validator-tester", + "role": "validator", + "modelLevel": "{{validator_level}}", + "timeout": "{{timeout}}", + "maxRetries": 3, + "outputFormat": "json", + "jsonSchema": { + "type": "object", + "properties": { + "approved": { + "type": "boolean" + }, + "summary": { + "type": "string" + }, + "errors": { + "type": "array", + "items": { + "type": "string" + } + }, + "testResults": { + "type": "string" + }, + "skippedTests": { + "type": "array", + "description": "Tests that couldn't run due to missing env vars or external service unavailability. Treated as WARNING, not failure.", + "items": { + "type": "object", + "properties": { + "test": { + "type": "string", + "description": "Test file or suite name" + }, + "reason": { + "type": "string", + "description": "Why it was skipped (missing env var, unavailable service, etc.)" + } + }, + "required": ["test", "reason"] + } + } + }, + "required": ["approved", "summary"] + }, + "prompt": { + "system": "## šŸ”“ OUTPUT FORMAT (CRITICAL - READ FIRST)\n\nYour output MUST be MINIMAL and STRUCTURED:\n- Output ONLY the required JSON schema fields\n- NO preambles (\"Here is my analysis...\", \"Let me explain...\")\n- NO verbose summaries - be CONCISE (max 100 chars per string field)\n- NO redundant information\n- NO explanations before or after the JSON\n- testResults field: ONLY include pass/fail counts and key errors, NOT full test output\n\n## šŸš« YOU ARE A VALIDATOR - READ-ONLY MODE\n\nYour ONLY job is to VALIDATE and OUTPUT JSON.\n- NEVER use Edit, Write, or any file modification tools\n- NEVER fix code - REJECT with errors array instead\n- NEVER create files, modify files, or make commits\n\n## šŸš« YOU CANNOT ASK QUESTIONS\n\nYou are running non-interactively. There is NO USER to answer.\n- NEVER use AskUserQuestion tool\n- NEVER say \"Should I...\" or \"Would you like...\"\n- When unsure: Make the SAFER choice and proceed.\n\nYou are a TEST EXECUTOR. Your job is to RUN TESTS, not read them.\n\n## šŸ”“ CORE PRINCIPLE: RUN THE TESTS, DON'T JUST READ THEM\n\n**Reading test code is NOT verification. You must EXECUTE tests.**\n\n- 'Tests look correct' = NOT ACCEPTABLE\n- 'Test output shows 15/15 passing' = ACTUAL VERIFICATION\n\n## šŸ”“ STEP 1: DISCOVER TEST COMMANDS FROM REPO CONTEXT\n\n1. Read context files (README, CONTRIBUTING, Makefile, package.json, pyproject.toml, Cargo.toml, etc.)\n2. Find the test runner command - DO NOT assume any specific tool\n3. Common patterns: `npm test`, `pytest`, `go test ./...`, `cargo test`, `make test`\n4. Use what the repo specifies, not hardcoded commands\n\n## šŸ”“ STEP 2: RUN TARGETED TESTS (NOT FULL SUITE)\n\n1. Identify changed files from IMPLEMENTATION_READY message\n2. Find tests RELATED to changed files:\n - Same directory (e.g., `src/foo.ts` ā†’ `src/foo.test.ts`)\n - Test files that import the changed files\n - Naming convention matches (e.g., `auth.ts` ā†’ `tests/auth.test.ts`)\n3. Run ONLY related tests using repo's \"related tests\" feature if available:\n - Jest: `--findRelatedTests `\n - pytest: `pytest `\n - go test: `go test ./path/to/package`\n4. For CRITICAL tasks (auth, payments, security): broader test scope is acceptable\n5. Record output in testResults field\n6. If ANY tests fail ā†’ REJECT immediately\n\n**DO NOT run the full test suite when targeted tests suffice.**\n\n## šŸ”“ STEP 3: RUN REPO-SPECIFIC VALIDATION\n\nIf context files specify validation commands (validation script, make check, etc.):\n1. RUN THEM\n2. Record output\n3. If they fail ā†’ REJECT\n\n## FORBIDDEN PATTERNS\n\n- āŒ 'Tests appear to have good coverage' without running them\n- āŒ 'Test assertions look correct' without executing them\n- āŒ 'The test file exists' as evidence of testing\n- āŒ Approving without testResults containing actual test output\n- āŒ Hardcoding `npm test` or any specific test runner without checking repo context\n- āŒ Running full test suite + coverage when only 2 files changed\n\n## šŸ”“ HANDLING SKIPPED TESTS\n\nIf a test SKIPS due to missing prerequisites (env vars, external services, credentials):\n- Add to skippedTests array with test name and reason\n- Do NOT add to errors array\n- Do NOT reject for skipped tests alone\n- Core tests must still pass\n\nSkipped = WARNING, not FAILURE. Same as CANNOT_VALIDATE in requirements validation.\n\n## APPROVAL CRITERIA\n\nONLY approve if:\n1. You RAN tests related to changed files\n2. All RUNNABLE tests pass (skipped tests = warning only)\n3. Repo-specific validation commands pass\n4. Changed code has test coverage\n\n## Output\n- **approved**: true if runnable tests passed\n- **summary**: Assessment based on test execution\n- **errors**: Issues from running tests\n- **testResults**: Actual test output (REQUIRED)\n- **skippedTests**: Tests that couldn't run (warning only)\n\n## šŸ”“ DEBUGGING METHODOLOGY CHECK\n\nBefore approving, verify the worker didn't take shortcuts:\n\n### Ad Hoc Fix Detection\n- Did worker fix ONE instance? ā†’ Grep for similar patterns. If N > 1 exists, REJECT.\n- Example: Fixed null check in `auth.ts:42`? ā†’ `grep -r \"similar pattern\" .` - are there others?\n\n### Root Cause vs Symptom\n- Did worker add a workaround? ā†’ Find the ACTUAL bug. If workaround hides real issue, REJECT.\n- Example: Added `|| []` fallback? ā†’ WHY is it undefined? Fix THAT.\n\n### Lazy Debugging Red Flags (INSTANT REJECT)\n- Worker suggests \"restart the service\" ā†’ REJECT (hides the bug)\n- Worker suggests \"clear the cache\" ā†’ REJECT (hides the bug)\n- Worker says \"works on my machine\" ā†’ REJECT (not a fix)\n- Worker blames the test ā†’ REJECT unless they PROVE test is wrong with evidence" + }, + "contextStrategy": { + "sources": [ + { + "topic": "ISSUE_OPENED", + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "STATE_SNAPSHOT", + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "PLAN_READY", + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "IMPLEMENTATION_READY", + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "QUICK_VALIDATION_PASSED", + "priority": "required", + "strategy": "latest", + "amount": 1 + } + ], + "format": "chronological", + "maxTokens": "{{max_tokens}}" + }, + "triggers": [ + { + "topic": "QUICK_VALIDATION_PASSED", + "action": "execute_task" + } + ], + "hooks": { + "onComplete": { + "action": "publish_message", + "config": { + "topic": "HEAVY_VALIDATION_RESULT", + "content": { + "text": "{{result.summary}}", + "data": { + "approved": "{{result.approved}}", + "errors": "{{result.errors}}", + "testResults": "{{result.testResults}}", + "skippedTests": "{{result.skippedTests}}", + "validatorId": "validator-tester" + } + } + } + } + } + }, + { + "id": "consensus-coordinator", + "role": "coordinator", + "modelLevel": "level1", + "timeout": "{{timeout}}", + "outputFormat": "json", + "jsonSchema": { + "type": "object", + "properties": { + "allApproved": { + "type": "boolean" + }, + "summary": { + "type": "string" + } + }, + "required": ["allApproved", "summary"] + }, + "prompt": { + "system": "Check if both validators approved. Output: {\"allApproved\": boolean, \"summary\": \"<50 chars>\"}" + }, + "contextStrategy": { + "sources": [ + { + "topic": "HEAVY_VALIDATION_RESULT", + "priority": "required", + "strategy": "latest", + "amount": 2 + } + ], + "format": "chronological", + "maxTokens": "{{max_tokens}}" + }, + "triggers": [ + { + "topic": "HEAVY_VALIDATION_RESULT", + "logic": { + "engine": "javascript", + "script": "const validators = ['validator-security', 'validator-tester']; const stageStart = ledger.findLast({ topic: 'QUICK_VALIDATION_PASSED' }); if (!stageStart) return false; const lastHeavyConsensus = ledger.findLast({ topic: 'VALIDATION_RESULT', since: stageStart.timestamp }); const since = Math.max(stageStart.timestamp, lastHeavyConsensus?.timestamp || 0); const responses = ledger.query({ topic: 'HEAVY_VALIDATION_RESULT', since }); const latestByValidator = new Map(); for (const response of responses) { if (validators.includes(response.sender)) { latestByValidator.set(response.sender, response); } } return validators.every((validatorId) => latestByValidator.has(validatorId));" + }, + "action": "execute_task" + } + ], + "hooks": { + "onComplete": { + "action": "publish_message", + "transform": { + "engine": "javascript", + "script": "return { topic: 'VALIDATION_RESULT', content: { text: result.allApproved ? 'All validations passed' : 'Stage 2 rejected', data: { approved: result.allApproved, stage: 'heavy', summary: result.summary, errors: ledger.query({ topic: 'HEAVY_VALIDATION_RESULT' }).flatMap(r => r.content?.data?.errors || []) } } };" + } + } + } + } + ] +} diff --git a/cluster-templates/base-templates/quick-validation.json b/cluster-templates/base-templates/quick-validation.json new file mode 100644 index 00000000..8025f5c8 --- /dev/null +++ b/cluster-templates/base-templates/quick-validation.json @@ -0,0 +1,285 @@ +{ + "name": "Quick Validation", + "description": "Stage 1: Requirements + Code validation. Fast feedback (30-60s).", + "params": { + "validator_level": { + "type": "string", + "enum": ["level1", "level2", "level3"], + "default": "level2" + }, + "max_tokens": { + "type": "number", + "default": 100000 + }, + "timeout": { + "type": "number", + "default": 0, + "description": "Task timeout in milliseconds (0 = no timeout)" + } + }, + "agents": [ + { + "id": "validator-requirements", + "role": "validator", + "modelLevel": "{{validator_level}}", + "timeout": "{{timeout}}", + "maxRetries": 3, + "outputFormat": "json", + "jsonSchema": { + "type": "object", + "properties": { + "approved": { + "type": "boolean" + }, + "summary": { + "type": "string" + }, + "errors": { + "type": "array", + "items": { + "type": "string" + } + }, + "criteriaResults": { + "type": "array", + "description": "Status for each acceptance criterion. PASS/FAIL require evidence. CANNOT_VALIDATE requires reason.", + "items": { + "type": "object", + "properties": { + "id": { + "type": "string", + "description": "AC1, AC2, etc. from plan" + }, + "status": { + "type": "string", + "enum": ["PASS", "FAIL", "SKIPPED", "CANNOT_VALIDATE"], + "description": "CANNOT_VALIDATE = verification impossible (missing tools, permissions, etc). Treated as PASS with warning." + }, + "evidence": { + "type": "object", + "description": "REQUIRED for PASS/FAIL. Proof of verification - actual command output.", + "properties": { + "command": { + "type": "string" + }, + "exitCode": { + "type": "integer" + }, + "output": { + "type": "string" + } + } + }, + "reason": { + "type": "string", + "description": "REQUIRED for CANNOT_VALIDATE. WHY verification is impossible (e.g., 'kubectl not installed', 'no SSH access')." + } + }, + "required": ["id", "status"] + } + } + }, + "required": ["approved", "summary"] + }, + "prompt": { + "system": "# REQUIREMENTS VALIDATOR\n\n## šŸš« YOU ARE A VALIDATOR - READ-ONLY MODE\n\nYour ONLY job is to VALIDATE and OUTPUT JSON.\n- NEVER use Edit, Write, or any file modification tools\n- NEVER fix code - REJECT with errors array instead\n- NEVER create files, modify files, or make commits\n\nVerify implementation meets ALL requirements from issue. Hold a HIGH BAR.\n\n## WORKFLOW\n1. Read context files (README, CONTRIBUTING, Makefile, package.json, etc.) for repo-specific validation\n2. Parse acceptanceCriteria from PLAN_READY\n3. For EACH criterion: run verification, record evidence\n4. If repo has validation script (validation script, make check, etc.), RUN IT\n\n## VERIFICATION\n- SEARCH before claiming 'missing' (Glob, Grep, Read)\n- RUN commands, capture output as evidence\n- CANNOT_VALIDATE only for: tool not installed, no network, permission denied\n\n## INSTANT REJECT\n- TODO/FIXME/placeholder = REJECT\n- Silent error swallowing = REJECT\n- 'Phase 2 deferred' = REJECT\n- 'Will add tests later' = REJECT\n- ANY priority=MUST criterion fails = REJECT\n\n## APPROVAL\n- approved:true = ALL MUST criteria pass + no blocking issues\n- approved:false = any MUST fails OR incomplete implementation\n\nšŸš« NO questions. Make safe choice and proceed.\n\n## šŸ”“ OUTPUT FORMAT (CRITICAL)\n\nYou MUST return valid JSON with these REQUIRED fields:\n```json\n{\n \"approved\": boolean,\n \"summary\": \"<100 chars max>\",\n \"errors\": [\"blocking issue 1\", \"blocking issue 2\"],\n \"criteriaResults\": [{\"id\": \"AC1\", \"status\": \"PASS|FAIL|CANNOT_VALIDATE\", \"evidence\": {\"command\": \"...\", \"exitCode\": 0, \"output\": \"<200 chars>\"}, \"reason\": \"for CANNOT_VALIDATE only\"}]\n}\n```\nNo preamble. JSON only." + }, + "contextStrategy": { + "sources": [ + { + "topic": "ISSUE_OPENED", + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "STATE_SNAPSHOT", + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "PLAN_READY", + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "IMPLEMENTATION_READY", + "priority": "high", + "since": "last_agent_start", + "strategy": "latest", + "amount": 1 + } + ], + "format": "chronological", + "maxTokens": "{{max_tokens}}" + }, + "triggers": [ + { + "topic": "IMPLEMENTATION_READY", + "action": "execute_task" + } + ], + "hooks": { + "onComplete": { + "action": "publish_message", + "config": { + "topic": "QUICK_VALIDATION_RESULT", + "content": { + "text": "{{result.summary}}", + "data": { + "approved": "{{result.approved}}", + "errors": "{{result.errors}}", + "criteriaResults": "{{result.criteriaResults}}", + "validatorId": "validator-requirements" + } + } + } + } + } + }, + { + "id": "validator-code", + "role": "validator", + "modelLevel": "{{validator_level}}", + "timeout": "{{timeout}}", + "maxRetries": 3, + "outputFormat": "json", + "jsonSchema": { + "type": "object", + "properties": { + "approved": { + "type": "boolean" + }, + "summary": { + "type": "string" + }, + "errors": { + "type": "array", + "items": { + "type": "string" + } + } + }, + "required": ["approved", "summary"] + }, + "prompt": { + "system": "# CODE VALIDATOR\n\n## šŸš« YOU ARE A VALIDATOR - READ-ONLY MODE\n\nYour ONLY job is to VALIDATE and OUTPUT JSON.\n- NEVER use Edit, Write, or any file modification tools\n- NEVER fix code - REJECT with errors array instead\n- NEVER create files, modify files, or make commits\n\nSenior engineer code review. Catch REAL bugs, not style preferences.\n\n## WORKFLOW\n1. Read context files (README, CONTRIBUTING, Makefile, package.json, etc.) for repo-specific validation\n2. SEARCH before claiming 'missing' (Glob, Grep, Read)\n3. RUN validation scripts if specified\n\n## INSTANT REJECT\n- TODO/FIXME/placeholder = REJECT\n- Silent error swallowing = REJECT\n- Dangerous fallbacks hiding failures = REJECT\n\n## šŸ”“ GENERALIZATION CHECK (CRITICAL)\nWorker fixed a bug? Verify they fixed ALL instances:\n1. Identify the PATTERN (not just the line)\n2. `grep -rn \"pattern\" .` - search codebase\n3. If N > 1 exists ā†’ Did worker fix ALL? If NO ā†’ REJECT\n\nExamples: null check in one handler? Check ALL. SQL injection in one query? Check ALL. A fix that leaves identical bugs elsewhere is NOT a fix.\n\n## šŸ”“ CLEAN DESIGN CHECK (CRITICAL)\nUnless the issue EXPLICITLY requests backwards compatibility, REJECT any of these:\n- Old function/class kept alongside new one (\"deprecated but still works\") ā†’ REJECT. Delete old, update ALL callers.\n- Re-exports or wrappers forwarding to new implementation ā†’ REJECT. Update imports at call sites.\n- `_unused` parameter renames to preserve old signatures ā†’ REJECT. Change the signature, update callers.\n- Fallback paths handling \"old format\" or \"legacy data\" ā†’ REJECT. Migrate the data, remove the fallback.\n- Comments like \"kept for backwards compatibility\" ā†’ REJECT. Remove the old code.\n- Feature flags toggling between old and new behavior ā†’ REJECT. Ship the new behavior. Delete the old.\n\nThe CLEAN solution: delete the old, update all callers, ship only the current implementation. A senior architect demolishes old scaffolding - they don't leave it standing \"just in case\".\n\n## BLOCKING (reject with WHAT/HOW/WHY)\n- Logic/off-by-one bugs\n- Race conditions\n- Security holes (injection, auth bypass)\n- Resource leaks (timers, connections)\n- God functions (>50 lines) - SPLIT\n- DRY violation (same logic 2+ places)\n- Missing error handling\n- Hardcoded values that should be config\n- Backwards compatibility shims or legacy wrappers (see CLEAN DESIGN CHECK)\n\n## NOT BLOCKING (summary only)\n- Style/naming preferences\n- 'Could theoretically...' without proof\n\nšŸš« NO questions. Make safe choice and proceed.\n\n## šŸ”“ OUTPUT FORMAT (CRITICAL)\n\nYou MUST return valid JSON:\n```json\n{\n \"approved\": boolean,\n \"summary\": \"<100 chars max>\",\n \"errors\": [\"WHAT: X. HOW: Y. WHY: Z\"]\n}\n```\nNo preamble. JSON only." + }, + "contextStrategy": { + "sources": [ + { + "topic": "ISSUE_OPENED", + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "STATE_SNAPSHOT", + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "PLAN_READY", + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "IMPLEMENTATION_READY", + "priority": "high", + "since": "last_agent_start", + "strategy": "latest", + "amount": 1 + } + ], + "format": "chronological", + "maxTokens": "{{max_tokens}}" + }, + "triggers": [ + { + "topic": "IMPLEMENTATION_READY", + "action": "execute_task" + } + ], + "hooks": { + "onComplete": { + "action": "publish_message", + "config": { + "topic": "QUICK_VALIDATION_RESULT", + "content": { + "text": "{{result.summary}}", + "data": { + "approved": "{{result.approved}}", + "errors": "{{result.errors}}", + "validatorId": "validator-code" + } + } + } + } + } + }, + { + "id": "consensus-coordinator", + "role": "coordinator", + "modelLevel": "level1", + "timeout": "{{timeout}}", + "outputFormat": "json", + "jsonSchema": { + "type": "object", + "properties": { + "allApproved": { + "type": "boolean" + }, + "summary": { + "type": "string" + } + }, + "required": ["allApproved", "summary"] + }, + "prompt": { + "system": "Check if both validators approved. Output: {\"allApproved\": boolean, \"summary\": \"<50 chars>\"}" + }, + "contextStrategy": { + "sources": [ + { + "topic": "QUICK_VALIDATION_RESULT", + "priority": "required", + "strategy": "latest", + "amount": 2 + } + ], + "format": "chronological", + "maxTokens": "{{max_tokens}}" + }, + "triggers": [ + { + "topic": "QUICK_VALIDATION_RESULT", + "logic": { + "engine": "javascript", + "script": "const lastImpl = ledger.findLast({ topic: 'IMPLEMENTATION_READY' }); const since = lastImpl?.timestamp || 0; return helpers.allResponded(['validator-requirements', 'validator-code'], 'QUICK_VALIDATION_RESULT', since);" + }, + "action": "execute_task" + } + ], + "hooks": { + "onComplete": { + "action": "publish_message", + "config": { + "topic": "QUICK_VALIDATION_PASSED", + "content": { + "text": "Stage 1 passed", + "data": {} + } + }, + "logic": { + "engine": "javascript", + "script": "if (!result.allApproved) { return { topic: 'VALIDATION_RESULT', content: { text: 'Stage 1 rejected', data: { approved: false, stage: 'quick', errors: ledger.query({ topic: 'QUICK_VALIDATION_RESULT' }).flatMap(r => r.content?.data?.errors || []) } } }; }" + } + } + } + } + ] +} diff --git a/cluster-templates/base-templates/single-worker.json b/cluster-templates/base-templates/single-worker.json index 93446801..6378b208 100644 --- a/cluster-templates/base-templates/single-worker.json +++ b/cluster-templates/base-templates/single-worker.json @@ -35,7 +35,15 @@ "sources": [ { "topic": "ISSUE_OPENED", - "limit": 1 + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "STATE_SNAPSHOT", + "priority": "required", + "strategy": "latest", + "amount": 1 } ], "format": "chronological", diff --git a/cluster-templates/base-templates/worker-validator.json b/cluster-templates/base-templates/worker-validator.json index f7a3aa19..16f23bd7 100644 --- a/cluster-templates/base-templates/worker-validator.json +++ b/cluster-templates/base-templates/worker-validator.json @@ -80,17 +80,29 @@ "sources": [ { "topic": "ISSUE_OPENED", - "limit": 1 + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "STATE_SNAPSHOT", + "priority": "required", + "strategy": "latest", + "amount": 1 }, { "topic": "WORKER_PROGRESS", + "priority": "medium", "since": "last_task_end", - "limit": 3 + "strategy": "latest", + "amount": 3 }, { "topic": "VALIDATION_RESULT", + "priority": "high", "since": "last_task_end", - "limit": 3 + "strategy": "latest", + "amount": 3 } ], "format": "chronological", @@ -166,17 +178,27 @@ "required": ["approved", "summary", "errors"] }, "prompt": { - "system": "## šŸ”“ OUTPUT FORMAT (CRITICAL - READ FIRST)\n\nYour output MUST be MINIMAL and STRUCTURED:\n- Output ONLY the required JSON schema fields\n- NO preambles (\"Here is my analysis...\", \"Let me explain...\")\n- NO verbose summaries - be CONCISE (max 100 chars per string field)\n- NO redundant information\n- NO explanations before or after the JSON\n\n## šŸš« YOU CANNOT ASK QUESTIONS\n\nYou are running non-interactively. There is NO USER to answer.\n- NEVER use AskUserQuestion tool\n- NEVER say \"Should I...\" or \"Would you like...\"\n- When unsure: Make the SAFER choice and proceed.\n\nYou are a validator for a SIMPLE {{task_type}} task.\n\n## šŸ”“ VERIFICATION PROTOCOL (REQUIRED - PREVENTS FALSE CLAIMS)\n\nBefore making ANY claim about missing functionality or code issues:\n\n1. **SEARCH FIRST** - Use Glob to find ALL relevant files\n2. **READ THE CODE** - Use Read to inspect actual implementation\n3. **GREP FOR PATTERNS** - Use Grep to search for specific code (function names, endpoints, etc.)\n\n**NEVER claim something doesn't exist without FIRST searching for it.**\n\nThe worker may have implemented features in different files than originally planned. If you claim '/api/metrics endpoint is missing' without searching, you may miss that it exists in 'server/routes/health.ts' instead of 'server/routes/api.ts'.\n\n### Example Verification Flow:\n```\n1. Claim: 'Missing error handling for network failures'\n2. BEFORE claiming ā†’ Grep for 'catch', 'error', 'try' in relevant files\n3. BEFORE claiming ā†’ Read the actual implementation\n4. ONLY IF NOT FOUND ā†’ Add to errors array\n```\n\n## VALIDATION CRITERIA\n\n**APPROVE** if:\n- Core functionality works as requested\n- Implementation is correct and complete\n- No obvious bugs or critical issues\n\n**REJECT** if:\n- Major functionality is missing or broken (VERIFIED by searching)\n- Implementation doesn't match requirements (VERIFIED by reading code)\n- Critical bugs present (VERIFIED by inspection)\n\n## TASK TYPE: {{task_type}}\n\n{{#if task_type == 'TASK'}}\nVerify the feature/change works correctly.\n{{/if}}\n\n{{#if task_type == 'DEBUG'}}\nVerify the bug is actually fixed at root cause.\n{{/if}}\n\n{{#if task_type == 'INQUIRY'}}\nVerify the information is accurate and complete.\n{{/if}}\n\nFor SIMPLE tasks, don't nitpick. Focus on: Does it work and meet requirements?\n\n## šŸ”“ DEBUGGING METHODOLOGY CHECK\n\nBefore approving, verify the worker didn't take shortcuts:\n\n### Ad Hoc Fix Detection\n- Did worker fix ONE instance? ā†’ Grep for similar patterns. If N > 1 exists, REJECT.\n- Example: Fixed null check in `auth.ts:42`? ā†’ `grep -r \"similar pattern\" .` - are there others?\n\n### Root Cause vs Symptom\n- Did worker add a workaround? ā†’ Find the ACTUAL bug. If workaround hides real issue, REJECT.\n- Example: Added `|| []` fallback? ā†’ WHY is it undefined? Fix THAT.\n\n### Lazy Debugging Red Flags (INSTANT REJECT)\n- Worker suggests \"restart the service\" ā†’ REJECT (hides the bug)\n- Worker suggests \"clear the cache\" ā†’ REJECT (hides the bug)\n- Worker says \"works on my machine\" ā†’ REJECT (not a fix)\n- Worker blames the test ā†’ REJECT unless they PROVE test is wrong with evidence" + "system": "## šŸ”“ OUTPUT FORMAT (CRITICAL - READ FIRST)\n\nYour output MUST be MINIMAL and STRUCTURED:\n- Output ONLY the required JSON schema fields\n- NO preambles (\"Here is my analysis...\", \"Let me explain...\")\n- NO verbose summaries - be CONCISE (max 100 chars per string field)\n- NO redundant information\n- NO explanations before or after the JSON\n\n## šŸš« YOU CANNOT ASK QUESTIONS\n\nYou are running non-interactively. There is NO USER to answer.\n- NEVER use AskUserQuestion tool\n- NEVER say \"Should I...\" or \"Would you like...\"\n- When unsure: Make the SAFER choice and proceed.\n\nYou are a validator for a SIMPLE {{task_type}} task.\n\n## šŸ”“ VERIFICATION PROTOCOL (REQUIRED - PREVENTS FALSE CLAIMS)\n\nBefore making ANY claim about missing functionality or code issues:\n\n1. **SEARCH FIRST** - Use Glob to find ALL relevant files\n2. **READ THE CODE** - Use Read to inspect actual implementation\n3. **GREP FOR PATTERNS** - Use Grep to search for specific code (function names, endpoints, etc.)\n\n**NEVER claim something doesn't exist without FIRST searching for it.**\n\nThe worker may have implemented features in different files than originally planned. If you claim '/api/metrics endpoint is missing' without searching, you may miss that it exists in 'server/routes/health.ts' instead of 'server/routes/api.ts'.\n\n### Example Verification Flow:\n```\n1. Claim: 'Missing error handling for network failures'\n2. BEFORE claiming ā†’ Grep for 'catch', 'error', 'try' in relevant files\n3. BEFORE claiming ā†’ Read the actual implementation\n4. ONLY IF NOT FOUND ā†’ Add to errors array\n```\n\n## šŸ”“ CLEAN DESIGN CHECK (CRITICAL)\nUnless the issue EXPLICITLY requests backwards compatibility, REJECT any of these:\n- Old function/class kept alongside new one (\"deprecated but still works\") ā†’ REJECT. Delete old, update ALL callers.\n- Re-exports or wrappers forwarding to new implementation ā†’ REJECT. Update imports at call sites.\n- `_unused` parameter renames to preserve old signatures ā†’ REJECT. Change the signature, update callers.\n- Fallback paths handling \"old format\" or \"legacy data\" ā†’ REJECT. Migrate the data, remove the fallback.\n- Comments like \"kept for backwards compatibility\" ā†’ REJECT. Remove the old code.\n- Feature flags toggling between old and new behavior ā†’ REJECT. Ship the new behavior. Delete the old.\n\nThe CLEAN solution: delete the old, update all callers, ship only the current implementation.\n\n## VALIDATION CRITERIA\n\n**APPROVE** if:\n- Core functionality works as requested\n- Implementation is correct and complete\n- No obvious bugs or critical issues\n- No backwards compatibility cruft (unless issue explicitly requests it)\n\n**REJECT** if:\n- Major functionality is missing or broken (VERIFIED by searching)\n- Implementation doesn't match requirements (VERIFIED by reading code)\n- Critical bugs present (VERIFIED by inspection)\n- Backwards compatibility shims, legacy wrappers, or deprecation patterns present (see CLEAN DESIGN CHECK)\n\n## TASK TYPE: {{task_type}}\n\n{{#if task_type == 'TASK'}}\nVerify the feature/change works correctly.\n{{/if}}\n\n{{#if task_type == 'DEBUG'}}\nVerify the bug is actually fixed at root cause.\n{{/if}}\n\n{{#if task_type == 'INQUIRY'}}\nVerify the information is accurate and complete.\n{{/if}}\n\nFor SIMPLE tasks, don't nitpick. Focus on: Does it work and meet requirements?\n\n## šŸ”“ DEBUGGING METHODOLOGY CHECK\n\nBefore approving, verify the worker didn't take shortcuts:\n\n### Ad Hoc Fix Detection\n- Did worker fix ONE instance? ā†’ Grep for similar patterns. If N > 1 exists, REJECT.\n- Example: Fixed null check in `auth.ts:42`? ā†’ `grep -r \"similar pattern\" .` - are there others?\n\n### Root Cause vs Symptom\n- Did worker add a workaround? ā†’ Find the ACTUAL bug. If workaround hides real issue, REJECT.\n- Example: Added `|| []` fallback? ā†’ WHY is it undefined? Fix THAT.\n\n### Lazy Debugging Red Flags (INSTANT REJECT)\n- Worker suggests \"restart the service\" ā†’ REJECT (hides the bug)\n- Worker suggests \"clear the cache\" ā†’ REJECT (hides the bug)\n- Worker says \"works on my machine\" ā†’ REJECT (not a fix)\n- Worker blames the test ā†’ REJECT unless they PROVE test is wrong with evidence" }, "contextStrategy": { "sources": [ { "topic": "ISSUE_OPENED", - "limit": 1 + "priority": "required", + "strategy": "latest", + "amount": 1 + }, + { + "topic": "STATE_SNAPSHOT", + "priority": "required", + "strategy": "latest", + "amount": 1 }, { "topic": "IMPLEMENTATION_READY", - "limit": 1 + "priority": "high", + "strategy": "latest", + "amount": 1 } ], "format": "chronological", diff --git a/cluster-templates/conductor-bootstrap.json b/cluster-templates/conductor-bootstrap.json index f08106bf..2ec3fc4e 100644 --- a/cluster-templates/conductor-bootstrap.json +++ b/cluster-templates/conductor-bootstrap.json @@ -5,7 +5,7 @@ { "id": "junior-conductor", "role": "conductor", - "modelLevel": "level1", + "modelLevel": "level2", "useDirectApi": true, "outputFormat": "json", "jsonSchema": { @@ -32,7 +32,9 @@ "system": "You are the JUNIOR CONDUCTOR - fast task classification.\n\n## Your Job\nClassify the task on TWO dimensions.\n\n## šŸ”“ COST REMINDER\n- CRITICAL uses Opus ($15/M tokens) + 4 validators = EXPENSIVE\n- STANDARD uses Sonnet ($3/M tokens) + 2 validators = NORMAL\n- Don't waste money on false positives. CRITICAL is rare.\n\n## COMPLEXITY (pick ONE)\n- TRIVIAL - One file, mechanical change; no behavior change.\n- SIMPLE - Small change, 1-2 files, low risk.\n- STANDARD - Multi-file work or user-visible behavior. **DEFAULT CHOICE.**\n- CRITICAL - ONLY when code DIRECTLY modifies: (1) authentication/authorization LOGIC, (2) payment processing/billing calculations, (3) secrets/credentials handling, (4) destructive database operations (DROP, DELETE), (5) production deployment or live infrastructure, (6) PII processing (not just displaying it).\n- UNCERTAIN - Escalate to senior conductor.\n\n**šŸ”“ BIAS: If unsure between STANDARD and CRITICAL, choose STANDARD.** CRITICAL is expensive. Reserve it for actual risk.\n\n## NOT CRITICAL (Common False Positives)\n\nThese are STANDARD, not CRITICAL:\n- Refactoring code that MENTIONS auth/billing/security (not MODIFYING the logic)\n- Adding TypeScript types for existing structures\n- Code cleanup in infra-related files\n- Read-only queries to production data\n- Tests for auth/billing code (tests don't touch prod)\n- Extracting modules or services (code organization)\n- Factory patterns, dependency injection (architecture)\n- Config file reorganization (not production config values)\n\n## TASK TYPE (pick ONE)\n- INQUIRY - Questions, exploration, read-only\n- TASK - Implement something new\n- DEBUG - Fix something broken\n\n## Examples\n\nTask: \"Explain current auth flow (read-only)\"\n```json\n{\"complexity\": \"SIMPLE\", \"taskType\": \"INQUIRY\", \"reasoning\": \"Read-only explanation\"}\n```\n\nTask: \"Refactor auth service into smaller modules\"\n```json\n{\"complexity\": \"STANDARD\", \"taskType\": \"TASK\", \"reasoning\": \"Refactoring code organization, not modifying auth logic\"}\n```\n\nTask: \"Add TypeScript types to payment types\"\n```json\n{\"complexity\": \"STANDARD\", \"taskType\": \"TASK\", \"reasoning\": \"Adding types, not modifying billing logic\"}\n```\n\nTask: \"Fix bug in password validation logic\"\n```json\n{\"complexity\": \"CRITICAL\", \"taskType\": \"DEBUG\", \"reasoning\": \"Directly modifying authentication logic\"}\n```\n\nTask: \"Add new payment method integration\"\n```json\n{\"complexity\": \"CRITICAL\", \"taskType\": \"TASK\", \"reasoning\": \"New billing/payment processing code\"}\n```\n\nTask: \"Rotate production API keys\"\n```json\n{\"complexity\": \"CRITICAL\", \"taskType\": \"TASK\", \"reasoning\": \"Modifying production secrets\"}\n```\n\nTask: \"DROP TABLE users migration\"\n```json\n{\"complexity\": \"CRITICAL\", \"taskType\": \"TASK\", \"reasoning\": \"Destructive database operation\"}\n```\n\n## Critical Rules\n1. Output ONLY valid JSON - no other text\n2. complexity must be EXACTLY one of: TRIVIAL, SIMPLE, STANDARD, CRITICAL, UNCERTAIN\n3. taskType must be EXACTLY one of: INQUIRY, TASK, DEBUG\n\nTask: {{ISSUE_OPENED.content.text}}" }, "contextStrategy": { - "sources": [{ "topic": "ISSUE_OPENED", "limit": 1 }], + "sources": [ + { "topic": "ISSUE_OPENED", "priority": "required", "strategy": "latest", "amount": 1 } + ], "format": "chronological", "maxTokens": 100000 }, @@ -59,7 +61,7 @@ { "id": "senior-conductor", "role": "conductor", - "modelLevel": "level2", + "modelLevel": "level3", "useDirectApi": true, "outputFormat": "json", "jsonSchema": { @@ -87,16 +89,20 @@ }, "contextStrategy": { "sources": [ - { "topic": "ISSUE_OPENED", "limit": 1 }, + { "topic": "ISSUE_OPENED", "priority": "required", "strategy": "latest", "amount": 1 }, { "topic": "CONDUCTOR_ESCALATE", + "priority": "high", "since": "last_agent_start", - "limit": 1 + "strategy": "latest", + "amount": 1 }, { "topic": "CLUSTER_OPERATIONS_VALIDATION_FAILED", + "priority": "high", "since": "cluster_start", - "limit": 3 + "strategy": "latest", + "amount": 3 } ], "format": "chronological", diff --git a/codecov.yml b/codecov.yml index bed7d4a6..9c258698 100644 --- a/codecov.yml +++ b/codecov.yml @@ -9,6 +9,6 @@ coverage: target: auto comment: - layout: "reach,diff,flags,files" + layout: 'reach,diff,flags,files' behavior: default require_changes: true diff --git a/docker/zeroshot-cluster/Dockerfile b/docker/zeroshot-cluster/Dockerfile index 9e39221b..2929bc38 100644 --- a/docker/zeroshot-cluster/Dockerfile +++ b/docker/zeroshot-cluster/Dockerfile @@ -18,6 +18,7 @@ ARG HELM_VERSION=3.13.3 ARG INFRACOST_VERSION=0.10.32 ARG TFLINT_VERSION=0.50.0 ARG TFSEC_VERSION=1.28.4 +ARG CLAUDE_CODE_VERSION=2.1.20 # Install system dependencies for e2e testing and development RUN apt-get update && apt-get install -y --no-install-recommends \ @@ -127,8 +128,8 @@ COPY docker/zeroshot-cluster/pre-baked-deps.json /pre-baked-deps/package.json RUN cd /pre-baked-deps && npm install --ignore-scripts \ && chown -R node:node /pre-baked-deps -# Install Claude CLI globally -RUN npm install -g @anthropic-ai/claude-code +# Install Claude CLI globally (version pinned via ARG) +RUN npm install -g @anthropic-ai/claude-code@${CLAUDE_CODE_VERSION} # Install Playwright (uses system Chromium) RUN npx playwright install-deps chromium 2>/dev/null || true diff --git a/docs/LIVE_MONITORING_PLAN.md b/docs/LIVE_MONITORING_PLAN.md deleted file mode 100644 index aa4aca0a..00000000 --- a/docs/LIVE_MONITORING_PLAN.md +++ /dev/null @@ -1,233 +0,0 @@ -# Live Monitoring for Zeroshot - -## Problem Statement - -When running `zeroshot run` or `zeroshot task`, users can only see agent output. They have no visibility into: -- Which workers are active vs idle -- CPU/memory usage per worker -- Network activity (API calls in progress) -- Whether a worker is stuck or making progress - -## Design Goals - -1. **Always visible** - Status footer shown during ALL zeroshot executions (not just attach) -2. **Non-intrusive** - Footer doesn't disrupt terminal output scrolling -3. **Real-time** - Update every 1-2 seconds -4. **Low overhead** - Minimal CPU cost for monitoring itself -5. **Cross-platform** - Linux-first (/proc), graceful degradation on macOS - -## Architecture Options - -### Option A: Status Bar in AttachClient (Recommended for MVP) - -Add a persistent header showing metrics for the attached agent: - -``` -ā”Œā”€ worker [sonnet] ā”€ CPU: 12% ā”‚ Mem: 45MB ā”‚ Net: ā†“2.1KB/s ā†‘0.3KB/s ā”‚ Tokens: 1.2K ā”€ā” -ā”‚ ... terminal output ... ā”‚ -ā””ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”˜ -``` - -**Pros:** -- Simple implementation -- Natural extension of attach -- Single agent focus - -**Cons:** -- Only one agent visible -- Requires terminal manipulation - -### Option B: Enhance `zeroshot watch` (Recommended for Full Solution) - -Extend existing TUI dashboard with per-agent metrics: - -``` -ā”Œā”€ Cluster: cosmic-meteor-87 ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā” -ā”‚ ā”‚ -ā”‚ AGENT STATE CPU MEM NET I/O TOKENS LAST OUTPUT ā”‚ -ā”‚ ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ ā”‚ -ā”‚ worker executing 23% 67MB ā†“4.2K ā†‘1.1K 3.4K 2s ago ā”‚ -ā”‚ validator-1 idle 0% 12MB - 0 waiting ā”‚ -ā”‚ validator-2 idle 0% 12MB - 0 waiting ā”‚ -ā”‚ ā”‚ -ā”‚ [CPU CHART] [NETWORK CHART] ā”‚ -ā”‚ ā–ˆā–ˆā–ˆā–ˆā–ˆā–ˆā–ˆā–‘ā–‘ā–‘ā–‘ā–‘ā–‘ā–‘ā–‘ā–‘ā–‘ā–‘ā–‘ 23% ā–ā–‚ā–ƒā–…ā–‡ā–ˆā–…ā–ƒā–‚ā– 4.2KB/s ā”‚ -ā”‚ ā”‚ -ā””ā”€ Press 'q' to quit ā”‚ 'a' to attach ā”‚ 'l' for logs ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”˜ -``` - -**Pros:** -- All agents visible -- Charts for trends -- Interactive (attach from dashboard) - -**Cons:** -- More complex -- Can't see terminal output simultaneously - -### Option C: Side-by-Side Split (Future) - -Terminal multiplexing with tmux-style panes. - -## Implementation Phases - -### Phase 1: Process Metrics Module (1-2 hours) - -Create `src/process-metrics.js`: - -```javascript -/** - * Get real-time metrics for a process and its children - * @param {number} pid - Process ID - * @param {number} [samplePeriodMs=1000] - Sampling period for rate calculations - * @returns {Promise} - */ -async function getProcessMetrics(pid, samplePeriodMs = 1000) { - // Read /proc/{pid}/stat for CPU - // Read /proc/{pid}/status for memory - // Read /proc/{pid}/io for I/O rates - // Use ss for network state - // Aggregate across child processes -} - -interface ProcessMetrics { - pid: number; - cpuPercent: number; - memoryMB: number; - ioReadBytesPerSec: number; - ioWriteBytesPerSec: number; - networkState: { - established: number; - sendQueueBytes: number; - recvQueueBytes: number; - }; - childCount: number; -} -``` - -**Leverage existing code:** -- `agent-stuck-detector.js` already has `getProcessState()` and `getNetworkState()` -- Refactor into reusable module - -### Phase 2: AttachClient Status Bar (2-3 hours) - -Modify `src/attach/attach-client.js`: - -1. Add metrics polling interval (every 1s) -2. Render status bar using ANSI escape codes -3. Handle terminal resize to reposition bar -4. Add Ctrl+B m to toggle metrics display - -```javascript -class AttachClient extends EventEmitter { - constructor(options) { - // ... existing code ... - this.showMetrics = options.showMetrics ?? true; - this.metricsInterval = null; - } - - _startMetricsPolling() { - this.metricsInterval = setInterval(async () => { - const metrics = await getProcessMetrics(this.processPid); - this._renderStatusBar(metrics); - }, 1000); - } - - _renderStatusBar(metrics) { - // Save cursor position - // Move to line 1 - // Clear line - // Write formatted metrics - // Restore cursor position - } -} -``` - -### Phase 3: Enhanced `zeroshot watch` (3-4 hours) - -Extend `src/tui/dashboard.js`: - -1. Add metrics column to agent table -2. Add sparkline charts for CPU/network -3. Poll metrics for all agents in cluster -4. Add 'a' key to attach to selected agent - -```javascript -// New component: AgentMetricsTable -const table = blessed.listtable({ - headers: ['AGENT', 'STATE', 'CPU', 'MEM', 'NET', 'TOKENS', 'LAST'], - data: agents.map(a => [ - a.id, - a.state, - `${a.metrics.cpuPercent}%`, - `${a.metrics.memoryMB}MB`, - formatNetworkRate(a.metrics), - a.tokenCount || '?', - formatTimeSince(a.lastOutputTime) - ]) -}); -``` - -## File Changes - -| File | Change | -|------|--------| -| `src/process-metrics.js` | NEW - Metrics collection module | -| `src/attach/attach-client.js` | Add status bar rendering | -| `src/tui/dashboard.js` | Add metrics table and charts | -| `src/agent-wrapper.js` | Expose PID to callers | -| `cli/index.js` | Add `--no-metrics` flag to attach | - -## Testing Strategy - -1. **Unit tests** for process-metrics.js - - Mock /proc filesystem - - Test rate calculations - - Test child process aggregation - -2. **Integration tests** - - Spawn real process, verify metrics - - Test with Claude CLI running - -3. **Visual testing** - - Manual verification of TUI rendering - - Test terminal resize handling - -## Platform Support - -| Platform | CPU/Mem | Network | I/O | Support Level | -|----------|---------|---------|-----|---------------| -| Linux | `/proc/stat` | `ss -tunp` | `/proc/io` | Full | -| macOS | `ps -o %cpu,rss` | `lsof -i` | N/A (requires sudo) | Full (no I/O) | -| Windows WSL | `/proc/stat` | `ss -tunp` | `/proc/io` | Full | - -```javascript -// Platform-aware metrics collection -function getProcessMetrics(pid) { - if (process.platform === 'darwin') { - return getMetricsDarwin(pid); // ps + lsof - } - return getMetricsLinux(pid); // /proc + ss -} -``` - -## Open Questions - -1. **Token counting** - Can we parse Claude CLI output to track token usage? - - Yes, stream-json has `usage` events - -2. **Network traffic attribution** - Per-process network bytes? - - Not easily without eBPF - - Use socket queue sizes as proxy for activity - -3. **Historical data** - Keep history for charts? - - Ring buffer of last 60 samples (1 minute) - - Store in memory, not persisted - -## Success Criteria - -- [ ] `zeroshot attach` shows live metrics without disrupting output -- [ ] `zeroshot watch` shows all agents with metrics -- [ ] Metrics update every 1 second -- [ ] No noticeable CPU overhead (<1% additional) -- [ ] Works on Linux and degrades gracefully on macOS diff --git a/docs/ZEROSHOT-DISRUPTIVE-TUI-DECISIONS.md b/docs/ZEROSHOT-DISRUPTIVE-TUI-DECISIONS.md new file mode 100644 index 00000000..72238615 --- /dev/null +++ b/docs/ZEROSHOT-DISRUPTIVE-TUI-DECISIONS.md @@ -0,0 +1,72 @@ +# Zeroshot Disruptive TUI ā€” Pre-M3 Decisions + +Date: 2026-02-01 +Status: Accepted (M3 baseline) + +## Context + +M3 (Live Cluster Canvas v1) needs stable interaction decisions to avoid reworking navigation and rendering. These choices set the baseline for M3 implementation and can be revised post-M3. + +## Decisions + +### 1) Focus model + +**Decision:** Explicit focus ring navigation. + +- Focus moves only via explicit keys (Tab/Shift-Tab, Left/Right where applicable). +- The focused pane is visually distinct (border + title treatment); unfocused panes do not capture navigation inputs. +- No automatic ā€œnearestā€ focus selection based on cursor position or layout changes. + +**Rationale:** Predictable keyboard navigation across resizes/layout tweaks; reduces accidental focus shifts and layout-driven churn. + +**Revisit if:** We add pointer/hover interactions or multi-focus gestures that justify spatial focus heuristics. + +### 2) Label strategy + +**Decision:** Always show stable identifiers; no hover/tooltip dependence. + +- Primary labels show stable IDs (cluster id, agent id/role) everywhere labels appear. +- Long labels are truncated with ellipsis; detailed metadata is shown in a detail pane or status line. +- Hover/tooltips are not a dependency (terminal UX). + +**Rationale:** Terminal UIs lack reliable hover, and IDs are the most unambiguous cross-view reference. + +**Revisit if:** We add a persistent details panel or toggled ā€œverbose labelsā€ mode that can replace truncation. + +### 3) Topology fidelity + +**Decision:** Stable, semantic layout (deterministic), not force-directed. + +- Use deterministic ordering and semantic positioning (e.g., workflow tiers or adjacency ordering). +- Avoid physics-based layouts that jitter across updates. + +**Rationale:** Stability beats visual fidelity in a text terminal; jitter breaks scanning and makes deltas hard to read. + +**Revisit if:** We add a dedicated ā€œexploreā€ mode with user-controlled layout and persistence. + +### 4) Scrub semantics + +**Decision:** Default scrub scope is per-cluster; per-agent in agent-focused views. + +- Cluster screen: scrub/scroll/timeline navigation applies to the cluster aggregate view. +- Agent screen: scrub/scroll is per-agent by default. +- Scope never switches implicitly; it follows the active screen/focus. + +**Rationale:** Matches user intent in each view and avoids hidden scope changes during navigation. + +**Revisit if:** We add multi-pane synchronized scrubbing or explicit cross-filtering controls. + +### 5) Spine height + +**Decision:** Strict 1-line spine; no persistent second hint line. + +- Core spine UI remains one line for maximum vertical space. +- Hints and transient guidance use the toast/status area instead of expanding the spine. + +**Rationale:** Preserves vertical space for live data and avoids layout reflow during bursts of guidance. + +**Revisit if:** Onboarding or accessibility testing shows a persistent hint line materially improves success rates. + +## Decision Outputs (for M3 issues) + +All M3 implementation issues should reference these decisions and keep behavior aligned unless a follow-up ADR explicitly revises them. diff --git a/docs/context-management.md b/docs/context-management.md new file mode 100644 index 00000000..4f33c7fb --- /dev/null +++ b/docs/context-management.md @@ -0,0 +1,182 @@ +# Context Management + +Zeroshot builds agent prompts from ledger history with explicit selection rules, a token budget, +and a durable state snapshot. This document explains the technology and how to configure it in +templates. + +## Technology stack + +- Ledger: SQLite message log per cluster (`~/.zeroshot/.db`). +- MessageBus: publish/subscribe wrapper over the ledger. +- AgentContextBuilder: assembles prompt sections and context sources. +- ContextPackBuilder: priority and budget based selection with compact variants. +- StateSnapshotter: derives `STATE_SNAPSHOT` from structured outputs. +- Context metrics: optional logs or ledger entries for observability. + +## Data flow + +```mermaid +flowchart LR + Issue[ISSUE_OPENED] -->|publish| Ledger[(SQLite ledger)] + Ledger --> Bus[MessageBus] + Bus --> Snapshotter[StateSnapshotter] + Snapshotter -->|STATE_SNAPSHOT| Ledger + Bus --> Builder[AgentContextBuilder] + Builder --> Packs[ContextPackBuilder] + Packs --> Prompt[Final agent context] + Prompt --> Agent[Agent execution] + Agent -->|publish topics| Ledger +``` + +## Prompt assembly sections + +The builder assembles the prompt as a set of packs: + +- Header and non-interactive rules +- Agent instructions and output schema +- Source sections from `contextStrategy.sources` +- Validator skip hints (validator role only) +- Triggering message (always included at the end) + +## Context strategy sources + +`contextStrategy.sources` controls which ledger messages are pulled into an agent prompt. + +| Field | Type | Purpose | Default | +| --------------- | --------------------------------- | --------------------------- | ----------------------------------------- | +| topic | string | Ledger topic name | required | +| sender | string | Filter by sender | none | +| since | string or timestamp | Lower bound for timestamps | none | +| strategy | latest \| oldest \| all | Selection semantics | latest if amount set, else all | +| amount | number | Max messages to include | none | +| limit | number | Deprecated alias for amount | none | +| priority | required \| high \| medium \| low | Pack priority | derived if missing | +| compactAmount | number | Amount for compact mode | 1 | +| compactStrategy | latest \| oldest \| all | Compact selection | latest if base strategy is all, else base | + +`since` accepts: `cluster_start`, `last_task_end`, `last_agent_start`, or an ISO timestamp string. + +### Selection semantics + +- `latest`: query DESC with limit, then reverse to render chronologically. +- `oldest`: query ASC with limit. +- `all`: query ASC with no limit (or a hard cap if amount is set). + +### Priority defaults + +If `priority` is missing, the builder assigns: + +- required: `STATE_SNAPSHOT`, `ISSUE_OPENED`, `PLAN_READY` +- high: `VALIDATION_RESULT`, `IMPLEMENTATION_READY` +- medium: everything else + +Templates should set priority explicitly for clarity. + +## Context packs and budgeting + +Context is built as a set of packs that are selected under a token budget. Each pack can provide +full and compact text. Required packs are always included first. + +```mermaid +flowchart TD + Start --> Required[Include required packs] + Required --> Optional[Evaluate optional packs by priority] + Optional --> Fits{Fits budget?} + Fits -- yes --> Include[Include full pack] + Fits -- no --> Compact{Compact fits?} + Compact -- yes --> IncludeCompact[Include compact pack] + Compact -- no --> Skip[Skip pack] + Include --> Next[Next pack] + IncludeCompact --> Next + Skip --> Next + Next --> MaxGuard[Apply max chars guard] + MaxGuard --> Done[Final context] +``` + +Budgeting details: + +- Token estimates use `estimateTokensFromChars` (chars / 4, rounded up). +- `maxTokens` controls selection; default is 100000 if unset. +- A defensive max chars guard caps the final context to 500000 chars. +- Required packs are preserved; if the max chars guard triggers, optional packs are compacted or + dropped first, then required packs are truncated as a last resort. + +## STATE_SNAPSHOT (durable working memory) + +`STATE_SNAPSHOT` is a structured summary of current cluster state. It is derived from ledger +messages and republished whenever relevant updates occur. + +### Update triggers + +State is updated from these topics: + +- `ISSUE_OPENED` +- `PLAN_READY` +- `WORKER_PROGRESS` +- `IMPLEMENTATION_READY` +- `VALIDATION_RESULT` +- `INVESTIGATION_COMPLETE` + +```mermaid +sequenceDiagram + participant Agent + participant Bus as MessageBus + participant Snapshotter + participant Ledger + Agent->>Bus: publish PLAN_READY + Bus->>Snapshotter: deliver PLAN_READY + Snapshotter->>Bus: publish STATE_SNAPSHOT + Bus->>Ledger: persist STATE_SNAPSHOT +``` + +Include `STATE_SNAPSHOT` as a required context source for workers, planners, and validators. + +## Context metrics and observability + +Enable context metrics: + +- `ZEROSHOT_CONTEXT_METRICS=1` prints metrics to stdout. +- `ZEROSHOT_CONTEXT_METRICS_LEDGER=1` publishes a `CONTEXT_METRICS` message. + +Metrics include: + +- total chars and estimated tokens +- per section and pack breakdown +- budget and truncation details + +## Example context strategy + +```json +{ + "contextStrategy": { + "sources": [ + { "topic": "ISSUE_OPENED", "priority": "required", "strategy": "latest", "amount": 1 }, + { "topic": "STATE_SNAPSHOT", "priority": "required", "strategy": "latest", "amount": 1 }, + { "topic": "PLAN_READY", "priority": "required", "strategy": "latest", "amount": 1 }, + { + "topic": "VALIDATION_RESULT", + "priority": "high", + "since": "last_task_end", + "strategy": "latest", + "amount": 5, + "compactAmount": 1 + }, + { + "topic": "WORKER_PROGRESS", + "priority": "medium", + "since": "last_task_end", + "strategy": "latest", + "amount": 3, + "compactAmount": 1 + } + ], + "maxTokens": 100000 + } +} +``` + +## Troubleshooting + +- Missing anchors: ensure required sources are present and priority is set to required. +- Stale messages: verify `strategy: \"latest\"` and that `amount` is set. +- Over budget: lower `maxTokens` or add `compactAmount` to optional sources. diff --git a/docs/postmortems/2026-01-31-pr-base-detached.md b/docs/postmortems/2026-01-31-pr-base-detached.md new file mode 100644 index 00000000..5eb96465 --- /dev/null +++ b/docs/postmortems/2026-01-31-pr-base-detached.md @@ -0,0 +1,49 @@ +# Postmortem: Detached PR base dropped (2026-01-31) + +## Timeline + +- 2026-01-31 13:45Z: Fix for detached `--pr-base` merged (PR #258). +- 2026-01-31 14:00Z: PR #259 created for issue #250. +- 2026-01-31 14:00Zā€“14:20Z: PR branch repeatedly rebased/pushed with commits from `main`. +- 2026-01-31 14:33Z: PR #259 merged after manual cleanup. + +## Impact + +- PR #259 included many unrelated commits from `main`, confusing review and CI. +- Time lost cleaning the branch and coordinating cluster shutdown. +- Reduced trust in automated PR creation. + +## Root Cause + +- Detached (`-d`) cluster runs did not forward `--pr-base` and related PR options to the daemon. +- The daemon therefore defaulted to `main` for rebase/PR base, even when `--pr-base dev` was supplied. + +## Contributing Factors + +- Cluster was started from a local checkout that did not include the 13:45Z fix yet. +- No single source of truth for run options in daemon mode; new flags required manual env wiring. +- No guardrail to alert when daemon options differ from CLI intent. + +## Detection + +- User observed PR #259 contained unrelated commits and questioned the base branch. + +## Resolution + +- Killed the active cluster to stop further pushes. +- Rebuilt the PR branch from a clean `origin/dev` base and re-opened PR #259 on `dev`. +- Merged fix PR #258 to preserve `--pr-base` in detached runs. + +## Prevention + +- Forward all run options via a single `ZEROSHOT_RUN_OPTIONS` payload in daemon mode. +- Parse daemon run options in `buildStartOptions` as a fallback for any missing CLI flags. +- Add unit coverage to ensure env-run-options are honored. +- Update operator docs to explicitly require forwarding options for detached runs. + +## Action Items + +- [x] Preserve `--pr-base`/`--merge-queue`/`--close-issue` in detached runs (PR #258). +- [x] Document daemon option forwarding in `AGENTS.md` and `CLAUDE.md`. +- [x] Add `ZEROSHOT_RUN_OPTIONS` fallback for daemon runs and unit tests. +- [ ] Add a preflight warning when local branch is behind remote base (optional follow-up). diff --git a/docs/tui-v2/IMPLEMENTATION_PLAN.md b/docs/tui-v2/IMPLEMENTATION_PLAN.md new file mode 100644 index 00000000..e3bc66d3 --- /dev/null +++ b/docs/tui-v2/IMPLEMENTATION_PLAN.md @@ -0,0 +1,583 @@ +# Zeroshot TUI v2 (Ink) - Multi-Stage Implementation Plan + +Date: 2026-01-25 +Status: Draft plan (intended to become a chain of GitHub issues) + +## Guiding Constraints + +- Ink + TypeScript for all new TUI code. +- Do not refactor unrelated JS during the TUI project. + - Small, targeted extraction is allowed only when it reduces duplication for TUI integration (e.g. shared `run` helpers). +- Replace the current dashboard feature (`zeroshot watch` / `src/tui/*`) with a new TUI experience launched by `zeroshot` (no args). +- Design/visual polish is explicitly deferred; prioritize UX flows and correctness. + +## Proposed Technical Approach (Replace Existing `src/tui`) + +### Code organization (proposed) + +- `src/tui/` (TypeScript source, Ink components) - replaces the existing blessed dashboard implementation + - `src/tui/app.tsx` - top-level Ink app, router, global keybindings + - `src/tui/views/*` - launcher/monitor/cluster/agent views + - `src/tui/commands/*` - slash command parser + dispatch + - `src/tui/services/*` - adapters around existing JS orchestrator/ledger + - `src/tui/domain/*` - typed models (ClusterRow, AgentRow, TimelineEvent, etc) +- `lib/tui/` (compiled JS output shipped in npm package) + +Rationale: + +- Keeps the existing JS runtime intact. +- Allows incremental TS adoption with minimal surface area. +- Avoids a repo-wide build migration. + +### Build strategy (minimal TS emission) + +Add a dedicated `tsconfig.tui.json` that: + +- includes only `src/tui/**/*.ts(x)` +- emits CJS output to `lib/tui/` +- keeps the existing `tsconfig.json` as "check JS, no emit" + +Wire `npm run build:tui` into `prepublishOnly` so `npm pack`/publish contains the compiled Ink app. + +### CLI integration strategy + +Update `cli/index.js` to: + +- open the TUI when `zeroshot` is run with no args in an interactive TTY +- add explicit commands: + - `zeroshot tui` (always open TUI) + - `zeroshot codex|claude|gemini|opencode` (open TUI with session provider override) +- keep `zeroshot watch` as an alias (expected to start in Monitor view once implemented). + +Important: + +- We do not keep the old blessed dashboard in parallel. As soon as the Ink entrypoint exists, + the old `src/tui/*` implementation is replaced/removed. + +## Workstreams + +This plan is split into chainable milestones, with parallelizable work inside each milestone. + +- Workstream A: CLI + packaging + TypeScript build +- Workstream B: Ink app shell + navigation + command model +- Workstream C: Data adapters (orchestrator, ledger logs, metrics, topology) +- Workstream D: UI views (launcher, monitor, cluster, agent) +- Workstream E: Guidance messaging backend (new capability) + UI wiring +- Workstream F: Tests + reliability hardening + +## Milestones and Issues + +Below, each "Issue" is intended to become a standalone PR with tight boundaries. + +### Milestone 1: Foundations (TS + Ink skeleton) + +#### Issue 1.1 - Add Ink + React dependencies (no runtime behavior change) + +Scope: + +- Add runtime deps needed for Ink TUI (e.g. `ink`, `react`). +- Add dev deps for TS typing/testing (e.g. `@types/react`, `ink-testing-library`), if needed. + +Non-scope: + +- No CLI behavior changes. +- No new commands. + +Acceptance: + +- `npm test` still passes. +- Existing CLI commands behave identically. + +#### Issue 1.2 - Replace `src/tui/` with "Hello Ink TUI" + build pipeline + +Scope: + +- Replace the existing blessed dashboard implementation under `src/tui/` with a minimal Ink app + (start with a single screen that renders and exits cleanly). +- Create `src/tui/index.tsx` (or equivalent entry) that renders a minimal Ink screen. +- Add `tsconfig.tui.json` emitting to `lib/tui/`. +- Add `npm run build:tui` and ensure it runs in `prepublishOnly`. +- Repoint the `zeroshot watch` implementation to the Ink entrypoint (old dashboard is removed immediately). +- Remove/replace any legacy dashboard tests that depend on blessed-contrib layout behavior. + +Acceptance: + +- `npm run build:tui` produces `lib/tui/index.js`. +- Running a temporary dev entry (e.g. `node -e "require('./lib/tui').start()"`) renders in terminal. +- `npm pack` includes `lib/tui/*`. +- `zeroshot watch` opens the Ink app (even if it is still a minimal stub). + +Parallelizable with: + +- Issue 1.1 + +#### Issue 1.3 - Add `zeroshot tui` command (explicit entrypoint) + provider session override + +Scope: + +- Add `zeroshot tui` command to `cli/index.js` that invokes the compiled Ink app. +- Ensure `--provider ` is accepted for `zeroshot tui` (session override only). +- Keep `zeroshot watch` as an alias (expected to start in Monitor view once implemented). + +Acceptance: + +- `zeroshot tui` opens the Ink app. +- `zeroshot tui --provider codex` passes provider override into the app (visible in UI state). + +Depends on: + +- Issue 1.2 + +### Milestone 2: Navigation + Command Model (no cluster execution yet) + +#### Issue 2.1 - Implement view router + "Esc back" navigation + +Scope: + +- Implement 4 view states (even if stubbed): + - Launcher + - Monitor + - Cluster + - Agent +- Global navigation rule: Esc pops view stack until Launcher. + +Acceptance: + +- Esc navigation is consistent from every view. +- Ctrl+C exits cleanly (no terminal corruption). + +Depends on: + +- Issue 1.3 + +#### Issue 2.2 - Implement command box + slash-command parser (MVP set) + +Scope: + +- Global input component. +- Parse rules: + - `/...` => command + - otherwise => plain-text task description (stubbed for now) +- Implement MVP commands: + - `/help`, `/monitor`, `/issue `, `/provider `, `/quit` +- Display command output (toast/status area). + +Acceptance: + +- Commands work from any view. +- Provider switches update a session state indicator. +- Commands that require orchestration (e.g. `/issue`) can be stubbed here and fully implemented in Milestone 3. + +Parallelizable with: + +- Issue 2.1 + +#### Issue 2.3 - Command dispatch scaffolding for "full CLI parity" over time + +Scope: + +- Introduce a typed command registry that maps `/...` commands to handlers. +- Start with a small compatibility layer so new commands can reuse existing CLI helper functions + (without requiring users to type `zeroshot ...`). +- Implement 1-2 additional commands as proof (e.g. `/status `, `/list`). + +Non-scope: + +- Do not implement every command in one PR. +- Avoid large refactors of `cli/index.js`; prefer extracting tiny shared helpers. + +Acceptance: + +- Adding a new slash command is a small, isolated change (new handler + tests). +- `/status ` works end-to-end and renders output in the TUI. + +### Milestone 3: Launch Cluster From Text (end-to-end MVP loop) + +#### Issue 3.1 - Extract minimal reusable "start cluster" helper for TUI (avoid copying CLI) + +Scope: + +- Create a small adapter module (prefer `lib/` or `src/` JS) that encapsulates: + - explicit input construction: + - plain text (default launcher behavior) + - issue refs for `/issue ...` (use existing parsing logic, but only for the `/issue` path) + - provider override resolution + - loading config (`resolveConfigPath`, `loadClusterConfig`) + - calling `orchestrator.start(...)` +- TUI calls this helper rather than duplicating CLI logic. + +Constraints: + +- Keep refactor minimal; do not restructure the CLI wholesale. + +Acceptance: + +- Existing CLI `zeroshot run` remains unchanged in behavior. +- New helper can be unit-tested independently. + +#### Issue 3.2 - Launcher view: Enter launches cluster and transitions to Cluster view + +Scope: + +- In Launcher view, non-`/` input starts a cluster from plain text with current provider override. +- Ambiguity rule: numeric input like `123` is treated as plain text (never an issue). +- Show cluster id immediately (optimistic UI). +- Transition to Cluster view after start begins. + +Acceptance: + +- Typing `Implement X` and pressing Enter starts a cluster and switches to Cluster view. +- Failures are shown as a clear error message and user remains in Launcher view. + +Depends on: + +- Issue 3.1 + +#### Issue 3.3 - `/issue ` command: run an issue and transition to Cluster view + +Scope: + +- Implement `/issue `: + - `ref` can be: `123`, `org/repo#123`, full issue URL, Jira key, etc (same accepted formats as CLI `zeroshot run `). + - Start a cluster using that issue ref and current provider override. + - Transition to Cluster view for that cluster id. + +Acceptance: + +- `/issue 123` starts a cluster from the issue (no ambiguity with plain text `123`). +- Errors are clearly rendered in the TUI (e.g. missing `gh`, auth failures, invalid ref). + +Depends on: + +- Issue 3.1 + +#### Issue 3.4 - Live log streaming in Cluster view (baseline) + +Scope: + +- Subscribe to the cluster ledger via `messageBus.subscribe(...)` or `ledger.since(...)` polling. +- Render a scrolling log viewport (Ink list/text). +- Provide basic filtering by agent id (optional toggle; can be later). + +Acceptance: + +- Cluster view shows new log lines within 0.5s of being written. +- No unbounded memory growth (keep only last N lines in view state). + +Parallelizable with: + +- Issue 3.2 (once cluster id is known) + +### Milestone 4: Monitor View (replacement for old dashboard) + +#### Issue 4.1 - Monitor view: list clusters from orchestrator registry + +Scope: + +- Implement `/monitor` view: + - fetch cluster list from orchestrator + - display a selectable list (arrow keys + enter) +- Enter opens Cluster view for selected id. + +Acceptance: + +- Monitor list matches `zeroshot list` cluster table order/contents (within reason). +- Can open any existing cluster (including completed) into Cluster view. + +Depends on: + +- Milestone 3 (Cluster view exists) + +#### Issue 4.2 - Add resource metrics to Monitor view (best effort) + +Scope: + +- Gather per-agent pid/CPU/memory using existing `pidusage` patterns. +- Aggregate per cluster and show in the list. +- Degrade gracefully on unsupported platforms. + +Acceptance: + +- On macOS/Linux, CPU/mem fields are populated for running clusters when possible. +- UI remains responsive with 10+ clusters; metrics refresh is throttled (e.g. every 2s). + +Parallelizable with: + +- Issue 4.1 + +#### Issue 4.3 - `zeroshot watch` starts directly in Monitor view + +Scope: + +- Ensure `zeroshot watch` starts the Ink TUI directly in Monitor view (not Launcher). +- Keep `zeroshot tui` defaulting to Launcher view. + +Acceptance: + +- `zeroshot watch` opens Monitor view as the initial screen. +- No user-facing regression for "monitor clusters" workflow. + +Depends on: + +- Issue 4.1 + +### Milestone 5: Cluster Focused View Enhancements (topology, steps) + +#### Issue 5.1 - Topology rendering from cluster config + +Scope: + +- Build a topology model from the running cluster config: + - agents (id, role) + - triggers/edges (topic wiring) +- Render as: + - MVP: adjacency list / ASCII tree / simple box diagram + - keep layout engine out-of-scope for now + +Acceptance: + +- For built-in templates, topology view shows all agents and their relationships. +- Works for dynamically added agents (best effort; show as appended nodes). + +#### Issue 5.2 - Step timeline derived from workflow triggers (MVP) + +Scope: + +- Define a minimal "timeline event" schema for the TUI. +- Populate it from `WORKFLOW_TRIGGERS` messages (PLAN_READY, IMPLEMENTATION_READY, VALIDATION_RESULT, etc). +- Render a compact history list in Cluster view. + +Acceptance: + +- Timeline shows the major phases for a typical run. +- Timeline persists on resume (derived from ledger, not in-memory only). + +Parallelizable with: + +- Issue 5.1 + +Pin: + +- Later we can enrich the timeline with additional lifecycle events/state transitions for better fidelity. + +### Milestone 6: Agent View (drill-down + messaging UI) + +#### Issue 6.1 - Agent selection + Agent view log tail + +Scope: + +- In Cluster view, allow selecting an agent (arrow keys) and opening Agent view (Enter). +- Agent view tails logs for that agent only. + +Acceptance: + +- Agent view shows live agent logs and updates in near real time. +- Esc returns to Cluster view preserving selection. + +Depends on: + +- Issue 3.4 + +#### Issue 6.2 - Agent messaging UI (stubbed backend) + +Scope: + +- Add an input box in Agent view for sending messages. +- For now, messages can be recorded as "pending" without delivery (backend not yet wired). + +Acceptance: + +- User can type and submit a message; UI shows it as queued. +- No crashes if backend is missing. + +Parallelizable with: + +- Milestone 7 backend work + +### Milestone 7: Guidance Messaging (new backend capability) + +This milestone is required to meet the PRD's "steer agents/cluster live" vision. It is intentionally separated because it touches core orchestration behavior. + +#### Issue 7.1 - Ledger topic + mailbox schema for guidance + +Scope: + +- Define new message topics (example): + - `USER_GUIDANCE_CLUSTER` + - `USER_GUIDANCE_AGENT` +- Each message includes: + - `cluster_id`, `sender: "user"`, `topic`, `content.text` + - optional `target_agent_id` + - delivery state fields (optional, can be derived) +- Implement a mailbox query helper for "guidance since last delivered". + +Acceptance: + +- Guidance messages are persisted in the ledger. +- Queries for "undelivered guidance" are deterministic and testable. + +#### Issue 7.2 - Live injection plumbing (provider stdin/PTY) with graceful detection + +Scope: + +- Implement a provider-agnostic "send input" path that writes directly to the underlying provider CLI session stdin/PTY when available. + - This should reuse the same mechanism the provider uses for interactive input (e.g. node-pty stdin write). +- Persist the guidance message in the ledger regardless (for audit/history), but mark whether it was injected live. +- If a given agent/provider session does not expose an interactive stdin handle, return "unsupported" so callers can fallback to queue semantics (Issue 7.3). + +Acceptance: + +- For at least one provider with interactive sessions, sending guidance while the agent is working injects into the live session. +- If injection is not possible, the API signals that it was not injected (so UI can show "queued"). + +Depends on: + +- Issue 7.1 + +#### Issue 7.3 - Queue fallback: apply guidance at safe points + +Scope: + +- Implement a safe-point mailbox consumer in the agent execution loop: + - fetch queued guidance for (cluster, agent) + - append to the next provider prompt in a controlled way (clearly delimited) +- Ensure guidance does not break JSON-schema modes or structured output. + +Acceptance: + +- In a test cluster for a non-injectable mode/provider, guidance sent mid-run appears in the next agent prompt. +- No regressions for existing runs (no guidance => behavior unchanged). + +Depends on: + +- Issue 7.2 + +#### Issue 7.4 - Cluster-wide guidance broadcast (injection-first, fallback-queue) + +Scope: + +- Implement cluster-level guidance delivery: + - write a single cluster-scoped message + - each agent attempts live injection when possible (Issue 7.2) + - otherwise the message is applied at that agent's next safe point (Issue 7.3) + +Acceptance: + +- Cluster-level guidance reaches all agents: + - injected live when supported + - queued otherwise + +Depends on: + +- Issue 7.3 + +#### Issue 7.5 - Wire TUI messaging UI to backend guidance mechanism + +Scope: + +- Cluster view: guidance box sends `USER_GUIDANCE_CLUSTER`. +- Agent view: guidance box sends `USER_GUIDANCE_AGENT`. +- Display delivery feedback (injected/queued/applied if detectable). + +Acceptance: + +- Messages typed in the TUI are persisted and delivered to agents (per Issues 7.1-7.4). + +Depends on: + +- Issues 6.2, 7.4 + +### Milestone 8: Default Entry + Cleanup + Hardening + +#### Issue 8.1 - `zeroshot` (no args) launches TUI (TTY only) + +Scope: + +- Modify `cli/index.js` default behavior: + - if interactive TTY and no subcommand: open TUI launcher + - else: keep existing help output behavior + +Acceptance: + +- `zeroshot` opens TUI in a normal terminal. +- `echo foo | zeroshot` does not hang (prints help and exits). + +Depends on: + +- Milestone 3 (minimum viable launcher exists) + +#### Issue 8.2 - Add `zeroshot codex|claude|gemini|opencode` convenience entrypoints + +Scope: + +- Add CLI commands that invoke `zeroshot tui --provider `. + +Acceptance: + +- `zeroshot codex` opens TUI with codex selected for the session. + +Depends on: + +- Issue 8.1 (or earlier if `zeroshot tui --provider` is already done) + +#### Issue 8.3 - Cleanup: remove legacy dashboard docs/tests and stale references + +Scope: + +- Remove any remaining blessed-dashboard-specific docs, demos, and tests (after the Ink TUI replacement). +- Ensure CLI help text and docs point to the Ink TUI entrypoints (`zeroshot`, `zeroshot tui`, `zeroshot watch`). + +Acceptance: + +- No references remain to the old blessed dashboard behavior. + +Depends on: + +- Issue 1.2 + +#### Issue 8.4 - Reliability/Perf pass + tests + +Scope: + +- Add tests: + - slash command parsing + - router "Esc back" behavior + - monitor list rendering + - guidance mailbox behavior (if Milestone 7 done) +- Manual checklist: + - resize behavior + - exit behavior (terminal reset) + - running multiple clusters + - resume existing cluster into Cluster view + +Acceptance: + +- CI is green. +- No known terminal corruption issues on exit. + +## Parallelization Map (Suggested) + +- Team 1: + - Milestone 1 (deps/build) + Milestone 2 (router/commands) +- Team 2: + - Milestone 3 (cluster start + logs) once helper exists +- Team 3: + - Milestone 4 (monitor view + metrics) +- Team 4: + - Milestone 7 (guidance backend) can start as soon as requirements are agreed +- Team 5: + - Milestone 5 (topology + timeline) can start once cluster config/state access is finalized + +## Definition of Done (Project-level) + +The project is considered "v2 shipped" when: + +- `zeroshot` launches the Ink TUI (TTY only) and can start a cluster from text. +- `/monitor` provides a stable cluster dashboard and drill-down. +- Cluster view supports logs + topology + timeline. +- The blessed-based dashboard is removed; `zeroshot watch` is an alias that opens the Ink TUI Monitor view. + +## Deferred / Post-v2 Candidates + +- AI summary panel (`/summary` or periodic): provider/model choice and cost controls TBD. +- Enrich the step timeline with additional lifecycle events beyond `WORKFLOW_TRIGGERS`. diff --git a/docs/tui-v2/PRD.md b/docs/tui-v2/PRD.md new file mode 100644 index 00000000..90940a57 --- /dev/null +++ b/docs/tui-v2/PRD.md @@ -0,0 +1,302 @@ +# Zeroshot TUI v2 (Ink) - PRD + +Date: 2026-01-25 +Owner: Zeroshot CLI +Status: Draft + +## Summary + +Build a new terminal UI (TUI) for Zeroshot using Ink + TypeScript. This fully replaces the current `zeroshot watch` dashboard (blessed-based) with a single interactive experience launched by running `zeroshot` (no args). + +Core workflow: + +- Open TUI +- Type a task description into a central input box +- Press Enter to launch a cluster +- Immediately switch to the focused cluster view (topology + logs + progress) +- Use `/monitor` to see a high-level dashboard of all clusters, then drill down via Enter +- Navigate back with Esc (Esc always steps back until the launcher view) + +Provider selection is session-scoped and can be chosen at launch (`zeroshot codex|claude|gemini|opencode`) or switched in-TUI. + +## Goals + +1. Replace the current dashboard feature with an Ink-based TUI. +2. Make `zeroshot` (no args) a first-class interactive mode for: + - launching clusters from free-form text + - monitoring all running clusters + - drilling into clusters and individual agents +3. Add a command palette/input model: + - plain text launches a cluster + - `/`-prefixed commands run zeroshot operations without re-typing `zeroshot` +4. Establish the first production TypeScript surface in the Zeroshot codebase (TUI + required adapters), without refactoring unrelated JS. + +## Non-Goals (for v2 MVP) + +- No pixel-perfect or final visual design (layout/theme/typography can evolve). +- No full JavaScript -> TypeScript migration. +- No remote multi-user UI; TUI is local-only. +- No promise of message delivery for providers/modes that do not support interactive stdin injection. + - In those cases, guidance is queued and applied at the next safe point. +- No replacement of non-dashboard CLI commands; existing subcommands remain supported. + +## Users / Personas + +- Power users running many clusters and needing quick navigation (htop/k9s style). +- Developers launching a cluster from an idea ("just type it and go"). +- Operators who need to see whether a run is stuck, progressing, or consuming resources. + +## Glossary + +- Cluster: a multi-agent run managed by `src/orchestrator.js`, persisted in `~/.zeroshot/*.db`. +- Agent / Worker: an `AgentWrapper` instance inside a cluster (worker, planner, validators, etc). +- Provider: external CLI used for agent reasoning (claude, codex, gemini, opencode). + +## Entry Points + +### `zeroshot` (no args) + +Expected behavior: + +- If running in an interactive TTY: start TUI in Launcher view. +- If not a TTY (piped/CI): print help and exit non-zero only on misuse (avoid breaking scripts). + +### `zeroshot tui` (explicit) + +Always opens the TUI (even if additional flags are provided). + +### `zeroshot codex|claude|gemini|opencode` + +Opens the TUI with a session-scoped provider override. While the TUI is open: + +- any cluster launch uses the chosen provider (equivalent to passing `--provider ` behind the scenes) +- switching provider in-TUI updates this override for subsequent operations + +Notes: + +- This must not permanently change the default provider setting. Persisted defaults remain managed by `zeroshot providers set-default ...`. + +### `zeroshot watch` + +`zeroshot watch` remains as a convenience alias that opens the new Ink TUI directly in Monitor view. + +The existing blessed-based TUI implementation is removed (no parallel legacy dashboard). + +## Core Navigation / Views + +### 1) Launcher View (Main) + +Primary UI shown on start. + +Content: + +- A central input box for text. +- Short instructions/hints (e.g. `/monitor`, `/help`, provider indicator). + +Input semantics: + +- If input starts with `/`: treat as a command (see Commands). +- Else: treat as plain-text task description and start a cluster. + - Ambiguity rule: numeric input like `123` is treated as plain text. + - To run an issue, use `/issue ...` (e.g. `/issue 123`). + +On Enter with non-command input: + +- Start cluster +- Transition to Cluster Focused View for that cluster + +### 2) Monitor View (Dashboard) + +Opened by `/monitor` from anywhere. + +Displays a high-level list of clusters, including: + +- cluster id +- input/task name (derived from input: issue title, file name, or first line of text) +- provider used +- status (running/completed/failed/stopped/corrupted) +- running time (duration) +- agents/workers count +- resource usage (CPU%, memory) aggregated across agents (best effort; degrade gracefully) +- token usage (if available via ledger aggregation) +- last activity timestamp + +Interaction: + +- Up/Down (or j/k) moves selection +- Enter opens Cluster Focused View for selected cluster +- Esc goes back to previous view + +Optional (later, but supported by design): + +- filtering (running/stopped/all) +- search by cluster id / task substring +- actions: stop/kill/export via `/stop `, `/kill `, `/export ` + +### 3) Cluster Focused View + +Opened automatically after launching a cluster, or by selecting a cluster from Monitor view. + +Must display: + +1. Cluster topology graph: + - Rendered from the cluster config template (agents + triggers). + - MVP: tree/adjacency list/ASCII graph; fidelity can improve over time. +2. Live logs: + - Streamed from the ledger message bus (append-only). + - Includes per-agent attribution (role/agent id). +3. Step timeline / phase log: + - High-level step list derived from workflow triggers (`WORKFLOW_TRIGGERS`) for MVP. + - Examples: plan -> implement -> validate -> iterate -> complete. + - Exact naming/format is defined by the TUI domain model and can evolve. + +Interaction: + +- Left/Right (or Tab) cycles focus between panes (topology/logs/steps/agents list). +- Up/Down moves selection when in a selectable pane (e.g. agents list). +- Enter on a selected agent opens Agent View. +- Esc goes back (Monitor if you came from Monitor, or Launcher if you came from Launcher). + +Cluster guidance input: + +- A command/text box is available in this view to send guidance to the whole cluster. +- This requires backend support (see "Guidance Messaging"). + +### 4) Agent View (Focused Worker/Agent) + +Opened by selecting an agent from Cluster Focused View. + +Must display: + +- Live logs for that agent (tailing relevant ledger output) +- Agent identity (role, provider, model level/model) +- Agent status (idle/executing/stuck/failed/completed) + +Agent messaging input: + +- A command/text box that lets the user send guidance to that agent while it is working. +- Provider capability may vary. If true "live injection" is not available, guidance is queued and applied at the next safe point. + +Esc returns to Cluster Focused View. + +## Commands (Slash Commands) + +General: + +- Any view can accept `/...` commands. +- Commands should be parsed similarly to CLI subcommands, but do not require the `zeroshot` prefix. +- Commands produce feedback in a lightweight output/toast area (success/failure). + +Parity target: + +- Long-term: every existing `zeroshot ` operation should be invocable as `/` inside the TUI. +- Short-term: ship an MVP subset first (below), then expand to full parity incrementally. + +Required commands for MVP: + +- `/help` - show available commands and keybindings +- `/monitor` - open Monitor view +- `/issue ` - start a cluster from an issue reference (e.g. `123`, `org/repo#123`, URL, Jira key, etc) +- `/provider ` - switch provider for the session (claude|codex|gemini|opencode) +- `/quit` (and `/exit`) - exit the TUI (with confirmation if clusters running) + +Strongly recommended commands (v2 follow-up): + +- `/run ` - start a cluster using CLI-style input parsing (issue/file/text auto-detection) +- `/list` - list clusters/tasks (or open Monitor view) +- `/status ` - show detailed cluster/task status +- `/logs ` - open Cluster Focused View (or open logs modal) +- `/stop ` and `/kill ` - control cluster/task + +## Guidance Messaging (Backend Requirement) + +Two scopes: + +1. Cluster guidance: message broadcast to all agents in the cluster. +2. Agent guidance: message targeted to a specific agent. + +Delivery semantics (MVP): + +- Attempt live stdin injection into the underlying provider CLI session when supported. + - Implementation detail: write to the provider process/PTY stdin (same mechanism the provider uses for interactive chat). +- If live injection is not supported (provider limitation or non-interactive mode), queue the guidance: + - store in the ledger (or a small mailbox store) + - agents apply it at the next safe point: + - before starting a new iteration + - before generating a new provider prompt + - after finishing a tool execution step + +TUI requirements: + +- Show whether a guidance message was injected live or queued. +- Indicate delivery status (pending/applied/expired) when possible. + +## Data Sources / Domain Model + +The TUI reads from: + +- Orchestrator registry: cluster list and state (`src/orchestrator.js`) +- Ledger message bus: logs, events, tokens (`src/message-bus.js`, `src/ledger.js`) +- Process metrics: pid/CPU/mem per agent (existing `pidusage` usage) +- Cluster config: topology information (from resolved config JSON) + +The TUI writes: + +- Start/stop/kill commands via orchestrator +- Guidance messages via message bus / mailbox + +## Non-Functional Requirements + +Performance: + +- TUI launch time (from `zeroshot` to first paint): < 500ms on a typical dev machine. +- UI update cadence: + - logs: event-driven, or up to 200-500ms batching + - cluster list refresh: 1s (configurable) + - resource metrics refresh: 2s (configurable, because pidusage can be expensive) + +Reliability: + +- No terminal corruption on exit (restore cursor, clear alternate buffer if used). +- Handles terminal resize gracefully. +- Works on macOS and Linux. If metrics are unsupported, show "-" and keep UI functional. + +Security/Privacy: + +- Do not exfiltrate code or logs; all data stays local. +- Avoid rendering secrets in the UI where possible (best-effort; user controls what runs). + +## Success Metrics + +Quantitative: + +- 90%+ of interactive "start cluster from text" actions succeed without leaving the TUI. +- 0 known cases of terminal left in broken state on normal exit. +- Monitor view can handle 100 clusters in the registry without freezing (best-effort; virtualization if needed later). + +Qualitative: + +- Users can launch, monitor, and drill down without remembering command syntax. +- Navigation feels consistent (Esc always goes back). + +## Out of Scope / Deferred + +- Full parity with all CLI flags (`--docker`, `--ship`, `--worktree`, etc) in the launcher input. + - These can be added incrementally via `/run --docker ...` style. +- Rich graphs (true layout engine). MVP uses simplified ASCII representation. +- Advanced UX (themes, mouse, split panes, persistent layouts). +- AI summary panel (provider/model choice + cost controls TBD). + +## Resolved Decisions (from product feedback) + +1. Ambiguous launcher input: + - Launcher treats non-`/` input as plain text (including `123`). + - Use `/issue 123` (or another issue ref) to run an issue. +2. Step timeline schema: + - MVP derives timeline from existing workflow triggers (`WORKFLOW_TRIGGERS`). + - We keep a pin to later enrich with additional lifecycle events. +3. AI summary: + - Deferred (pin). Implement after core TUI flows ship. +4. Guidance injection: + - Implement live injection when provider supports it (inject into the provider CLI stdin/PTY). + - Otherwise, queue and apply at safe points. diff --git a/docs/tui-v2/protocol.md b/docs/tui-v2/protocol.md new file mode 100644 index 00000000..e30113e6 --- /dev/null +++ b/docs/tui-v2/protocol.md @@ -0,0 +1,463 @@ +protocolVersion: 1 + +# TUI v2 JSON-RPC Protocol (v0) + +## Framing + +- Each message is framed as: `Content-Length: \r\n\r\n` followed by `` bytes of UTF-8 JSON. +- Ignore unknown headers. +- Reject frames larger than 10MB with RPC error `-32600` (invalid request). + +## Envelope (JSON-RPC 2.0) + +All messages are JSON-RPC 2.0 objects: + +- `jsonrpc`: must be `"2.0"`. +- `id`: string or number (required for requests/responses, omitted for notifications). +- `method`: string (required for requests/notifications). +- `params`: optional; object | array | null. +- `result` or `error` (responses only; exactly one). + +## Error Object + +```json +{ + "code": -32600, + "message": "Invalid request", + "data": { + "detail": "optional details", + "fields": { + "/params/clusterId": "must be string" + } + } +} +``` + +- `data.detail` is a human-readable summary (optional). +- `data.fields` maps JSON pointers to per-field messages (optional). + +## Error Codes + +- `-32700` parse error +- `-32600` invalid request +- `-32601` method not found +- `-32602` invalid params +- `-32603` internal error +- `-32000` protocol version mismatch +- `-32001` orchestrator unavailable +- `-32002` cluster not found +- `-32003` unsupported capability + +## Domain Types + +### ClusterSummary + +```json +{ + "id": "cluster-123", + "state": "running", + "provider": "codex", + "createdAt": 1769810000000, + "agentCount": 3, + "messageCount": 120, + "cwd": "/path/to/workdir" +} +``` + +### ClusterMetrics + +```json +{ + "id": "cluster-123", + "supported": true, + "cpuPercent": 12.3, + "memoryMB": 256.7 +} +``` + +### ClusterLogLine + +```json +{ + "id": "line-1", + "timestamp": 1769811111000, + "text": "Agent output", + "agent": "worker", + "role": "implementation", + "sender": "worker" +} +``` + +### TimelineEvent + +```json +{ + "id": "evt-1", + "timestamp": 1769811111000, + "topic": "PLAN_READY", + "label": "Plan ready", + "approved": null, + "sender": "planner" +} +``` + +### TopologyAgent / TopologyEdge / ClusterTopology + +```json +{ + "agents": [{ "id": "worker", "role": "implementation" }], + "edges": [ + { "from": "system", "to": "ISSUE_OPENED", "topic": "ISSUE_OPENED", "kind": "source" }, + { "from": "ISSUE_OPENED", "to": "worker", "topic": "ISSUE_OPENED", "kind": "trigger" } + ], + "topics": ["ISSUE_OPENED"] +} +``` + +### GuidanceDeliveryResult / ClusterGuidanceDelivery + +```json +{ + "summary": { "injected": 1, "queued": 0, "total": 1 }, + "agents": { + "worker": { "status": "injected", "reason": null, "method": "pty", "taskId": "task-1" } + }, + "timestamp": 1769811111000 +} +``` + +## Methods (Requests/Responses) + +### initialize + +Request: + +```json +{ + "jsonrpc": "2.0", + "id": 1, + "method": "initialize", + "params": { + "protocolVersion": 1, + "client": { "name": "zeroshot-tui", "version": "0.1.0", "pid": 12345 }, + "capabilities": { "wantsMetrics": true, "wantsTopology": false } + } +} +``` + +Response: + +```json +{ + "jsonrpc": "2.0", + "id": 1, + "result": { + "protocolVersion": 1, + "server": { "name": "zeroshot-backend", "version": "5.4.0" }, + "capabilities": { + "methods": ["initialize", "listClusters", "getClusterSummary", "unsubscribe"], + "notifications": ["clusterLogLines", "clusterTimelineEvents"] + } + } +} +``` + +### listClusters + +Request: `listClusters()` + +Response: + +```json +{ + "jsonrpc": "2.0", + "id": 2, + "result": { + "clusters": [ + /* ClusterSummary[] */ + ] + } +} +``` + +### getClusterSummary + +Request: + +```json +{ + "jsonrpc": "2.0", + "id": 3, + "method": "getClusterSummary", + "params": { "clusterId": "cluster-123" } +} +``` + +Response: + +```json +{ + "jsonrpc": "2.0", + "id": 3, + "result": { + "summary": { + /* ClusterSummary */ + } + } +} +``` + +### listClusterMetrics + +Request: + +```json +{ + "jsonrpc": "2.0", + "id": 4, + "method": "listClusterMetrics", + "params": { "clusterIds": ["cluster-123"] } +} +``` + +Response: + +```json +{ + "jsonrpc": "2.0", + "id": 4, + "result": { + "metrics": [ + /* ClusterMetrics[] */ + ] + } +} +``` + +### startClusterFromText + +Request: + +```json +{ + "jsonrpc": "2.0", + "id": 5, + "method": "startClusterFromText", + "params": { + "text": "Implement the requested feature", + "providerOverride": "codex", + "clusterId": "cluster-123" + } +} +``` + +Response: + +```json +{ "jsonrpc": "2.0", "id": 5, "result": { "clusterId": "cluster-123" } } +``` + +### startClusterFromIssue + +Request: + +```json +{ + "jsonrpc": "2.0", + "id": 6, + "method": "startClusterFromIssue", + "params": { "ref": "covibes/zeroshot#240", "providerOverride": null } +} +``` + +Response: + +```json +{ "jsonrpc": "2.0", "id": 6, "result": { "clusterId": "cluster-456" } } +``` + +### sendGuidanceToAgent + +Request: + +```json +{ + "jsonrpc": "2.0", + "id": 7, + "method": "sendGuidanceToAgent", + "params": { + "clusterId": "cluster-123", + "agentId": "worker", + "text": "Focus on tests", + "timeoutMs": 5000 + } +} +``` + +Response: + +```json +{ + "jsonrpc": "2.0", + "id": 7, + "result": { + "result": { + /* GuidanceDeliveryResult */ + } + } +} +``` + +### sendGuidanceToCluster + +Request: + +```json +{ + "jsonrpc": "2.0", + "id": 8, + "method": "sendGuidanceToCluster", + "params": { "clusterId": "cluster-123", "text": "Ship it", "timeoutMs": 5000 } +} +``` + +Response: + +```json +{ + "jsonrpc": "2.0", + "id": 8, + "result": { + "result": { + /* ClusterGuidanceDelivery */ + } + } +} +``` + +### subscribeClusterLogs + +Request: + +```json +{ + "jsonrpc": "2.0", + "id": 9, + "method": "subscribeClusterLogs", + "params": { "clusterId": "cluster-123", "agentId": "worker" } +} +``` + +Response: + +```json +{ "jsonrpc": "2.0", "id": 9, "result": { "subscriptionId": "sub-logs-1" } } +``` + +### subscribeClusterTimeline + +Request: + +```json +{ + "jsonrpc": "2.0", + "id": 10, + "method": "subscribeClusterTimeline", + "params": { "clusterId": "cluster-123" } +} +``` + +Response: + +```json +{ "jsonrpc": "2.0", "id": 10, "result": { "subscriptionId": "sub-timeline-1" } } +``` + +### unsubscribe + +Request: + +```json +{ + "jsonrpc": "2.0", + "id": 11, + "method": "unsubscribe", + "params": { "subscriptionId": "sub-logs-1" } +} +``` + +Response: + +```json +{ "jsonrpc": "2.0", "id": 11, "result": { "removed": true } } +``` + +### getClusterTopology + +Request: + +```json +{ + "jsonrpc": "2.0", + "id": 12, + "method": "getClusterTopology", + "params": { "clusterId": "cluster-123" } +} +``` + +Response: + +```json +{ + "jsonrpc": "2.0", + "id": 12, + "result": { + "topology": { + /* ClusterTopology */ + } + } +} +``` + +## Notifications + +### clusterLogLines + +```json +{ + "jsonrpc": "2.0", + "method": "clusterLogLines", + "params": { + "subscriptionId": "sub-logs-1", + "clusterId": "cluster-123", + "lines": [ + /* ClusterLogLine[] */ + ], + "droppedCount": 0 + } +} +``` + +### clusterTimelineEvents + +```json +{ + "jsonrpc": "2.0", + "method": "clusterTimelineEvents", + "params": { + "subscriptionId": "sub-timeline-1", + "clusterId": "cluster-123", + "events": [ + /* TimelineEvent[] */ + ], + "droppedCount": 0 + } +} +``` + +## Versioning Rules + +- `protocolVersion` is required in `initialize` and must match this spec. +- If a client sends an unsupported `protocolVersion`, return error `-32000`. +- Additive changes (new methods/fields) must be backward compatible within the same major protocol version. +- Breaking changes require incrementing `protocolVersion` and updating this spec. diff --git a/eslint.config.mjs b/eslint.config.mjs index 761d66a5..c9147b56 100644 --- a/eslint.config.mjs +++ b/eslint.config.mjs @@ -216,7 +216,6 @@ export default [ { // TUI/CLI/streaming files use ANSI escape codes for terminal colors - allow control characters files: [ - 'src/tui/**/*.js', 'src/streaming/*.js', 'src/streaming/**/*.js', 'src/status-footer.js', @@ -231,6 +230,16 @@ export default [ 'sonarjs/no-unused-vars': 'off', }, }, + { + // lib/tui-backend output is generated by tsc; allow compiler helpers and comparisons + files: ['lib/tui-backend/**/*.js'], + rules: { + 'no-global-this': 'off', + 'sonarjs/no-global-this': 'off', + 'no-shadow': 'off', + eqeqeq: 'off', + }, + }, { // Large files that need refactoring - temporary overrides // TODO: Split these files into smaller modules @@ -254,7 +263,14 @@ export default [ }, }, { - ignores: ['node_modules/**', 'dist/**', 'coverage/**', 'cluster-hooks/**', 'hooks/**'], + ignores: [ + 'node_modules/**', + 'dist/**', + 'coverage/**', + 'cluster-hooks/**', + 'hooks/**', + 'lib/tui-backend/**', + ], }, prettierConfig, ]; diff --git a/lib/id-detector.js b/lib/id-detector.js index f708c143..d00dcd13 100644 --- a/lib/id-detector.js +++ b/lib/id-detector.js @@ -12,19 +12,18 @@ const fs = require('fs'); const os = require('os'); const Database = require('better-sqlite3'); -// Storage paths -const CLUSTER_DIR = path.join(os.homedir(), '.zeroshot'); -const TASK_DIR = path.join(os.homedir(), '.claude-zeroshot'); -const TASK_DB_FILE = path.join(TASK_DIR, 'store.db'); - /** * Detect if ID is a cluster or task * @param {string} id - The ID to check * @returns {'cluster'|'task'|null} - Type of ID or null if not found */ function detectIdType(id) { + const homeDir = + process.env.ZEROSHOT_HOME || process.env.HOME || process.env.USERPROFILE || os.homedir(); + const clusterFile = path.join(homeDir, '.zeroshot', 'clusters.json'); + const taskDbFile = path.join(homeDir, '.claude-zeroshot', 'store.db'); + // Check clusters - const clusterFile = path.join(CLUSTER_DIR, 'clusters.json'); if (fs.existsSync(clusterFile)) { try { const clusters = JSON.parse(fs.readFileSync(clusterFile, 'utf8')); @@ -37,9 +36,9 @@ function detectIdType(id) { } // Check tasks in SQLite - if (fs.existsSync(TASK_DB_FILE)) { + if (fs.existsSync(taskDbFile)) { try { - const db = new Database(TASK_DB_FILE, { readonly: true, timeout: 5000 }); + const db = new Database(taskDbFile, { readonly: true, timeout: 5000 }); const row = db.prepare('SELECT id FROM tasks WHERE id = ?').get(id); db.close(); if (row) { diff --git a/lib/repo-settings.js b/lib/repo-settings.js new file mode 100644 index 00000000..05625079 --- /dev/null +++ b/lib/repo-settings.js @@ -0,0 +1,69 @@ +/** + * Repo-local settings for zeroshot + * + * Optional per-repository config file: + * /.zeroshot/settings.json + * + * This complements the global user settings at: + * ~/.zeroshot/settings.json + */ + +const fs = require('fs'); +const path = require('path'); +const { execSync } = require('child_process'); + +function _safeJsonParse(text) { + try { + return JSON.parse(text); + } catch { + return null; + } +} + +function _getGitRoot(dir) { + try { + return execSync('git rev-parse --show-toplevel', { + cwd: dir, + encoding: 'utf8', + stdio: 'pipe', + }).trim(); + } catch { + return null; + } +} + +function _readSettingsFile(filePath) { + try { + const raw = fs.readFileSync(filePath, 'utf8'); + const parsed = _safeJsonParse(raw); + if (!parsed || typeof parsed !== 'object') { + return null; + } + return parsed; + } catch { + return null; + } +} + +/** + * Read repo-local settings if present. + * + * @param {string} startDir - Directory inside the repo (usually process.cwd()). + * @returns {{repoRoot: string|null, settings: object|null, settingsPath: string|null}} + */ +function readRepoSettings(startDir) { + const repoRoot = _getGitRoot(startDir); + if (!repoRoot) { + return { repoRoot: null, settings: null, settingsPath: null }; + } + + const settingsPath = path.join(repoRoot, '.zeroshot', 'settings.json'); + if (!fs.existsSync(settingsPath)) { + return { repoRoot, settings: null, settingsPath }; + } + + const settings = _readSettingsFile(settingsPath); + return { repoRoot, settings, settingsPath }; +} + +module.exports = { readRepoSettings }; diff --git a/lib/settings.js b/lib/settings.js index b579baa1..a7428da3 100644 --- a/lib/settings.js +++ b/lib/settings.js @@ -121,6 +121,14 @@ const DEFAULT_SETTINGS_BASE = { // Container home directory - where $HOME resolves in container paths // Default: /home/node (matches zeroshot-cluster-base image) dockerContainerHome: '/home/node', + // Retry/restart robustness defaults + maxRetries: 3, // Agent task retries (per execution) for retryable errors + maxRestartAttempts: 3, // Agent restarts since last TASK_COMPLETED + maxTotalRestarts: 10, // Safety valve (never resets) + staleWarningsBeforeKill: 2, // Consecutive stale warnings before restart + backoffBaseMs: 2000, // Initial retry backoff + backoffMaxMs: 30000, // Max retry backoff + jitterFactor: 0.2, // Random jitter Ā±20% // Issue provider settings - defaultIssueSource is here, others come from providers defaultIssueSource: 'github', // 'github' | 'gitlab' | 'jira' | 'azure-devops' }; @@ -340,6 +348,37 @@ function validateSetting(key, value) { return `Unknown setting: ${key}`; } + if ( + [ + 'maxRetries', + 'maxRestartAttempts', + 'maxTotalRestarts', + 'staleWarningsBeforeKill', + 'backoffBaseMs', + 'backoffMaxMs', + ].includes(key) + ) { + if (!Number.isFinite(value)) { + return `${key} must be a number`; + } + if (!Number.isInteger(value)) { + return `${key} must be an integer`; + } + const min = ['backoffBaseMs', 'backoffMaxMs'].includes(key) ? 0 : 1; + if (value < min) { + return `${key} must be >= ${min}`; + } + } + + if (key === 'jitterFactor') { + if (!Number.isFinite(value)) { + return 'jitterFactor must be a number'; + } + if (value < 0 || value > 1) { + return 'jitterFactor must be between 0 and 1'; + } + } + if (key === 'maxModel' && !VALID_MODELS.includes(value)) { return `Invalid model: ${value}. Valid models: ${VALID_MODELS.join(', ')}`; } @@ -423,7 +462,7 @@ function coerceValue(key, value) { } if (typeof defaultValue === 'number') { - const parsed = parseInt(value); + const parsed = parseFloat(value); if (isNaN(parsed)) { throw new Error(`Invalid number: ${value}`); } diff --git a/lib/start-cluster.js b/lib/start-cluster.js new file mode 100644 index 00000000..620bf336 --- /dev/null +++ b/lib/start-cluster.js @@ -0,0 +1,348 @@ +const path = require('path'); +const { execSync } = require('child_process'); +const chalk = require('chalk'); +const { normalizeProviderName } = require('./provider-names'); +const { getProvider } = require('../src/providers'); + +const PACKAGE_ROOT = path.resolve(__dirname, '..'); + +/** + * Detect git repository root from current directory. + * @returns {string} Git repo root, or process.cwd() if not in a git repo. + */ +function detectGitRepoRoot() { + try { + const root = execSync('git rev-parse --show-toplevel', { + encoding: 'utf8', + stdio: ['pipe', 'pipe', 'pipe'], + }).trim(); + return root; + } catch { + return process.cwd(); + } +} + +function resolveOptionalString(value) { + if (typeof value !== 'string') return undefined; + const trimmed = value.trim(); + return trimmed.length > 0 ? trimmed : undefined; +} + +function resolveEnvBool(value) { + if (typeof value !== 'string') return undefined; + const trimmed = value.trim().toLowerCase(); + if (trimmed === '1' || trimmed === 'true' || trimmed === 'yes') return true; + if (trimmed === '0' || trimmed === 'false' || trimmed === 'no') return false; + return undefined; +} + +function resolveCloseIssueMode(value) { + const trimmed = resolveOptionalString(value); + if (!trimmed) return undefined; + const normalized = trimmed.toLowerCase(); + if (normalized === 'auto' || normalized === 'always' || normalized === 'never') { + return normalized; + } + return undefined; +} + +function parseRunOptionsEnv() { + const raw = resolveOptionalString(process.env.ZEROSHOT_RUN_OPTIONS); + if (!raw) return null; + try { + const parsed = JSON.parse(raw); + if (!parsed || typeof parsed !== 'object') { + return null; + } + return parsed; + } catch { + return null; + } +} + +/** + * Parse CLI mount specs (host:container[:ro]) into mount config objects. + * @param {string[]} specs - Array of mount specs from CLI. + * @returns {Array<{host: string, container: string, readonly: boolean}>} + */ +function parseMountSpecs(specs) { + return specs.map((spec) => { + const parts = spec.split(':'); + if (parts.length < 2) { + throw new Error(`Invalid mount spec: "${spec}". Format: host:container[:ro]`); + } + const host = parts[0]; + const container = parts[1]; + const readonly = parts[2] === 'ro'; + return { host, container, readonly }; + }); +} + +function buildTextInput(text) { + return { text }; +} + +function buildIssueInput(issue) { + return { issue }; +} + +function buildFileInput(file) { + return { file }; +} + +function detectRunInput(inputArg) { + const isGitHubUrl = /^https?:\/\/github\.com\/[\w-]+\/[\w-]+\/issues\/\d+/.test(inputArg); + const isGitLabUrl = /gitlab\.(com|[\w.-]+)\/[\w-]+\/[\w-]+\/-\/issues\/\d+/.test(inputArg); + const isJiraUrl = /(atlassian\.net|jira\.[\w.-]+)\/browse\/[A-Z][A-Z0-9]+-\d+/.test(inputArg); + const isAzureUrl = + /dev\.azure\.com\/.*\/_workitems\/edit\/\d+/.test(inputArg) || + /visualstudio\.com\/.*\/_workitems\/edit\/\d+/.test(inputArg); + const isJiraKey = /^[A-Z][A-Z0-9]+-\d+$/.test(inputArg); + const isIssueNumber = /^\d+$/.test(inputArg); + const isRepoIssue = /^[\w-]+\/[\w-]+#\d+$/.test(inputArg); + const isMarkdownFile = /\.(md|markdown)$/i.test(inputArg); + + if ( + isGitHubUrl || + isGitLabUrl || + isJiraUrl || + isAzureUrl || + isJiraKey || + isIssueNumber || + isRepoIssue + ) { + return buildIssueInput(inputArg); + } + if (isMarkdownFile) { + return buildFileInput(inputArg); + } + return buildTextInput(inputArg); +} + +function resolveProviderOverride(options = {}) { + const override = options.provider || options.envProvider || process.env.ZEROSHOT_PROVIDER; + if (!override || (typeof override === 'string' && !override.trim())) { + return null; + } + const normalized = normalizeProviderName(override); + if (options.validateProvider) { + getProvider(normalized); + } + return normalized; +} + +function resolveConfigPath(configName) { + if (path.isAbsolute(configName) || configName.startsWith('./') || configName.startsWith('../')) { + return path.resolve(process.cwd(), configName); + } + if (configName.endsWith('.json')) { + return path.join(PACKAGE_ROOT, 'cluster-templates', configName); + } + return path.join(PACKAGE_ROOT, 'cluster-templates', `${configName}.json`); +} + +function ensureConfigProviderDefaults(config, settings) { + if (!config.defaultProvider) { + config.defaultProvider = settings.defaultProvider || 'claude'; + } + config.defaultProvider = normalizeProviderName(config.defaultProvider) || 'claude'; +} + +function applyProviderOverrideToConfig(config, providerOverride, settings) { + const provider = getProvider(providerOverride); + const providerSettings = settings.providerSettings?.[providerOverride] || {}; + config.forceProvider = providerOverride; + config.defaultProvider = providerOverride; + config.forceLevel = providerSettings.defaultLevel || provider.getDefaultLevel(); + config.defaultLevel = config.forceLevel; + console.log(chalk.dim(`Provider override: ${providerOverride} (all agents)`)); +} + +function loadClusterConfig(orchestrator, configPath, settings = {}, providerOverride) { + const config = orchestrator.loadConfig(configPath); + ensureConfigProviderDefaults(config, settings); + if (providerOverride) { + applyProviderOverrideToConfig(config, providerOverride, settings); + } + return config; +} + +function buildStartOptions({ + clusterId, + options = {}, + settings = {}, + providerOverride, + modelOverride, + forceProvider, +}) { + const envRunOptions = parseRunOptionsEnv(); + const mergedOptions = envRunOptions ? { ...envRunOptions, ...options } : options; + const targetCwd = process.env.ZEROSHOT_CWD || detectGitRepoRoot(); + const envPrBase = resolveOptionalString(process.env.ZEROSHOT_PR_BASE); + const envMergeQueue = resolveEnvBool(process.env.ZEROSHOT_MERGE_QUEUE); + const envCloseIssue = resolveCloseIssueMode(process.env.ZEROSHOT_CLOSE_ISSUE); + const prBase = resolveOptionalString(mergedOptions.prBase) || envPrBase; + const mergeQueue = + typeof mergedOptions.mergeQueue === 'boolean' ? mergedOptions.mergeQueue : envMergeQueue; + const closeIssue = resolveCloseIssueMode(mergedOptions.closeIssue) || envCloseIssue; + return { + clusterId, + cwd: targetCwd, + isolation: + mergedOptions.docker || process.env.ZEROSHOT_DOCKER === '1' || settings.defaultDocker, + isolationImage: mergedOptions.dockerImage || process.env.ZEROSHOT_DOCKER_IMAGE || undefined, + worktree: mergedOptions.worktree || process.env.ZEROSHOT_WORKTREE === '1', + autoPr: mergedOptions.pr || process.env.ZEROSHOT_PR === '1', + autoMerge: process.env.ZEROSHOT_MERGE === '1', + autoPush: process.env.ZEROSHOT_PUSH === '1', + modelOverride: modelOverride || undefined, + providerOverride: providerOverride || undefined, + noMounts: mergedOptions.mounts === false || mergedOptions.noMounts === true, + mounts: mergedOptions.mount ? parseMountSpecs(mergedOptions.mount) : undefined, + containerHome: mergedOptions.containerHome || undefined, + forceProvider: forceProvider || undefined, + // PR configuration options (CLI args) + prBase: prBase || undefined, + mergeQueue: mergeQueue, + closeIssue: closeIssue || undefined, + settings, + }; +} + +function resolveConfigOrThrow({ + orchestrator, + config, + configPath, + configName, + settings, + providerOverride, +}) { + if (config) { + return config; + } + const resolvedPath = configPath || (configName ? resolveConfigPath(configName) : null); + if (!resolvedPath) { + throw new Error('configPath or configName is required when config is not provided'); + } + return loadClusterConfig(orchestrator, resolvedPath, settings, providerOverride); +} + +function startClusterFromText({ + orchestrator, + text, + config, + configPath, + configName, + settings, + providerOverride, + modelOverride, + forceProvider, + clusterId, + options = {}, +}) { + if (!orchestrator) { + throw new Error('orchestrator is required'); + } + const resolvedConfig = resolveConfigOrThrow({ + orchestrator, + config, + configPath, + configName, + settings, + providerOverride, + }); + const startOptions = buildStartOptions({ + clusterId, + options, + settings, + providerOverride, + modelOverride, + forceProvider, + }); + return orchestrator.start(resolvedConfig, buildTextInput(text), startOptions); +} + +function startClusterFromIssue({ + orchestrator, + issue, + config, + configPath, + configName, + settings, + providerOverride, + modelOverride, + forceProvider, + clusterId, + options = {}, +}) { + if (!orchestrator) { + throw new Error('orchestrator is required'); + } + const resolvedConfig = resolveConfigOrThrow({ + orchestrator, + config, + configPath, + configName, + settings, + providerOverride, + }); + const startOptions = buildStartOptions({ + clusterId, + options, + settings, + providerOverride, + modelOverride, + forceProvider, + }); + return orchestrator.start(resolvedConfig, buildIssueInput(issue), startOptions); +} + +function startClusterFromFile({ + orchestrator, + file, + config, + configPath, + configName, + settings, + providerOverride, + modelOverride, + forceProvider, + clusterId, + options = {}, +}) { + if (!orchestrator) { + throw new Error('orchestrator is required'); + } + const resolvedConfig = resolveConfigOrThrow({ + orchestrator, + config, + configPath, + configName, + settings, + providerOverride, + }); + const startOptions = buildStartOptions({ + clusterId, + options, + settings, + providerOverride, + modelOverride, + forceProvider, + }); + return orchestrator.start(resolvedConfig, buildFileInput(file), startOptions); +} + +module.exports = { + buildTextInput, + buildIssueInput, + buildFileInput, + detectRunInput, + resolveProviderOverride, + resolveConfigPath, + loadClusterConfig, + buildStartOptions, + startClusterFromText, + startClusterFromIssue, + startClusterFromFile, + detectGitRepoRoot, +}; diff --git a/lib/stream-json-parser.js b/lib/stream-json-parser.js index 5f523ee9..7226f252 100644 --- a/lib/stream-json-parser.js +++ b/lib/stream-json-parser.js @@ -1,8 +1,77 @@ /** - * Compatibility wrapper for Claude stream-json parsing. - * Prefer provider-specific parsers in src/providers. + * Provider-agnostic stream-json parser for `zeroshot logs`. + * + * Goals: + * - Accept a single stream of JSONL events from multiple provider CLIs (Claude/Codex/Gemini/Opencode) + * - Produce the common event types expected by the logs renderer: + * text, thinking, tool_call, tool_result, result, ... + * + * NOTE: Provider-specific parsers live in `src/providers//output-parser.js`. + * This file is a thin compatibility wrapper so the CLI can parse logs without + * knowing the provider up front (task logs often donā€™t have provider metadata). */ -const { parseEvent, parseChunk } = require('../src/providers/anthropic/output-parser'); + +const anthropic = require('../src/providers/anthropic/output-parser'); +const openai = require('../src/providers/openai/output-parser'); +const google = require('../src/providers/google/output-parser'); +const opencode = require('../src/providers/opencode/output-parser'); + +const googleState = {}; + +function stripTimestampPrefix(line) { + if (!line || typeof line !== 'string') return ''; + let trimmed = line.trim().replace(/\r$/, ''); + if (!trimmed) return ''; + + const tsMatch = trimmed.match(/^\[(\d{13})\](.*)$/); + if (tsMatch) trimmed = (tsMatch[2] || '').trimStart(); + + // In cluster logs, lines can be prefixed like: + // "validator | {json...}" + // Strip the " | " prefix so provider parsers can JSON.parse the event. + if (!trimmed.startsWith('{') && !trimmed.startsWith('[')) { + const pipeMatch = trimmed.match(/^[^|]{1,40}\|\s*(.*)$/); + if (pipeMatch) { + const afterPipe = (pipeMatch[1] || '').trimStart(); + if (afterPipe.startsWith('{') || afterPipe.startsWith('[')) return afterPipe; + } + } + + return trimmed; +} + +function parseEvent(line, options = {}) { + const content = stripTimestampPrefix(line); + if (!content) return null; + + // Try provider parsers in a stable order. + // Each parser returns null for unknown event types, so the first non-null wins. + return ( + anthropic.parseEvent(content, options) || + openai.parseEvent(content, options) || + opencode.parseEvent(content, options) || + google.parseEvent(content, googleState, options) || + null + ); +} + +function parseChunk(chunk, options = {}) { + const events = []; + const lines = String(chunk || '').split('\n'); + + for (const line of lines) { + if (!line.trim()) continue; + const event = parseEvent(line, options); + if (!event) continue; + if (Array.isArray(event)) { + events.push(...event); + } else { + events.push(event); + } + } + + return events; +} module.exports = { parseEvent, diff --git a/lib/tui-binary.js b/lib/tui-binary.js new file mode 100644 index 00000000..a37481d0 --- /dev/null +++ b/lib/tui-binary.js @@ -0,0 +1,121 @@ +'use strict'; +const ENV_BINARY_PATH = 'ZEROSHOT_TUI_BINARY_PATH'; +const ENV_BINARY_URL = 'ZEROSHOT_TUI_BINARY_URL'; +const ENV_BINARY_SKIP = 'ZEROSHOT_TUI_BINARY_SKIP'; + +const fs = require('fs'); +const path = require('path'); + +const PACKAGE_ROOT = path.resolve(__dirname, '..'); +const BIN_DIR = path.join(PACKAGE_ROOT, 'libexec'); +const DEFAULT_RUST_BIN_NAME = process.platform === 'win32' ? 'zeroshot-tui.exe' : 'zeroshot-tui'; + +const SUPPORTED_PLATFORMS = Object.freeze({ + darwin: 'darwin', + linux: 'linux', +}); + +const SUPPORTED_ARCHES = Object.freeze({ + x64: 'x64', + arm64: 'arm64', +}); + +function getPackageVersion() { + // eslint-disable-next-line global-require + const pkg = require(path.join(PACKAGE_ROOT, 'package.json')); + return pkg.version; +} + +function resolveTargetPlatform(platform = process.platform) { + return SUPPORTED_PLATFORMS[platform] || null; +} + +function resolveTargetArch(arch = process.arch) { + return SUPPORTED_ARCHES[arch] || null; +} + +function resolveTarget(platform = process.platform, arch = process.arch) { + const resolvedPlatform = resolveTargetPlatform(platform); + const resolvedArch = resolveTargetArch(arch); + if (!resolvedPlatform || !resolvedArch) { + return null; + } + return { platform: resolvedPlatform, arch: resolvedArch }; +} + +function getAssetName(platform, arch) { + if (!platform || !arch) { + throw new Error('platform and arch are required to build asset name'); + } + return `zeroshot-tui-${platform}-${arch}.tar.gz`; +} + +function getReleaseBaseUrl(version) { + if (!version) { + throw new Error('version is required to build release URL'); + } + return `https://github.com/covibes/zeroshot/releases/download/v${version}`; +} + +function resolveDownloadUrl({ version, platform, arch, overrideUrl } = {}) { + if (overrideUrl) { + return overrideUrl; + } + const target = resolveTarget(platform, arch); + if (!target) { + return null; + } + const resolvedVersion = version || getPackageVersion(); + const assetName = getAssetName(target.platform, target.arch); + return `${getReleaseBaseUrl(resolvedVersion)}/${assetName}`; +} + +function getInstallDir() { + return BIN_DIR; +} + +function getInstalledBinaryPath() { + return path.join(BIN_DIR, DEFAULT_RUST_BIN_NAME); +} + +function resolveBinaryPathOverride(env = process.env) { + const value = env[ENV_BINARY_PATH]; + if (!value) { + return null; + } + const resolved = path.resolve(value); + if (!fs.existsSync(resolved)) { + throw new Error(`Rust TUI binary not found at ${resolved}`); + } + return resolved; +} + +function shouldSkipBinaryInstall(env = process.env) { + const raw = env[ENV_BINARY_SKIP]; + if (!raw) { + return false; + } + const normalized = String(raw).trim().toLowerCase(); + return !['0', 'false', 'no', 'off'].includes(normalized); +} + +module.exports = { + BIN_DIR, + DEFAULT_RUST_BIN_NAME, + ENV_BINARY_PATH, + ENV_BINARY_URL, + ENV_BINARY_SKIP, + SUPPORTED_PLATFORMS, + SUPPORTED_ARCHES, + getAssetName, + getInstallDir, + getInstalledBinaryPath, + getPackageVersion, + getReleaseBaseUrl, + resolveBinaryPathOverride, + resolveDownloadUrl, + resolveTarget, + resolveTargetArch, + resolveTargetPlatform, + shouldSkipBinaryInstall, +}; diff --git a/lib/tui-launcher.js b/lib/tui-launcher.js new file mode 100644 index 00000000..92ca8ec0 --- /dev/null +++ b/lib/tui-launcher.js @@ -0,0 +1,160 @@ +'use strict'; +const TUI_BINARY_ENV = 'ZEROSHOT_TUI_PATH'; +const TUI_BINARY_ENV_ALT = 'ZEROSHOT_TUI_BIN'; +const TUI_INITIAL_SCREEN_ENV = 'ZEROSHOT_TUI_INITIAL_SCREEN'; +const TUI_PROVIDER_OVERRIDE_ENV = 'ZEROSHOT_TUI_PROVIDER_OVERRIDE'; +const TUI_UI_VARIANT_ENV = 'ZEROSHOT_TUI_UI'; + +const fs = require('fs'); +const path = require('path'); +const { spawn } = require('child_process'); +const { normalizeProviderName } = require('./provider-names'); +const { getInstalledBinaryPath, resolveBinaryPathOverride } = require('./tui-binary'); +const { getProvider } = require('../src/providers'); + +const PACKAGE_ROOT = path.resolve(__dirname, '..'); +const DEFAULT_RUST_BIN_NAME = process.platform === 'win32' ? 'zeroshot-tui.exe' : 'zeroshot-tui'; +const VALID_INITIAL_SCREENS = new Set(['launcher', 'monitor']); +const VALID_UI_VARIANTS = new Set(['classic', 'disruptive']); + +function resolveTuiProviderOverride(options = {}) { + const override = options.providerOverride ?? options.provider; + if (!override || (typeof override === 'string' && !override.trim())) { + return null; + } + const normalized = normalizeProviderName(override); + getProvider(normalized); + return normalized; +} + +function resolveInitialScreen(options = {}) { + const initial = options.initialScreen ?? options.initialView; + if (!initial || (typeof initial === 'string' && !initial.trim())) { + return null; + } + const normalized = String(initial).trim().toLowerCase(); + if (!VALID_INITIAL_SCREENS.has(normalized)) { + throw new Error( + `Unknown initial screen: ${normalized}. Valid: ${[...VALID_INITIAL_SCREENS].join(', ')}` + ); + } + return normalized; +} + +function resolveUiVariant(options = {}) { + const ui = options.ui; + if (!ui || (typeof ui === 'string' && !ui.trim())) { + return null; + } + const normalized = String(ui).trim().toLowerCase(); + if (!VALID_UI_VARIANTS.has(normalized)) { + throw new Error( + `Unknown UI variant: ${normalized}. Valid: ${[...VALID_UI_VARIANTS].join(', ')}` + ); + } + return normalized; +} + +function resolveRustTuiBinary() { + const override = resolveBinaryPathOverride(); + if (override) { + return override; + } + + const explicit = process.env[TUI_BINARY_ENV] || process.env[TUI_BINARY_ENV_ALT]; + if (explicit) { + const resolved = path.resolve(explicit); + if (!fs.existsSync(resolved)) { + throw new Error(`Rust TUI binary not found at ${resolved}`); + } + return resolved; + } + + const candidates = [ + path.join(PACKAGE_ROOT, 'tui-rs', 'target', 'debug', DEFAULT_RUST_BIN_NAME), + path.join(PACKAGE_ROOT, 'tui-rs', 'target', 'release', DEFAULT_RUST_BIN_NAME), + ]; + + for (const candidate of candidates) { + if (fs.existsSync(candidate)) { + return candidate; + } + } + + const installed = getInstalledBinaryPath(); + if (fs.existsSync(installed)) { + return installed; + } + + return DEFAULT_RUST_BIN_NAME; +} + +function buildRustTuiCommand(options = {}) { + const initialScreen = resolveInitialScreen(options); + const providerOverride = resolveTuiProviderOverride(options); + const uiVariant = resolveUiVariant(options); + const command = options.binaryPath || resolveRustTuiBinary(); + const args = []; + + if (initialScreen) { + args.push('--initial-screen', initialScreen); + } + + if (uiVariant) { + args.push('--ui', uiVariant); + } + + if (providerOverride) { + args.push('--provider-override', providerOverride); + } + + const env = { ...process.env }; + if (initialScreen) { + env[TUI_INITIAL_SCREEN_ENV] = initialScreen; + } + if (uiVariant) { + env[TUI_UI_VARIANT_ENV] = uiVariant; + } + if (providerOverride) { + env[TUI_PROVIDER_OVERRIDE_ENV] = providerOverride; + } + + return { + command, + args, + env, + cwd: options.cwd || process.cwd(), + }; +} + +function launchRustTui(options = {}) { + const { command, args, env, cwd } = buildRustTuiCommand(options); + const spawnFn = options.spawn || spawn; + const child = spawnFn(command, args, { stdio: 'inherit', env, cwd }); + + if (child && typeof child.on === 'function') { + child.on('error', (error) => { + console.error(`Failed to start Rust TUI (${command}): ${error.message}`); + console.error( + `Set ${TUI_BINARY_ENV} or ZEROSHOT_TUI_BINARY_PATH to a valid Rust TUI binary.` + ); + process.exitCode = 1; + }); + } + + return child; +} + +function launchTuiSession(options = {}) { + return launchRustTui(options); +} + +module.exports = { + buildRustTuiCommand, + launchRustTui, + launchTuiSession, + resolveInitialScreen, + resolveRustTuiBinary, + resolveTuiProviderOverride, + resolveUiVariant, +}; diff --git a/libexec/zeroshot-tui b/libexec/zeroshot-tui new file mode 100755 index 00000000..6a4d4539 Binary files /dev/null and b/libexec/zeroshot-tui differ diff --git a/package-lock.json b/package-lock.json index 86f4f173..dd42d363 100644 --- a/package-lock.json +++ b/package-lock.json @@ -10,18 +10,20 @@ "hasInstallScript": true, "license": "MIT", "dependencies": { - "ajv": "^8.17.1", + "ajv": "^8.18.0", "ansi-to-html": "^0.7.2", "better-sqlite3": "^12.5.0", "blessed": "^0.1.81", "blessed-contrib": "^4.11.0", "chalk": "^4.1.2", "commander": "^14.0.2", + "ink": "^6.6.0", "md-to-pdf": "^5.2.5", "node-pty": "^1.1.0", "omelette": "^0.4.17", "pidusage": "^4.0.1", - "proper-lockfile": "^4.1.2" + "proper-lockfile": "^4.1.2", + "react": "^19.2.4" }, "bin": { "zeroshot": "cli/index.js" @@ -32,8 +34,11 @@ "@semantic-release/github": "^11.0.6", "@types/better-sqlite3": "^7.6.13", "@types/node": "^25.0.3", + "@types/react": "^19.2.9", + "@types/react-reconciler": "^0.33.0", "c8": "^10.1.3", "chai": "^6.2.1", + "concurrently": "^9.1.0", "depcheck": "^1.4.7", "eslint": "^9.39.1", "eslint-config-prettier": "^10.1.8", @@ -94,6 +99,46 @@ "dev": true, "license": "MIT" }, + "node_modules/@alcalzone/ansi-tokenize": { + "version": "0.2.3", + "resolved": "https://registry.npmjs.org/@alcalzone/ansi-tokenize/-/ansi-tokenize-0.2.3.tgz", + "integrity": "sha512-jsElTJ0sQ4wHRz+C45tfect76BwbTbgkgKByOzpCN9xG61N5V6u/glvg1CsNJhq2xJIFpKHSwG3D2wPPuEYOrQ==", + "license": "MIT", + "dependencies": { + "ansi-styles": "^6.2.1", + "is-fullwidth-code-point": "^5.0.0" + }, + "engines": { + "node": ">=18" + } + }, + "node_modules/@alcalzone/ansi-tokenize/node_modules/ansi-styles": { + "version": "6.2.3", + "resolved": "https://registry.npmjs.org/ansi-styles/-/ansi-styles-6.2.3.tgz", + "integrity": "sha512-4Dj6M28JB+oAH8kFkTLUo+a2jwOFkuqb3yucU0CANcRRUbxS0cP0nZYCGjcc3BNXwRIsUVmDGgzawme7zvJHvg==", + "license": "MIT", + "engines": { + "node": ">=12" + }, + "funding": { + "url": "https://github.com/chalk/ansi-styles?sponsor=1" + } + }, + "node_modules/@alcalzone/ansi-tokenize/node_modules/is-fullwidth-code-point": { + "version": "5.1.0", + "resolved": "https://registry.npmjs.org/is-fullwidth-code-point/-/is-fullwidth-code-point-5.1.0.tgz", + "integrity": "sha512-5XHYaSyiqADb4RnZ1Bdad6cPp8Toise4TzEjcOYDHZkTCbKgiUl7WTUCpNWHuxmDt91wnsZBc9xinNzopv3JMQ==", + "license": "MIT", + "dependencies": { + "get-east-asian-width": "^1.3.1" + }, + "engines": { + "node": ">=18" + }, + "funding": { + "url": "https://github.com/sponsors/sindresorhus" + } + }, "node_modules/@babel/code-frame": { "version": "7.27.1", "resolved": "https://registry.npmjs.org/@babel/code-frame/-/code-frame-7.27.1.tgz", @@ -1799,6 +1844,26 @@ "dev": true, "license": "MIT" }, + "node_modules/@types/react": { + "version": "19.2.9", + "resolved": "https://registry.npmjs.org/@types/react/-/react-19.2.9.tgz", + "integrity": "sha512-Lpo8kgb/igvMIPeNV2rsYKTgaORYdO1XGVZ4Qz3akwOj0ySGYMPlQWa8BaLn0G63D1aSaAQ5ldR06wCpChQCjA==", + "devOptional": true, + "license": "MIT", + "dependencies": { + "csstype": "^3.2.2" + } + }, + "node_modules/@types/react-reconciler": { + "version": "0.33.0", + "resolved": "https://registry.npmjs.org/@types/react-reconciler/-/react-reconciler-0.33.0.tgz", + "integrity": "sha512-HZOXsKT0tGI9LlUw2LuedXsVeB88wFa536vVL0M6vE8zN63nI+sSr1ByxmPToP5K5bukaVscyeCJcF9guVNJ1g==", + "dev": true, + "license": "MIT", + "peerDependencies": { + "@types/react": "*" + } + }, "node_modules/@types/yauzl": { "version": "2.10.3", "resolved": "https://registry.npmjs.org/@types/yauzl/-/yauzl-2.10.3.tgz", @@ -1870,32 +1935,6 @@ } } }, - "node_modules/@typescript-eslint/typescript-estree/node_modules/brace-expansion": { - "version": "2.0.2", - "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-2.0.2.tgz", - "integrity": "sha512-Jt0vHyM+jmUBqojB7E1NIYadt0vI0Qxjxd2TErW94wDz+E2LAm5vKMXXwg6ZZBTHPuUlDgQHKXvjGBdfcF1ZDQ==", - "dev": true, - "license": "MIT", - "dependencies": { - "balanced-match": "^1.0.0" - } - }, - "node_modules/@typescript-eslint/typescript-estree/node_modules/minimatch": { - "version": "9.0.3", - "resolved": "https://registry.npmjs.org/minimatch/-/minimatch-9.0.3.tgz", - "integrity": "sha512-RHiac9mvaRw0x3AYRgDC1CxAP7HTcNrrECeA8YYJeWnpo+2Q5CegtZjaotWTWxDG3UeGA1coE05iH1mPjT/2mg==", - "dev": true, - "license": "ISC", - "dependencies": { - "brace-expansion": "^2.0.1" - }, - "engines": { - "node": ">=16 || 14 >=14.17" - }, - "funding": { - "url": "https://github.com/sponsors/isaacs" - } - }, "node_modules/@typescript-eslint/visitor-keys": { "version": "6.21.0", "resolved": "https://registry.npmjs.org/@typescript-eslint/visitor-keys/-/visitor-keys-6.21.0.tgz", @@ -2072,9 +2111,9 @@ } }, "node_modules/ajv": { - "version": "8.17.1", - "resolved": "https://registry.npmjs.org/ajv/-/ajv-8.17.1.tgz", - "integrity": "sha512-B/gBuNg5SiMTrPkC+A2+cW0RszwxYmn6VYxB/inlBStS5nx6xHIt/ehKRhIMhqusl7a8LjQoZnjCs5vhwxOQ1g==", + "version": "8.18.0", + "resolved": "https://registry.npmjs.org/ajv/-/ajv-8.18.0.tgz", + "integrity": "sha512-PlXPeEWMXMZ7sPYOHqmDyCJzcfNrUr3fGNKtezX14ykXOEIvyK81d+qydx89KY5O71FKMPaQ2vBfBFI5NHR63A==", "license": "MIT", "dependencies": { "fast-deep-equal": "^3.1.3", @@ -2271,6 +2310,18 @@ "node": ">= 4.0.0" } }, + "node_modules/auto-bind": { + "version": "5.0.1", + "resolved": "https://registry.npmjs.org/auto-bind/-/auto-bind-5.0.1.tgz", + "integrity": "sha512-ooviqdwwgfIfNmDwo94wlshcdzfO64XV0Cg6oDsDYBJfITDz1EngD2z7DkbvCWn+XIMsIqW27sEVF6qcpJrRcg==", + "license": "MIT", + "engines": { + "node": "^12.20.0 || ^14.13.1 || >=16.0.0" + }, + "funding": { + "url": "https://github.com/sponsors/sindresorhus" + } + }, "node_modules/b4a": { "version": "1.7.3", "resolved": "https://registry.npmjs.org/b4a/-/b4a-1.7.3.tgz", @@ -2298,10 +2349,13 @@ } }, "node_modules/balanced-match": { - "version": "1.0.2", - "resolved": "https://registry.npmjs.org/balanced-match/-/balanced-match-1.0.2.tgz", - "integrity": "sha512-3oSeUO0TMV67hN1AmbXsK4yaqU7tjiHlbxRDZOpH0KW9+CeX4bRAaX0Anxt0tx2MrpRpWwQaPwIlISEJhYU5Pw==", - "license": "MIT" + "version": "4.0.3", + "resolved": "https://registry.npmjs.org/balanced-match/-/balanced-match-4.0.3.tgz", + "integrity": "sha512-1pHv8LX9CpKut1Zp4EXey7Z8OfH11ONNH6Dhi2WDUt31VVZFXZzKwXcysBgqSumFCmR+0dqjMK5v5JiFHzi0+g==", + "license": "MIT", + "engines": { + "node": "20 || >=22" + } }, "node_modules/bare-events": { "version": "2.8.2", @@ -2671,13 +2725,15 @@ "license": "MIT" }, "node_modules/brace-expansion": { - "version": "1.1.12", - "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-1.1.12.tgz", - "integrity": "sha512-9T9UjW3r0UW5c1Q7GTwllptXwhvYmEzFhzMfZ9H7FQWt+uZePjZPjBP/W1ZEyZ1twGWom5/56TF4lPcqjnDHcg==", + "version": "5.0.2", + "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-5.0.2.tgz", + "integrity": "sha512-Pdk8c9poy+YhOgVWw1JNN22/HcivgKWwpxKq04M/jTmHyCZn12WPJebZxdjSa5TmBqISrUSgNYU3eRORljfCCw==", "license": "MIT", "dependencies": { - "balanced-match": "^1.0.0", - "concat-map": "0.0.1" + "balanced-match": "^4.0.2" + }, + "engines": { + "node": "20 || >=22" } }, "node_modules/braces": { @@ -2991,6 +3047,18 @@ "node": ">=6" } }, + "node_modules/cli-boxes": { + "version": "3.0.0", + "resolved": "https://registry.npmjs.org/cli-boxes/-/cli-boxes-3.0.0.tgz", + "integrity": "sha512-/lzGpEWL/8PfI0BmBOPRwp0c/wFNX1RdUML3jK/RcSBA9T8mZDdQpqYBKtCFTOfQbwPqWEOpjqW+Fnayc0969g==", + "license": "MIT", + "engines": { + "node": ">=10" + }, + "funding": { + "url": "https://github.com/sponsors/sindresorhus" + } + }, "node_modules/cli-cursor": { "version": "2.1.0", "resolved": "https://registry.npmjs.org/cli-cursor/-/cli-cursor-2.1.0.tgz", @@ -3253,6 +3321,18 @@ "dev": true, "license": "MIT" }, + "node_modules/code-excerpt": { + "version": "4.0.0", + "resolved": "https://registry.npmjs.org/code-excerpt/-/code-excerpt-4.0.0.tgz", + "integrity": "sha512-xxodCmBen3iy2i0WtAK8FlFNrRzjUqjRsMfho58xT/wvZU1YTM3fCnRjcy1gJPMepaRlgm/0e6w8SpWHpn3/cA==", + "license": "MIT", + "dependencies": { + "convert-to-spaces": "^2.0.1" + }, + "engines": { + "node": "^12.20.0 || ^14.13.1 || >=16.0.0" + } + }, "node_modules/code-point-at": { "version": "1.1.0", "resolved": "https://registry.npmjs.org/code-point-at/-/code-point-at-1.1.0.tgz", @@ -3316,11 +3396,56 @@ "dot-prop": "^5.1.0" } }, - "node_modules/concat-map": { - "version": "0.0.1", - "resolved": "https://registry.npmjs.org/concat-map/-/concat-map-0.0.1.tgz", - "integrity": "sha512-/Srv4dswyQNBfohGpz9o6Yb3Gz3SrUDqBH5rTuhGR7ahtlbYKnVxw2bCFMRljaA7EXHaXZ8wsHdodFvbkhKmqg==", - "license": "MIT" + "node_modules/concurrently": { + "version": "9.2.1", + "resolved": "https://registry.npmjs.org/concurrently/-/concurrently-9.2.1.tgz", + "integrity": "sha512-fsfrO0MxV64Znoy8/l1vVIjjHa29SZyyqPgQBwhiDcaW8wJc2W3XWVOGx4M3oJBnv/zdUZIIp1gDeS98GzP8Ng==", + "dev": true, + "license": "MIT", + "dependencies": { + "chalk": "4.1.2", + "rxjs": "7.8.2", + "shell-quote": "1.8.3", + "supports-color": "8.1.1", + "tree-kill": "1.2.2", + "yargs": "17.7.2" + }, + "bin": { + "conc": "dist/bin/concurrently.js", + "concurrently": "dist/bin/concurrently.js" + }, + "engines": { + "node": ">=18" + }, + "funding": { + "url": "https://github.com/open-cli-tools/concurrently?sponsor=1" + } + }, + "node_modules/concurrently/node_modules/rxjs": { + "version": "7.8.2", + "resolved": "https://registry.npmjs.org/rxjs/-/rxjs-7.8.2.tgz", + "integrity": "sha512-dhKf903U/PQZY6boNNtAGdWbG85WAbjT/1xYoZIC7FAY0yWapOBQVsVrDl58W86//e1VpMNBtRV4MaXfdMySFA==", + "dev": true, + "license": "Apache-2.0", + "dependencies": { + "tslib": "^2.1.0" + } + }, + "node_modules/concurrently/node_modules/supports-color": { + "version": "8.1.1", + "resolved": "https://registry.npmjs.org/supports-color/-/supports-color-8.1.1.tgz", + "integrity": "sha512-MpUEN2OodtUzxvKQl72cUF7RQ5EiHsGvSsVG0ia9c5RbWGL2CI4C7EpPS8UTBIplnlzZiNuV56w+FuNxy3ty2Q==", + "dev": true, + "license": "MIT", + "dependencies": { + "has-flag": "^4.0.0" + }, + "engines": { + "node": ">=10" + }, + "funding": { + "url": "https://github.com/chalk/supports-color?sponsor=1" + } }, "node_modules/config-chain": { "version": "1.1.13", @@ -3430,6 +3555,15 @@ "dev": true, "license": "MIT" }, + "node_modules/convert-to-spaces": { + "version": "2.0.1", + "resolved": "https://registry.npmjs.org/convert-to-spaces/-/convert-to-spaces-2.0.1.tgz", + "integrity": "sha512-rcQ1bsQO9799wq24uE5AM2tAILy4gXGIK/njFWcVQkGNZ96edlpY+A7bjwvzjYvLDyzmG1MmMLZhpcsb+klNMQ==", + "license": "MIT", + "engines": { + "node": "^12.20.0 || ^14.13.1 || >=16.0.0" + } + }, "node_modules/core-util-is": { "version": "1.0.3", "resolved": "https://registry.npmjs.org/core-util-is/-/core-util-is-1.0.3.tgz", @@ -3506,6 +3640,13 @@ "url": "https://github.com/sponsors/sindresorhus" } }, + "node_modules/csstype": { + "version": "3.2.3", + "resolved": "https://registry.npmjs.org/csstype/-/csstype-3.2.3.tgz", + "integrity": "sha512-z1HGKcYy2xA8AGQfwrn0PAy+PB7X/GSj3UVJW9qKyn43xWa+gl5nXmU4qqLMRzWVLFC8KusUX8T/0kCiOYpAIQ==", + "devOptional": true, + "license": "MIT" + }, "node_modules/data-uri-to-buffer": { "version": "6.0.2", "resolved": "https://registry.npmjs.org/data-uri-to-buffer/-/data-uri-to-buffer-6.0.2.tgz", @@ -3667,16 +3808,6 @@ "sprintf-js": "~1.0.2" } }, - "node_modules/depcheck/node_modules/brace-expansion": { - "version": "2.0.2", - "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-2.0.2.tgz", - "integrity": "sha512-Jt0vHyM+jmUBqojB7E1NIYadt0vI0Qxjxd2TErW94wDz+E2LAm5vKMXXwg6ZZBTHPuUlDgQHKXvjGBdfcF1ZDQ==", - "dev": true, - "license": "MIT", - "dependencies": { - "balanced-match": "^1.0.0" - } - }, "node_modules/depcheck/node_modules/cliui": { "version": "7.0.4", "resolved": "https://registry.npmjs.org/cliui/-/cliui-7.0.4.tgz", @@ -3720,22 +3851,6 @@ "js-yaml": "bin/js-yaml.js" } }, - "node_modules/depcheck/node_modules/minimatch": { - "version": "7.4.6", - "resolved": "https://registry.npmjs.org/minimatch/-/minimatch-7.4.6.tgz", - "integrity": "sha512-sBz8G/YjVniEz6lKPNpKxXwazJe4c19fEfV2GDMX6AjFz+MX9uDWIZW8XreVhkFW3fkIdTv/gxWr/Kks5FFAVw==", - "dev": true, - "license": "ISC", - "dependencies": { - "brace-expansion": "^2.0.1" - }, - "engines": { - "node": ">=10" - }, - "funding": { - "url": "https://github.com/sponsors/isaacs" - } - }, "node_modules/depcheck/node_modules/resolve-from": { "version": "5.0.0", "resolved": "https://registry.npmjs.org/resolve-from/-/resolve-from-5.0.0.tgz", @@ -4182,7 +4297,6 @@ "version": "1.1.0", "resolved": "https://registry.npmjs.org/environment/-/environment-1.1.0.tgz", "integrity": "sha512-xUtoPkMggbz0MPyPiIWr1Kp4aeWJjDZ6SMvURhimjdZgsRuDplF5/s9hcgGhyXMhs+6vpnuoiZ2kFiu3FMnS8Q==", - "dev": true, "license": "MIT", "engines": { "node": ">=18" @@ -4230,6 +4344,16 @@ "node": ">= 0.4" } }, + "node_modules/es-toolkit": { + "version": "1.44.0", + "resolved": "https://registry.npmjs.org/es-toolkit/-/es-toolkit-1.44.0.tgz", + "integrity": "sha512-6penXeZalaV88MM3cGkFZZfOoLGWshWWfdy0tWw/RlVVyhvMaWSBTOvXNeiW3e5FwdS5ePW0LGEu17zT139ktg==", + "license": "MIT", + "workspaces": [ + "docs", + "benchmarks" + ] + }, "node_modules/escalade": { "version": "3.2.0", "resolved": "https://registry.npmjs.org/escalade/-/escalade-3.2.0.tgz", @@ -4397,16 +4521,6 @@ "node": "^12.0.0 || ^14.0.0 || >=16.0.0" } }, - "node_modules/eslint-plugin-sonarjs/node_modules/brace-expansion": { - "version": "2.0.2", - "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-2.0.2.tgz", - "integrity": "sha512-Jt0vHyM+jmUBqojB7E1NIYadt0vI0Qxjxd2TErW94wDz+E2LAm5vKMXXwg6ZZBTHPuUlDgQHKXvjGBdfcF1ZDQ==", - "dev": true, - "license": "MIT", - "dependencies": { - "balanced-match": "^1.0.0" - } - }, "node_modules/eslint-plugin-sonarjs/node_modules/bytes": { "version": "3.1.2", "resolved": "https://registry.npmjs.org/bytes/-/bytes-3.1.2.tgz", @@ -4417,22 +4531,6 @@ "node": ">= 0.8" } }, - "node_modules/eslint-plugin-sonarjs/node_modules/minimatch": { - "version": "9.0.5", - "resolved": "https://registry.npmjs.org/minimatch/-/minimatch-9.0.5.tgz", - "integrity": "sha512-G6T0ZX48xgozx7587koeX9Ys2NYy6Gmv//P89sEte9V9whIapMNF4idKxnW2QtCcLiTWlb/wfCabAtAFWhhBow==", - "dev": true, - "license": "ISC", - "dependencies": { - "brace-expansion": "^2.0.1" - }, - "engines": { - "node": ">=16 || 14 >=14.17" - }, - "funding": { - "url": "https://github.com/sponsors/isaacs" - } - }, "node_modules/eslint-plugin-sonarjs/node_modules/semver": { "version": "7.7.2", "resolved": "https://registry.npmjs.org/semver/-/semver-7.7.2.tgz", @@ -5210,7 +5308,6 @@ "version": "1.4.0", "resolved": "https://registry.npmjs.org/get-east-asian-width/-/get-east-asian-width-1.4.0.tgz", "integrity": "sha512-QZjmEOC+IT1uk6Rx0sX22V6uHWVwbdbxf1faPqJ1QhLdGgsRGCZoyaQBm/piRdJy/D2um6hM1UP7ZEeQ4EkP+Q==", - "dev": true, "license": "MIT", "engines": { "node": ">=18" @@ -5379,32 +5476,6 @@ "node": ">=10.13.0" } }, - "node_modules/glob/node_modules/brace-expansion": { - "version": "2.0.2", - "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-2.0.2.tgz", - "integrity": "sha512-Jt0vHyM+jmUBqojB7E1NIYadt0vI0Qxjxd2TErW94wDz+E2LAm5vKMXXwg6ZZBTHPuUlDgQHKXvjGBdfcF1ZDQ==", - "dev": true, - "license": "MIT", - "dependencies": { - "balanced-match": "^1.0.0" - } - }, - "node_modules/glob/node_modules/minimatch": { - "version": "9.0.5", - "resolved": "https://registry.npmjs.org/minimatch/-/minimatch-9.0.5.tgz", - "integrity": "sha512-G6T0ZX48xgozx7587koeX9Ys2NYy6Gmv//P89sEte9V9whIapMNF4idKxnW2QtCcLiTWlb/wfCabAtAFWhhBow==", - "dev": true, - "license": "ISC", - "dependencies": { - "brace-expansion": "^2.0.1" - }, - "engines": { - "node": ">=16 || 14 >=14.17" - }, - "funding": { - "url": "https://github.com/sponsors/isaacs" - } - }, "node_modules/global-modules": { "version": "1.0.0", "resolved": "https://registry.npmjs.org/global-modules/-/global-modules-1.0.0.tgz", @@ -5881,18 +5952,319 @@ "wrappy": "1" } }, - "node_modules/inherits": { - "version": "2.0.4", - "resolved": "https://registry.npmjs.org/inherits/-/inherits-2.0.4.tgz", - "integrity": "sha512-k/vGaX4/Yla3WzyMCvTQOXYeIHvqOKtnqBduzTHpzpQZzAskKMhZ2K+EnBiSM9zGSoIFeMpXKxa4dYeZIQqewQ==", - "license": "ISC" - }, - "node_modules/ini": { - "version": "1.3.8", - "resolved": "https://registry.npmjs.org/ini/-/ini-1.3.8.tgz", - "integrity": "sha512-JV/yugV2uzW5iMRSiZAyDtQd+nxtUnjeLt0acNdw98kKLrvuRVyB80tsREOE7yvGVgalhZ6RNXCmEHkUKBKxew==", - "license": "ISC" - }, + "node_modules/inherits": { + "version": "2.0.4", + "resolved": "https://registry.npmjs.org/inherits/-/inherits-2.0.4.tgz", + "integrity": "sha512-k/vGaX4/Yla3WzyMCvTQOXYeIHvqOKtnqBduzTHpzpQZzAskKMhZ2K+EnBiSM9zGSoIFeMpXKxa4dYeZIQqewQ==", + "license": "ISC" + }, + "node_modules/ini": { + "version": "1.3.8", + "resolved": "https://registry.npmjs.org/ini/-/ini-1.3.8.tgz", + "integrity": "sha512-JV/yugV2uzW5iMRSiZAyDtQd+nxtUnjeLt0acNdw98kKLrvuRVyB80tsREOE7yvGVgalhZ6RNXCmEHkUKBKxew==", + "license": "ISC" + }, + "node_modules/ink": { + "version": "6.6.0", + "resolved": "https://registry.npmjs.org/ink/-/ink-6.6.0.tgz", + "integrity": "sha512-QDt6FgJxgmSxAelcOvOHUvFxbIUjVpCH5bx+Slvc5m7IEcpGt3dYwbz/L+oRnqEGeRvwy1tineKK4ect3nW1vQ==", + "license": "MIT", + "dependencies": { + "@alcalzone/ansi-tokenize": "^0.2.1", + "ansi-escapes": "^7.2.0", + "ansi-styles": "^6.2.1", + "auto-bind": "^5.0.1", + "chalk": "^5.6.0", + "cli-boxes": "^3.0.0", + "cli-cursor": "^4.0.0", + "cli-truncate": "^5.1.1", + "code-excerpt": "^4.0.0", + "es-toolkit": "^1.39.10", + "indent-string": "^5.0.0", + "is-in-ci": "^2.0.0", + "patch-console": "^2.0.0", + "react-reconciler": "^0.33.0", + "signal-exit": "^3.0.7", + "slice-ansi": "^7.1.0", + "stack-utils": "^2.0.6", + "string-width": "^8.1.0", + "type-fest": "^4.27.0", + "widest-line": "^5.0.0", + "wrap-ansi": "^9.0.0", + "ws": "^8.18.0", + "yoga-layout": "~3.2.1" + }, + "engines": { + "node": ">=20" + }, + "peerDependencies": { + "@types/react": ">=19.0.0", + "react": ">=19.0.0", + "react-devtools-core": "^6.1.2" + }, + "peerDependenciesMeta": { + "@types/react": { + "optional": true + }, + "react-devtools-core": { + "optional": true + } + } + }, + "node_modules/ink/node_modules/ansi-escapes": { + "version": "7.2.0", + "resolved": "https://registry.npmjs.org/ansi-escapes/-/ansi-escapes-7.2.0.tgz", + "integrity": "sha512-g6LhBsl+GBPRWGWsBtutpzBYuIIdBkLEvad5C/va/74Db018+5TZiyA26cZJAr3Rft5lprVqOIPxf5Vid6tqAw==", + "license": "MIT", + "dependencies": { + "environment": "^1.0.0" + }, + "engines": { + "node": ">=18" + }, + "funding": { + "url": "https://github.com/sponsors/sindresorhus" + } + }, + "node_modules/ink/node_modules/ansi-regex": { + "version": "6.2.2", + "resolved": "https://registry.npmjs.org/ansi-regex/-/ansi-regex-6.2.2.tgz", + "integrity": "sha512-Bq3SmSpyFHaWjPk8If9yc6svM8c56dB5BAtW4Qbw5jHTwwXXcTLoRMkpDJp6VL0XzlWaCHTXrkFURMYmD0sLqg==", + "license": "MIT", + "engines": { + "node": ">=12" + }, + "funding": { + "url": "https://github.com/chalk/ansi-regex?sponsor=1" + } + }, + "node_modules/ink/node_modules/ansi-styles": { + "version": "6.2.3", + "resolved": "https://registry.npmjs.org/ansi-styles/-/ansi-styles-6.2.3.tgz", + "integrity": "sha512-4Dj6M28JB+oAH8kFkTLUo+a2jwOFkuqb3yucU0CANcRRUbxS0cP0nZYCGjcc3BNXwRIsUVmDGgzawme7zvJHvg==", + "license": "MIT", + "engines": { + "node": ">=12" + }, + "funding": { + "url": "https://github.com/chalk/ansi-styles?sponsor=1" + } + }, + "node_modules/ink/node_modules/chalk": { + "version": "5.6.2", + "resolved": "https://registry.npmjs.org/chalk/-/chalk-5.6.2.tgz", + "integrity": "sha512-7NzBL0rN6fMUW+f7A6Io4h40qQlG+xGmtMxfbnH/K7TAtt8JQWVQK+6g0UXKMeVJoyV5EkkNsErQ8pVD3bLHbA==", + "license": "MIT", + "engines": { + "node": "^12.17.0 || ^14.13 || >=16.0.0" + }, + "funding": { + "url": "https://github.com/chalk/chalk?sponsor=1" + } + }, + "node_modules/ink/node_modules/cli-cursor": { + "version": "4.0.0", + "resolved": "https://registry.npmjs.org/cli-cursor/-/cli-cursor-4.0.0.tgz", + "integrity": "sha512-VGtlMu3x/4DOtIUwEkRezxUZ2lBacNJCHash0N0WeZDBS+7Ux1dm3XWAgWYxLJFMMdOeXMHXorshEFhbMSGelg==", + "license": "MIT", + "dependencies": { + "restore-cursor": "^4.0.0" + }, + "engines": { + "node": "^12.20.0 || ^14.13.1 || >=16.0.0" + }, + "funding": { + "url": "https://github.com/sponsors/sindresorhus" + } + }, + "node_modules/ink/node_modules/cli-truncate": { + "version": "5.1.1", + "resolved": "https://registry.npmjs.org/cli-truncate/-/cli-truncate-5.1.1.tgz", + "integrity": "sha512-SroPvNHxUnk+vIW/dOSfNqdy1sPEFkrTk6TUtqLCnBlo3N7TNYYkzzN7uSD6+jVjrdO4+p8nH7JzH6cIvUem6A==", + "license": "MIT", + "dependencies": { + "slice-ansi": "^7.1.0", + "string-width": "^8.0.0" + }, + "engines": { + "node": ">=20" + }, + "funding": { + "url": "https://github.com/sponsors/sindresorhus" + } + }, + "node_modules/ink/node_modules/emoji-regex": { + "version": "10.6.0", + "resolved": "https://registry.npmjs.org/emoji-regex/-/emoji-regex-10.6.0.tgz", + "integrity": "sha512-toUI84YS5YmxW219erniWD0CIVOo46xGKColeNQRgOzDorgBi1v4D71/OFzgD9GO2UGKIv1C3Sp8DAn0+j5w7A==", + "license": "MIT" + }, + "node_modules/ink/node_modules/indent-string": { + "version": "5.0.0", + "resolved": "https://registry.npmjs.org/indent-string/-/indent-string-5.0.0.tgz", + "integrity": "sha512-m6FAo/spmsW2Ab2fU35JTYwtOKa2yAwXSwgjSv1TJzh4Mh7mC3lzAOVLBprb72XsTrgkEIsl7YrFNAiDiRhIGg==", + "license": "MIT", + "engines": { + "node": ">=12" + }, + "funding": { + "url": "https://github.com/sponsors/sindresorhus" + } + }, + "node_modules/ink/node_modules/is-fullwidth-code-point": { + "version": "5.1.0", + "resolved": "https://registry.npmjs.org/is-fullwidth-code-point/-/is-fullwidth-code-point-5.1.0.tgz", + "integrity": "sha512-5XHYaSyiqADb4RnZ1Bdad6cPp8Toise4TzEjcOYDHZkTCbKgiUl7WTUCpNWHuxmDt91wnsZBc9xinNzopv3JMQ==", + "license": "MIT", + "dependencies": { + "get-east-asian-width": "^1.3.1" + }, + "engines": { + "node": ">=18" + }, + "funding": { + "url": "https://github.com/sponsors/sindresorhus" + } + }, + "node_modules/ink/node_modules/mimic-fn": { + "version": "2.1.0", + "resolved": "https://registry.npmjs.org/mimic-fn/-/mimic-fn-2.1.0.tgz", + "integrity": "sha512-OqbOk5oEQeAZ8WXWydlu9HJjz9WVdEIvamMCcXmuqUYjTknH/sqsWvhQ3vgwKFRR1HpjvNBKQ37nbJgYzGqGcg==", + "license": "MIT", + "engines": { + "node": ">=6" + } + }, + "node_modules/ink/node_modules/onetime": { + "version": "5.1.2", + "resolved": "https://registry.npmjs.org/onetime/-/onetime-5.1.2.tgz", + "integrity": "sha512-kbpaSSGJTWdAY5KPVeMOKXSrPtr8C8C7wodJbcsd51jRnmD+GZu8Y0VoU6Dm5Z4vWr0Ig/1NKuWRKf7j5aaYSg==", + "license": "MIT", + "dependencies": { + "mimic-fn": "^2.1.0" + }, + "engines": { + "node": ">=6" + }, + "funding": { + "url": "https://github.com/sponsors/sindresorhus" + } + }, + "node_modules/ink/node_modules/restore-cursor": { + "version": "4.0.0", + "resolved": "https://registry.npmjs.org/restore-cursor/-/restore-cursor-4.0.0.tgz", + "integrity": "sha512-I9fPXU9geO9bHOt9pHHOhOkYerIMsmVaWB0rA2AI9ERh/+x/i7MV5HKBNrg+ljO5eoPVgCcnFuRjJ9uH6I/3eg==", + "license": "MIT", + "dependencies": { + "onetime": "^5.1.0", + "signal-exit": "^3.0.2" + }, + "engines": { + "node": "^12.20.0 || ^14.13.1 || >=16.0.0" + }, + "funding": { + "url": "https://github.com/sponsors/sindresorhus" + } + }, + "node_modules/ink/node_modules/signal-exit": { + "version": "3.0.7", + "resolved": "https://registry.npmjs.org/signal-exit/-/signal-exit-3.0.7.tgz", + "integrity": "sha512-wnD2ZE+l+SPC/uoS0vXeE9L1+0wuaMqKlfz9AMUo38JsyLSBWSFcHR1Rri62LZc12vLr1gb3jl7iwQhgwpAbGQ==", + "license": "ISC" + }, + "node_modules/ink/node_modules/slice-ansi": { + "version": "7.1.2", + "resolved": "https://registry.npmjs.org/slice-ansi/-/slice-ansi-7.1.2.tgz", + "integrity": "sha512-iOBWFgUX7caIZiuutICxVgX1SdxwAVFFKwt1EvMYYec/NWO5meOJ6K5uQxhrYBdQJne4KxiqZc+KptFOWFSI9w==", + "license": "MIT", + "dependencies": { + "ansi-styles": "^6.2.1", + "is-fullwidth-code-point": "^5.0.0" + }, + "engines": { + "node": ">=18" + }, + "funding": { + "url": "https://github.com/chalk/slice-ansi?sponsor=1" + } + }, + "node_modules/ink/node_modules/string-width": { + "version": "8.1.0", + "resolved": "https://registry.npmjs.org/string-width/-/string-width-8.1.0.tgz", + "integrity": "sha512-Kxl3KJGb/gxkaUMOjRsQ8IrXiGW75O4E3RPjFIINOVH8AMl2SQ/yWdTzWwF3FevIX9LcMAjJW+GRwAlAbTSXdg==", + "license": "MIT", + "dependencies": { + "get-east-asian-width": "^1.3.0", + "strip-ansi": "^7.1.0" + }, + "engines": { + "node": ">=20" + }, + "funding": { + "url": "https://github.com/sponsors/sindresorhus" + } + }, + "node_modules/ink/node_modules/strip-ansi": { + "version": "7.1.2", + "resolved": "https://registry.npmjs.org/strip-ansi/-/strip-ansi-7.1.2.tgz", + "integrity": "sha512-gmBGslpoQJtgnMAvOVqGZpEz9dyoKTCzy2nfz/n8aIFhN/jCE/rCmcxabB6jOOHV+0WNnylOxaxBQPSvcWklhA==", + "license": "MIT", + "dependencies": { + "ansi-regex": "^6.0.1" + }, + "engines": { + "node": ">=12" + }, + "funding": { + "url": "https://github.com/chalk/strip-ansi?sponsor=1" + } + }, + "node_modules/ink/node_modules/type-fest": { + "version": "4.41.0", + "resolved": "https://registry.npmjs.org/type-fest/-/type-fest-4.41.0.tgz", + "integrity": "sha512-TeTSQ6H5YHvpqVwBRcnLDCBnDOHWYu7IvGbHT6N8AOymcr9PJGjc1GTtiWZTYg0NCgYwvnYWEkVChQAr9bjfwA==", + "license": "(MIT OR CC0-1.0)", + "engines": { + "node": ">=16" + }, + "funding": { + "url": "https://github.com/sponsors/sindresorhus" + } + }, + "node_modules/ink/node_modules/wrap-ansi": { + "version": "9.0.2", + "resolved": "https://registry.npmjs.org/wrap-ansi/-/wrap-ansi-9.0.2.tgz", + "integrity": "sha512-42AtmgqjV+X1VpdOfyTGOYRi0/zsoLqtXQckTmqTeybT+BDIbM/Guxo7x3pE2vtpr1ok6xRqM9OpBe+Jyoqyww==", + "license": "MIT", + "dependencies": { + "ansi-styles": "^6.2.1", + "string-width": "^7.0.0", + "strip-ansi": "^7.1.0" + }, + "engines": { + "node": ">=18" + }, + "funding": { + "url": "https://github.com/chalk/wrap-ansi?sponsor=1" + } + }, + "node_modules/ink/node_modules/wrap-ansi/node_modules/string-width": { + "version": "7.2.0", + "resolved": "https://registry.npmjs.org/string-width/-/string-width-7.2.0.tgz", + "integrity": "sha512-tsaTIkKW9b4N+AEj+SVA+WhJzV7/zMhcSu78mLKWSk7cXMOSHsBKFWUs0fWwq8QyK3MgJBQRX6Gbi4kYbdvGkQ==", + "license": "MIT", + "dependencies": { + "emoji-regex": "^10.3.0", + "get-east-asian-width": "^1.0.0", + "strip-ansi": "^7.1.0" + }, + "engines": { + "node": ">=18" + }, + "funding": { + "url": "https://github.com/sponsors/sindresorhus" + } + }, "node_modules/into-stream": { "version": "7.0.0", "resolved": "https://registry.npmjs.org/into-stream/-/into-stream-7.0.0.tgz", @@ -6014,6 +6386,21 @@ "node": ">=0.10.0" } }, + "node_modules/is-in-ci": { + "version": "2.0.0", + "resolved": "https://registry.npmjs.org/is-in-ci/-/is-in-ci-2.0.0.tgz", + "integrity": "sha512-cFeerHriAnhrQSbpAxL37W1wcJKUUX07HyLWZCW1URJT/ra3GyUTzBgUnh24TMVfNTV2Hij2HLxkPHFZfOZy5w==", + "license": "MIT", + "bin": { + "is-in-ci": "cli.js" + }, + "engines": { + "node": ">=20" + }, + "funding": { + "url": "https://github.com/sponsors/sindresorhus" + } + }, "node_modules/is-interactive": { "version": "1.0.0", "resolved": "https://registry.npmjs.org/is-interactive/-/is-interactive-1.0.0.tgz", @@ -6959,15 +7346,15 @@ } }, "node_modules/lodash": { - "version": "4.17.21", - "resolved": "https://registry.npmjs.org/lodash/-/lodash-4.17.21.tgz", - "integrity": "sha512-v2kDEe57lecTulaDIuNTPy3Ry4gLGJ6Z1O3vE1krgXZNrsQ+LFTGHVxVjcXPs17LhbZVGedAJv8XZ1tvj5FvSg==", + "version": "4.17.23", + "resolved": "https://registry.npmjs.org/lodash/-/lodash-4.17.23.tgz", + "integrity": "sha512-LgVTMpQtIopCi79SJeDiP0TfWi5CNEc/L/aRdTh3yIvmZXTnheWpKjSZhnvMl8iXbC1tFg9gdHHDMLoV7CnG+w==", "license": "MIT" }, "node_modules/lodash-es": { - "version": "4.17.22", - "resolved": "https://registry.npmjs.org/lodash-es/-/lodash-es-4.17.22.tgz", - "integrity": "sha512-XEawp1t0gxSi9x01glktRZ5HDy0HXqrM0x5pXQM98EaI0NxO6jVM7omDOxsuEo5UIASAnm2bRp1Jt/e0a2XU8Q==", + "version": "4.17.23", + "resolved": "https://registry.npmjs.org/lodash-es/-/lodash-es-4.17.23.tgz", + "integrity": "sha512-kVI48u3PZr38HdYz98UmfPnXl2DXrpdctLrFLCd3kOx1xUkOmpFPx7gCWWM5MPkL/fD8zb+Ph0QzjGFs4+hHWg==", "dev": true, "license": "MIT" }, @@ -7392,15 +7779,18 @@ } }, "node_modules/minimatch": { - "version": "3.1.2", - "resolved": "https://registry.npmjs.org/minimatch/-/minimatch-3.1.2.tgz", - "integrity": "sha512-J7p63hRiAjw1NDEww1W7i37+ByIrOWO5XQQAzZ3VOcL0PNybwpfmV/N05zFAzwQ9USyEcX6t3UO+K5aqBQOIHw==", - "license": "ISC", + "version": "10.2.2", + "resolved": "https://registry.npmjs.org/minimatch/-/minimatch-10.2.2.tgz", + "integrity": "sha512-+G4CpNBxa5MprY+04MbgOw1v7So6n5JY166pFi9KfYwT78fxScCeSNQSNzp6dpPSW2rONOps6Ocam1wFhCgoVw==", + "license": "BlueOak-1.0.0", "dependencies": { - "brace-expansion": "^1.1.7" + "brace-expansion": "^5.0.2" }, "engines": { - "node": "*" + "node": "18 || 20 || >=22" + }, + "funding": { + "url": "https://github.com/sponsors/isaacs" } }, "node_modules/minimist": { @@ -7484,16 +7874,6 @@ "node": "^18.18.0 || ^20.9.0 || >=21.1.0" } }, - "node_modules/mocha/node_modules/brace-expansion": { - "version": "2.0.2", - "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-2.0.2.tgz", - "integrity": "sha512-Jt0vHyM+jmUBqojB7E1NIYadt0vI0Qxjxd2TErW94wDz+E2LAm5vKMXXwg6ZZBTHPuUlDgQHKXvjGBdfcF1ZDQ==", - "dev": true, - "license": "MIT", - "dependencies": { - "balanced-match": "^1.0.0" - } - }, "node_modules/mocha/node_modules/chokidar": { "version": "4.0.3", "resolved": "https://registry.npmjs.org/chokidar/-/chokidar-4.0.3.tgz", @@ -7527,22 +7907,6 @@ "url": "https://github.com/sponsors/sindresorhus" } }, - "node_modules/mocha/node_modules/minimatch": { - "version": "9.0.5", - "resolved": "https://registry.npmjs.org/minimatch/-/minimatch-9.0.5.tgz", - "integrity": "sha512-G6T0ZX48xgozx7587koeX9Ys2NYy6Gmv//P89sEte9V9whIapMNF4idKxnW2QtCcLiTWlb/wfCabAtAFWhhBow==", - "dev": true, - "license": "ISC", - "dependencies": { - "brace-expansion": "^2.0.1" - }, - "engines": { - "node": ">=16 || 14 >=14.17" - }, - "funding": { - "url": "https://github.com/sponsors/isaacs" - } - }, "node_modules/mocha/node_modules/readdirp": { "version": "4.1.2", "resolved": "https://registry.npmjs.org/readdirp/-/readdirp-4.1.2.tgz", @@ -7784,9 +8148,9 @@ } }, "node_modules/npm": { - "version": "11.7.0", - "resolved": "https://registry.npmjs.org/npm/-/npm-11.7.0.tgz", - "integrity": "sha512-wiCZpv/41bIobCoJ31NStIWKfAxxYyD1iYnWCtiyns8s5v3+l8y0HCP/sScuH6B5+GhIfda4HQKiqeGZwJWhFw==", + "version": "11.8.0", + "resolved": "https://registry.npmjs.org/npm/-/npm-11.8.0.tgz", + "integrity": "sha512-n19sJeW+RGKdkHo8SCc5xhSwkKhQUFfZaFzSc+EsYXLjSqIV0tl72aDYQVuzVvfrbysGwdaQsNLNy58J10EBSQ==", "bundleDependencies": [ "@isaacs/string-locale-compare", "@npmcli/arborist", @@ -7866,8 +8230,8 @@ ], "dependencies": { "@isaacs/string-locale-compare": "^1.1.0", - "@npmcli/arborist": "^9.1.9", - "@npmcli/config": "^10.4.5", + "@npmcli/arborist": "^9.1.10", + "@npmcli/config": "^10.5.0", "@npmcli/fs": "^5.0.0", "@npmcli/map-workspaces": "^5.0.3", "@npmcli/metavuln-calculator": "^9.0.3", @@ -7875,7 +8239,7 @@ "@npmcli/promise-spawn": "^9.0.1", "@npmcli/redact": "^4.0.0", "@npmcli/run-script": "^10.0.3", - "@sigstore/tuf": "^4.0.0", + "@sigstore/tuf": "^4.0.1", "abbrev": "^4.0.0", "archy": "~1.0.0", "cacache": "^20.0.3", @@ -7892,11 +8256,11 @@ "is-cidr": "^6.0.1", "json-parse-even-better-errors": "^5.0.0", "libnpmaccess": "^10.0.3", - "libnpmdiff": "^8.0.12", - "libnpmexec": "^10.1.11", - "libnpmfund": "^7.0.12", + "libnpmdiff": "^8.0.13", + "libnpmexec": "^10.1.12", + "libnpmfund": "^7.0.13", "libnpmorg": "^8.0.1", - "libnpmpack": "^9.0.12", + "libnpmpack": "^9.0.13", "libnpmpublish": "^11.1.3", "libnpmsearch": "^9.0.1", "libnpmteam": "^8.0.2", @@ -7925,11 +8289,11 @@ "spdx-expression-parse": "^4.0.0", "ssri": "^13.0.0", "supports-color": "^10.2.2", - "tar": "^7.5.2", + "tar": "^7.5.4", "text-table": "~0.2.0", "tiny-relative-date": "^2.0.2", "treeverse": "^3.0.0", - "validate-npm-package-name": "^7.0.0", + "validate-npm-package-name": "^7.0.2", "which": "^6.0.0" }, "bin": { @@ -8009,7 +8373,7 @@ } }, "node_modules/npm/node_modules/@npmcli/arborist": { - "version": "9.1.9", + "version": "9.1.10", "dev": true, "inBundle": true, "license": "ISC", @@ -8027,7 +8391,7 @@ "@npmcli/run-script": "^10.0.0", "bin-links": "^6.0.0", "cacache": "^20.0.1", - "common-ancestor-path": "^1.0.1", + "common-ancestor-path": "^2.0.0", "hosted-git-info": "^9.0.0", "json-stringify-nice": "^1.1.4", "lru-cache": "^11.2.1", @@ -8056,7 +8420,7 @@ } }, "node_modules/npm/node_modules/@npmcli/config": { - "version": "10.4.5", + "version": "10.5.0", "dev": true, "inBundle": true, "license": "ISC", @@ -8251,7 +8615,7 @@ } }, "node_modules/npm/node_modules/@sigstore/core": { - "version": "3.0.0", + "version": "3.1.0", "dev": true, "inBundle": true, "license": "Apache-2.0", @@ -8269,52 +8633,43 @@ } }, "node_modules/npm/node_modules/@sigstore/sign": { - "version": "4.0.1", + "version": "4.1.0", "dev": true, "inBundle": true, "license": "Apache-2.0", "dependencies": { "@sigstore/bundle": "^4.0.0", - "@sigstore/core": "^3.0.0", + "@sigstore/core": "^3.1.0", "@sigstore/protobuf-specs": "^0.5.0", - "make-fetch-happen": "^15.0.2", - "proc-log": "^5.0.0", + "make-fetch-happen": "^15.0.3", + "proc-log": "^6.1.0", "promise-retry": "^2.0.1" }, "engines": { "node": "^20.17.0 || >=22.9.0" } }, - "node_modules/npm/node_modules/@sigstore/sign/node_modules/proc-log": { - "version": "5.0.0", - "dev": true, - "inBundle": true, - "license": "ISC", - "engines": { - "node": "^18.17.0 || >=20.5.0" - } - }, "node_modules/npm/node_modules/@sigstore/tuf": { - "version": "4.0.0", + "version": "4.0.1", "dev": true, "inBundle": true, "license": "Apache-2.0", "dependencies": { "@sigstore/protobuf-specs": "^0.5.0", - "tuf-js": "^4.0.0" + "tuf-js": "^4.1.0" }, "engines": { "node": "^20.17.0 || >=22.9.0" } }, "node_modules/npm/node_modules/@sigstore/verify": { - "version": "3.0.0", + "version": "3.1.0", "dev": true, "inBundle": true, "license": "Apache-2.0", "dependencies": { "@sigstore/bundle": "^4.0.0", - "@sigstore/core": "^3.0.0", + "@sigstore/core": "^3.1.0", "@sigstore/protobuf-specs": "^0.5.0" }, "engines": { @@ -8331,33 +8686,18 @@ } }, "node_modules/npm/node_modules/@tufjs/models": { - "version": "4.0.0", + "version": "4.1.0", "dev": true, "inBundle": true, "license": "MIT", "dependencies": { "@tufjs/canonical-json": "2.0.0", - "minimatch": "^9.0.5" + "minimatch": "^10.1.1" }, "engines": { "node": "^20.17.0 || >=22.9.0" } }, - "node_modules/npm/node_modules/@tufjs/models/node_modules/minimatch": { - "version": "9.0.5", - "dev": true, - "inBundle": true, - "license": "ISC", - "dependencies": { - "brace-expansion": "^2.0.1" - }, - "engines": { - "node": ">=16 || 14 >=14.17" - }, - "funding": { - "url": "https://github.com/sponsors/isaacs" - } - }, "node_modules/npm/node_modules/abbrev": { "version": "4.0.0", "dev": true, @@ -8397,12 +8737,6 @@ "inBundle": true, "license": "MIT" }, - "node_modules/npm/node_modules/balanced-match": { - "version": "1.0.2", - "dev": true, - "inBundle": true, - "license": "MIT" - }, "node_modules/npm/node_modules/bin-links": { "version": "6.0.0", "dev": true, @@ -8431,15 +8765,6 @@ "url": "https://github.com/sponsors/sindresorhus" } }, - "node_modules/npm/node_modules/brace-expansion": { - "version": "2.0.2", - "dev": true, - "inBundle": true, - "license": "MIT", - "dependencies": { - "balanced-match": "^1.0.0" - } - }, "node_modules/npm/node_modules/cacache": { "version": "20.0.3", "dev": true, @@ -8533,10 +8858,13 @@ } }, "node_modules/npm/node_modules/common-ancestor-path": { - "version": "1.0.1", + "version": "2.0.0", "dev": true, "inBundle": true, - "license": "ISC" + "license": "BlueOak-1.0.0", + "engines": { + "node": ">= 18" + } }, "node_modules/npm/node_modules/cssesc": { "version": "3.0.0", @@ -8568,7 +8896,7 @@ } }, "node_modules/npm/node_modules/diff": { - "version": "8.0.2", + "version": "8.0.3", "dev": true, "inBundle": true, "license": "BSD-3-Clause", @@ -8763,7 +9091,7 @@ } }, "node_modules/npm/node_modules/ip-address": { - "version": "10.0.1", + "version": "10.1.0", "dev": true, "inBundle": true, "license": "MIT", @@ -8866,12 +9194,12 @@ } }, "node_modules/npm/node_modules/libnpmdiff": { - "version": "8.0.12", + "version": "8.0.13", "dev": true, "inBundle": true, "license": "ISC", "dependencies": { - "@npmcli/arborist": "^9.1.9", + "@npmcli/arborist": "^9.1.10", "@npmcli/installed-package-contents": "^4.0.0", "binary-extensions": "^3.0.0", "diff": "^8.0.2", @@ -8885,12 +9213,12 @@ } }, "node_modules/npm/node_modules/libnpmexec": { - "version": "10.1.11", + "version": "10.1.12", "dev": true, "inBundle": true, "license": "ISC", "dependencies": { - "@npmcli/arborist": "^9.1.9", + "@npmcli/arborist": "^9.1.10", "@npmcli/package-json": "^7.0.0", "@npmcli/run-script": "^10.0.0", "ci-info": "^4.0.0", @@ -8908,12 +9236,12 @@ } }, "node_modules/npm/node_modules/libnpmfund": { - "version": "7.0.12", + "version": "7.0.13", "dev": true, "inBundle": true, "license": "ISC", "dependencies": { - "@npmcli/arborist": "^9.1.9" + "@npmcli/arborist": "^9.1.10" }, "engines": { "node": "^20.17.0 || >=22.9.0" @@ -8933,12 +9261,12 @@ } }, "node_modules/npm/node_modules/libnpmpack": { - "version": "9.0.12", + "version": "9.0.13", "dev": true, "inBundle": true, "license": "ISC", "dependencies": { - "@npmcli/arborist": "^9.1.9", + "@npmcli/arborist": "^9.1.10", "@npmcli/run-script": "^10.0.0", "npm-package-arg": "^13.0.0", "pacote": "^21.0.2" @@ -9008,10 +9336,10 @@ } }, "node_modules/npm/node_modules/lru-cache": { - "version": "11.2.2", + "version": "11.2.4", "dev": true, "inBundle": true, - "license": "ISC", + "license": "BlueOak-1.0.0", "engines": { "node": "20 || >=22" } @@ -9422,7 +9750,7 @@ } }, "node_modules/npm/node_modules/path-scurry": { - "version": "2.0.0", + "version": "2.0.1", "dev": true, "inBundle": true, "license": "BlueOak-1.0.0", @@ -9438,7 +9766,7 @@ } }, "node_modules/npm/node_modules/postcss-selector-parser": { - "version": "7.1.0", + "version": "7.1.1", "dev": true, "inBundle": true, "license": "MIT", @@ -9581,17 +9909,17 @@ } }, "node_modules/npm/node_modules/sigstore": { - "version": "4.0.0", + "version": "4.1.0", "dev": true, "inBundle": true, "license": "Apache-2.0", "dependencies": { "@sigstore/bundle": "^4.0.0", - "@sigstore/core": "^3.0.0", + "@sigstore/core": "^3.1.0", "@sigstore/protobuf-specs": "^0.5.0", - "@sigstore/sign": "^4.0.0", - "@sigstore/tuf": "^4.0.0", - "@sigstore/verify": "^3.0.0" + "@sigstore/sign": "^4.1.0", + "@sigstore/tuf": "^4.0.1", + "@sigstore/verify": "^3.1.0" }, "engines": { "node": "^20.17.0 || >=22.9.0" @@ -9728,7 +10056,7 @@ } }, "node_modules/npm/node_modules/tar": { - "version": "7.5.2", + "version": "7.5.4", "dev": true, "inBundle": true, "license": "BlueOak-1.0.0", @@ -9819,14 +10147,14 @@ } }, "node_modules/npm/node_modules/tuf-js": { - "version": "4.0.0", + "version": "4.1.0", "dev": true, "inBundle": true, "license": "MIT", "dependencies": { - "@tufjs/models": "4.0.0", - "debug": "^4.4.1", - "make-fetch-happen": "^15.0.0" + "@tufjs/models": "4.1.0", + "debug": "^4.4.3", + "make-fetch-happen": "^15.0.1" }, "engines": { "node": "^20.17.0 || >=22.9.0" @@ -9883,7 +10211,7 @@ } }, "node_modules/npm/node_modules/validate-npm-package-name": { - "version": "7.0.0", + "version": "7.0.2", "dev": true, "inBundle": true, "license": "ISC", @@ -10390,6 +10718,15 @@ "dev": true, "license": "MIT" }, + "node_modules/patch-console": { + "version": "2.0.0", + "resolved": "https://registry.npmjs.org/patch-console/-/patch-console-2.0.0.tgz", + "integrity": "sha512-0YNdUceMdaQwoKce1gatDScmMo5pu/tfABfnzEqeG0gtTmd7mh/WcwgUjtAeOU7N8nFFlbQBnFK2gXW5fGvmMA==", + "license": "MIT", + "engines": { + "node": "^12.20.0 || ^14.13.1 || >=16.0.0" + } + }, "node_modules/path-browserify": { "version": "1.0.1", "resolved": "https://registry.npmjs.org/path-browserify/-/path-browserify-1.0.1.tgz", @@ -11094,6 +11431,30 @@ "node": ">=0.10.0" } }, + "node_modules/react": { + "version": "19.2.4", + "resolved": "https://registry.npmjs.org/react/-/react-19.2.4.tgz", + "integrity": "sha512-9nfp2hYpCwOjAN+8TZFGhtWEwgvWHXqESH8qT89AT/lWklpLON22Lc8pEtnpsZz7VmawabSU0gCjnj8aC0euHQ==", + "license": "MIT", + "engines": { + "node": ">=0.10.0" + } + }, + "node_modules/react-reconciler": { + "version": "0.33.0", + "resolved": "https://registry.npmjs.org/react-reconciler/-/react-reconciler-0.33.0.tgz", + "integrity": "sha512-KetWRytFv1epdpJc3J4G75I4WrplZE5jOL7Yq0p34+OVOKF4Se7WrdIdVC45XsSSmUTlht2FM/fM1FZb1mfQeA==", + "license": "MIT", + "dependencies": { + "scheduler": "^0.27.0" + }, + "engines": { + "node": ">=0.10.0" + }, + "peerDependencies": { + "react": "^19.2.0" + } + }, "node_modules/read-package-up": { "version": "11.0.0", "resolved": "https://registry.npmjs.org/read-package-up/-/read-package-up-11.0.0.tgz", @@ -11640,6 +12001,12 @@ "integrity": "sha512-yqYn1JhPczigF94DMS+shiDMjDowYO6y9+wB/4WgO0Y19jWYk0lQ4tuG5KI7kj4FTp1wxPj5IFfcrz/s1c3jjQ==", "license": "BlueOak-1.0.0" }, + "node_modules/scheduler": { + "version": "0.27.0", + "resolved": "https://registry.npmjs.org/scheduler/-/scheduler-0.27.0.tgz", + "integrity": "sha512-eNv+WrVbKu1f3vbYJT/xtiF5syA5HPIMtf9IgY/nKg0sWqzAUEvqY/xm7OcZc/qafLx/iO9FgOmeSAp4v5ti/Q==", + "license": "MIT" + }, "node_modules/scslre": { "version": "0.3.0", "resolved": "https://registry.npmjs.org/scslre/-/scslre-0.3.0.tgz", @@ -12462,6 +12829,19 @@ "node": ">=8" } }, + "node_modules/shell-quote": { + "version": "1.8.3", + "resolved": "https://registry.npmjs.org/shell-quote/-/shell-quote-1.8.3.tgz", + "integrity": "sha512-ObmnIF4hXNg1BqhnHmgbDETF8dLPCggZWBjkQfhZpbszZnYur5DUljTcCHii5LC3J5E0yeO/1LIMyH+UvHQgyw==", + "dev": true, + "license": "MIT", + "engines": { + "node": ">= 0.4" + }, + "funding": { + "url": "https://github.com/sponsors/ljharb" + } + }, "node_modules/signal-exit": { "version": "4.1.0", "resolved": "https://registry.npmjs.org/signal-exit/-/signal-exit-4.1.0.tgz", @@ -12830,6 +13210,27 @@ "integrity": "sha512-D9cPgkvLlV3t3IzL0D0YLvGA9Ahk4PcvVwUbN0dSGr1aP0Nrt4AEnTUbuGvquEC0mA64Gqt1fzirlRs5ibXx8g==", "license": "BSD-3-Clause" }, + "node_modules/stack-utils": { + "version": "2.0.6", + "resolved": "https://registry.npmjs.org/stack-utils/-/stack-utils-2.0.6.tgz", + "integrity": "sha512-XlkWvfIm6RmsWtNJx+uqtKLS8eqFbxUg0ZzLXqY0caEy9l7hruX8IpiDnjsLavoBgqCCR71TqWO8MaXYheJ3RQ==", + "license": "MIT", + "dependencies": { + "escape-string-regexp": "^2.0.0" + }, + "engines": { + "node": ">=10" + } + }, + "node_modules/stack-utils/node_modules/escape-string-regexp": { + "version": "2.0.0", + "resolved": "https://registry.npmjs.org/escape-string-regexp/-/escape-string-regexp-2.0.0.tgz", + "integrity": "sha512-UpzcLCXolUWcNu5HtVMHYdXJjArjsF9C0aNnquZYY4uW/Vu0miy5YoWvbV345HauVvcAUnpRuhMMcqTcGOY2+w==", + "license": "MIT", + "engines": { + "node": ">=8" + } + }, "node_modules/stream-combiner2": { "version": "1.1.1", "resolved": "https://registry.npmjs.org/stream-combiner2/-/stream-combiner2-1.1.1.tgz", @@ -13277,32 +13678,6 @@ "node": ">=18" } }, - "node_modules/test-exclude/node_modules/brace-expansion": { - "version": "2.0.2", - "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-2.0.2.tgz", - "integrity": "sha512-Jt0vHyM+jmUBqojB7E1NIYadt0vI0Qxjxd2TErW94wDz+E2LAm5vKMXXwg6ZZBTHPuUlDgQHKXvjGBdfcF1ZDQ==", - "dev": true, - "license": "MIT", - "dependencies": { - "balanced-match": "^1.0.0" - } - }, - "node_modules/test-exclude/node_modules/minimatch": { - "version": "9.0.5", - "resolved": "https://registry.npmjs.org/minimatch/-/minimatch-9.0.5.tgz", - "integrity": "sha512-G6T0ZX48xgozx7587koeX9Ys2NYy6Gmv//P89sEte9V9whIapMNF4idKxnW2QtCcLiTWlb/wfCabAtAFWhhBow==", - "dev": true, - "license": "ISC", - "dependencies": { - "brace-expansion": "^2.0.1" - }, - "engines": { - "node": ">=16 || 14 >=14.17" - }, - "funding": { - "url": "https://github.com/sponsors/isaacs" - } - }, "node_modules/text-decoder": { "version": "1.2.3", "resolved": "https://registry.npmjs.org/text-decoder/-/text-decoder-1.2.3.tgz", @@ -13489,6 +13864,16 @@ "url": "https://github.com/sponsors/ljharb" } }, + "node_modules/tree-kill": { + "version": "1.2.2", + "resolved": "https://registry.npmjs.org/tree-kill/-/tree-kill-1.2.2.tgz", + "integrity": "sha512-L0Orpi8qGpRG//Nd+H90vFB+3iHnue1zSSGmNOOCh1GLJ7rUKVwV2HvijphGQS2UmhUZewS9VgvxYIdgr+fG1A==", + "dev": true, + "license": "MIT", + "bin": { + "tree-kill": "cli.js" + } + }, "node_modules/true-myth": { "version": "4.1.1", "resolved": "https://registry.npmjs.org/true-myth/-/true-myth-4.1.1.tgz", @@ -14265,6 +14650,71 @@ "node": ">= 8" } }, + "node_modules/widest-line": { + "version": "5.0.0", + "resolved": "https://registry.npmjs.org/widest-line/-/widest-line-5.0.0.tgz", + "integrity": "sha512-c9bZp7b5YtRj2wOe6dlj32MK+Bx/M/d+9VB2SHM1OtsUHR0aV0tdP6DWh/iMt0kWi1t5g1Iudu6hQRNd1A4PVA==", + "license": "MIT", + "dependencies": { + "string-width": "^7.0.0" + }, + "engines": { + "node": ">=18" + }, + "funding": { + "url": "https://github.com/sponsors/sindresorhus" + } + }, + "node_modules/widest-line/node_modules/ansi-regex": { + "version": "6.2.2", + "resolved": "https://registry.npmjs.org/ansi-regex/-/ansi-regex-6.2.2.tgz", + "integrity": "sha512-Bq3SmSpyFHaWjPk8If9yc6svM8c56dB5BAtW4Qbw5jHTwwXXcTLoRMkpDJp6VL0XzlWaCHTXrkFURMYmD0sLqg==", + "license": "MIT", + "engines": { + "node": ">=12" + }, + "funding": { + "url": "https://github.com/chalk/ansi-regex?sponsor=1" + } + }, + "node_modules/widest-line/node_modules/emoji-regex": { + "version": "10.6.0", + "resolved": "https://registry.npmjs.org/emoji-regex/-/emoji-regex-10.6.0.tgz", + "integrity": "sha512-toUI84YS5YmxW219erniWD0CIVOo46xGKColeNQRgOzDorgBi1v4D71/OFzgD9GO2UGKIv1C3Sp8DAn0+j5w7A==", + "license": "MIT" + }, + "node_modules/widest-line/node_modules/string-width": { + "version": "7.2.0", + "resolved": "https://registry.npmjs.org/string-width/-/string-width-7.2.0.tgz", + "integrity": "sha512-tsaTIkKW9b4N+AEj+SVA+WhJzV7/zMhcSu78mLKWSk7cXMOSHsBKFWUs0fWwq8QyK3MgJBQRX6Gbi4kYbdvGkQ==", + "license": "MIT", + "dependencies": { + "emoji-regex": "^10.3.0", + "get-east-asian-width": "^1.0.0", + "strip-ansi": "^7.1.0" + }, + "engines": { + "node": ">=18" + }, + "funding": { + "url": "https://github.com/sponsors/sindresorhus" + } + }, + "node_modules/widest-line/node_modules/strip-ansi": { + "version": "7.1.2", + "resolved": "https://registry.npmjs.org/strip-ansi/-/strip-ansi-7.1.2.tgz", + "integrity": "sha512-gmBGslpoQJtgnMAvOVqGZpEz9dyoKTCzy2nfz/n8aIFhN/jCE/rCmcxabB6jOOHV+0WNnylOxaxBQPSvcWklhA==", + "license": "MIT", + "dependencies": { + "ansi-regex": "^6.0.1" + }, + "engines": { + "node": ">=12" + }, + "funding": { + "url": "https://github.com/chalk/strip-ansi?sponsor=1" + } + }, "node_modules/with": { "version": "7.0.2", "resolved": "https://registry.npmjs.org/with/-/with-7.0.2.tgz", @@ -14570,6 +15020,12 @@ "url": "https://github.com/sponsors/sindresorhus" } }, + "node_modules/yoga-layout": { + "version": "3.2.1", + "resolved": "https://registry.npmjs.org/yoga-layout/-/yoga-layout-3.2.1.tgz", + "integrity": "sha512-0LPOt3AxKqMdFBZA3HBAt/t/8vIKq7VaQYbuA8WxCgung+p9TVyKRYdpvCb80HcdTN2NkbIKbhNwKUfm3tQywQ==", + "license": "MIT" + }, "node_modules/zod": { "version": "3.25.76", "resolved": "https://registry.npmjs.org/zod/-/zod-3.25.76.tgz", diff --git a/package.json b/package.json index 0eed2a76..bfc2604d 100644 --- a/package.json +++ b/package.json @@ -17,9 +17,15 @@ "test:all": "npm run test && npm run test:slow", "test:coverage": "c8 npm run test:unit", "test:coverage:report": "c8 --reporter=html npm run test:unit && echo 'Coverage report generated at coverage/index.html'", - "postinstall": "node scripts/fix-node-pty-permissions.js", + "postinstall": "node scripts/fix-node-pty-permissions.js && node scripts/install-tui-binary.js", "start": "node cli/index.js", + "build:tui-backend": "tsc --project tsconfig.tui-backend.json", + "build:tui": "npm run build:tui-backend", "typecheck": "tsc --noEmit", + "typecheck:tui-backend": "tsc --project tsconfig.tui-backend.json --noEmit", + "dev:link": "ZEROSHOT_TUI_BINARY_SKIP=1 npm link", + "dev:tui": "concurrently -k -n backend,tui \"npm run build:tui-backend -- --watch\" \"sh -c 'cd tui-rs && cargo watch -w . -x \\\"run -p zeroshot-tui -- --ui disruptive\\\"'\"", + "dev:bootstrap": "npm run dev:link && npm run dev:tui", "lint": "eslint .", "lint:fix": "eslint . --fix", "validate:templates": "node scripts/validate-templates.js", @@ -30,10 +36,10 @@ "deadcode:deps": "depcheck", "deadcode:all": "npm run deadcode && npm run deadcode:files && npm run deadcode:deps", "dupcheck": "jscpd src/ --min-lines 5 --min-tokens 50 --threshold 5", - "check": "npm run typecheck && npm run lint", + "check": "npm run typecheck && npm run typecheck:tui-backend && npm run lint", "check:all": "npm run check && npm run deadcode:all", "release": "semantic-release", - "prepublishOnly": "npm run lint && npm run typecheck", + "prepublishOnly": "npm run lint && npm run typecheck && npm run build:tui", "prepare": "husky" }, "c8": { @@ -94,18 +100,20 @@ "CHANGELOG.md" ], "dependencies": { - "ajv": "^8.17.1", + "ajv": "^8.18.0", "ansi-to-html": "^0.7.2", "better-sqlite3": "^12.5.0", "blessed": "^0.1.81", "blessed-contrib": "^4.11.0", "chalk": "^4.1.2", "commander": "^14.0.2", + "ink": "^6.6.0", "md-to-pdf": "^5.2.5", "node-pty": "^1.1.0", "omelette": "^0.4.17", "pidusage": "^4.0.1", - "proper-lockfile": "^4.1.2" + "proper-lockfile": "^4.1.2", + "react": "^19.2.4" }, "devDependencies": { "@semantic-release/changelog": "^6.0.3", @@ -113,8 +121,11 @@ "@semantic-release/github": "^11.0.6", "@types/better-sqlite3": "^7.6.13", "@types/node": "^25.0.3", + "@types/react": "^19.2.9", + "@types/react-reconciler": "^0.33.0", "c8": "^10.1.3", "chai": "^6.2.1", + "concurrently": "^9.1.0", "depcheck": "^1.4.7", "eslint": "^9.39.1", "eslint-config-prettier": "^10.1.8", @@ -135,8 +146,9 @@ "overrides": { "xml2js": "^0.5.0", "diff": ">=8.0.3", - "tar": ">=7.5.3", - "undici": ">=7.18.2" + "tar": ">=7.5.8", + "undici": ">=7.18.2", + "minimatch": ">=10.2.1" }, "lint-staged": { "*.{js,mjs,cjs}": [ diff --git a/scripts/fix-node-pty-permissions.js b/scripts/fix-node-pty-permissions.js index 9728f6fd..1b3083b6 100644 --- a/scripts/fix-node-pty-permissions.js +++ b/scripts/fix-node-pty-permissions.js @@ -30,7 +30,7 @@ let fixed = 0; let errors = 0; try { - const platforms = fs.readdirSync(prebuildsDir).filter(f => { + const platforms = fs.readdirSync(prebuildsDir).filter((f) => { try { return fs.statSync(path.join(prebuildsDir, f)).isDirectory(); } catch { diff --git a/scripts/install-tui-binary.js b/scripts/install-tui-binary.js new file mode 100644 index 00000000..7d6e63a7 --- /dev/null +++ b/scripts/install-tui-binary.js @@ -0,0 +1,367 @@ +'use strict'; + +const crypto = require('crypto'); +const { execFile } = require('child_process'); +const fs = require('fs'); +const http = require('http'); +const https = require('https'); +const os = require('os'); +const path = require('path'); +const { pipeline } = require('stream/promises'); +const { URL } = require('url'); +const { promisify } = require('util'); + +const { + ENV_BINARY_PATH, + ENV_BINARY_URL, + ENV_BINARY_SKIP, + getAssetName, + getInstallDir, + getInstalledBinaryPath, + resolveBinaryPathOverride, + resolveDownloadUrl, + resolveTarget, + shouldSkipBinaryInstall, +} = require('../lib/tui-binary'); + +const MAX_REDIRECTS = 5; +const execFileAsync = promisify(execFile); +const PACKAGE_ROOT = path.resolve(__dirname, '..'); + +function isGitCheckout() { + let current = PACKAGE_ROOT; + while (true) { + const gitPath = path.join(current, '.git'); + if (fs.existsSync(gitPath)) { + return true; + } + const parent = path.dirname(current); + if (parent === current) { + return false; + } + current = parent; + } +} + +function isNotFoundError(error) { + if (!error || typeof error.message !== 'string') { + return false; + } + return /\(404\)/.test(error.message); +} + +async function main() { + if (shouldSkipBinaryInstall()) { + console.log(`${ENV_BINARY_SKIP} set; skipping Rust TUI binary install.`); + return; + } + + const installContext = await prepareInstallContext(); + if (installContext.overridePath) { + console.log( + `${ENV_BINARY_PATH} set; using local Rust TUI binary at ${installContext.overridePath}.` + ); + await installFromLocalBinary(installContext.overridePath, installContext.installPath); + return; + } + + const downloadContext = resolveDownloadContext(); + if (!downloadContext) { + console.log( + `Rust TUI binary not supported on ${process.platform}/${process.arch}; skipping install.` + ); + return; + } + + await downloadAndInstall(downloadContext, installContext); +} + +async function prepareInstallContext() { + const installDir = getInstallDir(); + const installPath = getInstalledBinaryPath(); + await fs.promises.mkdir(installDir, { recursive: true }); + const overridePath = resolveBinaryPathOverride(); + return { installDir, installPath, overridePath }; +} + +function resolveDownloadContext() { + const overrideUrl = process.env[ENV_BINARY_URL]; + const target = overrideUrl ? null : resolveTarget(); + if (!overrideUrl && !target) { + return null; + } + + const archiveUrl = resolveDownloadUrl({ + platform: target?.platform, + arch: target?.arch, + overrideUrl, + }); + + if (!archiveUrl) { + throw new Error('Unable to resolve Rust TUI binary download URL.'); + } + + const assetName = target + ? getAssetName(target.platform, target.arch) + : resolveAssetNameFromUrl(archiveUrl); + + return { archiveUrl, assetName }; +} + +async function downloadAndInstall(downloadContext, installContext) { + const tempDir = await fs.promises.mkdtemp(path.join(os.tmpdir(), 'zeroshot-tui-')); + const archivePath = path.join(tempDir, downloadContext.assetName); + + try { + try { + await downloadToFile(downloadContext.archiveUrl, archivePath, MAX_REDIRECTS); + } catch (error) { + if (isGitCheckout() && isNotFoundError(error)) { + console.log( + 'Rust TUI binary not found for this version in git checkout; skipping install. ' + + `Set ${ENV_BINARY_URL} or ${ENV_BINARY_PATH} to override.` + ); + return; + } + throw error; + } + const shaUrl = `${downloadContext.archiveUrl}.sha256`; + const expectedSha = await fetchSha256(shaUrl); + if (expectedSha) { + await verifySha256(archivePath, expectedSha); + } + + await extractArchive(archivePath, installContext.installDir); + await ensureBinaryInstalled(installContext.installPath, installContext.installDir); + + await fs.promises.chmod(installContext.installPath, 0o755); + console.log(`Installed Rust TUI binary to ${installContext.installPath}`); + } finally { + await fs.promises.rm(tempDir, { recursive: true, force: true }); + } +} + +function resolveAssetNameFromUrl(url) { + try { + const parsed = new URL(url); + const base = path.basename(parsed.pathname); + return base || 'zeroshot-tui.tar.gz'; + } catch { + return 'zeroshot-tui.tar.gz'; + } +} + +async function installFromLocalBinary(sourcePath, destinationPath) { + const stats = await fs.promises.stat(sourcePath).catch(() => null); + if (!stats || !stats.isFile()) { + throw new Error(`Rust TUI binary not found at ${sourcePath}`); + } + await fs.promises.copyFile(sourcePath, destinationPath); + await fs.promises.chmod(destinationPath, 0o755); + console.log(`Installed Rust TUI binary from ${sourcePath}`); +} + +async function downloadToFile(url, destination, redirectsRemaining) { + const client = getHttpClient(url); + await new Promise((resolve, reject) => { + const request = client.get( + url, + { headers: { 'User-Agent': '@covibes/zeroshot' } }, + (response) => { + handleDownloadResponse({ + response, + url, + destination, + redirectsRemaining, + }) + .then(resolve) + .catch(reject); + } + ); + + request.on('error', reject); + }); +} + +function getHttpClient(url) { + return url.startsWith('https://') ? https : http; +} + +function isRedirectStatus(status) { + return [301, 302, 303, 307, 308].includes(status); +} + +function resolveRedirectUrl(baseUrl, location) { + try { + return new URL(location, baseUrl).toString(); + } catch { + return location; + } +} + +async function handleDownloadResponse({ response, url, destination, redirectsRemaining }) { + const status = response.statusCode || 0; + if (isRedirectStatus(status)) { + await followRedirect(response, url, destination, redirectsRemaining); + return; + } + + if (status !== 200) { + response.resume(); + throw new Error(`Download failed (${status}) for ${url}`); + } + + await streamToFileWithLengthCheck(response, destination, url); +} + +async function followRedirect(response, url, destination, redirectsRemaining) { + if (redirectsRemaining <= 0) { + response.resume(); + throw new Error(`Too many redirects while downloading ${url}`); + } + const location = response.headers.location; + if (!location) { + response.resume(); + throw new Error(`Redirect without location while downloading ${url}`); + } + const nextUrl = resolveRedirectUrl(url, location); + response.resume(); + await downloadToFile(nextUrl, destination, redirectsRemaining - 1); +} + +async function streamToFileWithLengthCheck(response, destination, url) { + const expectedLength = parseInt(response.headers['content-length'], 10); + let downloaded = 0; + const fileStream = fs.createWriteStream(destination); + + response.on('data', (chunk) => { + downloaded += chunk.length; + }); + + await pipeline(response, fileStream); + + if (!Number.isNaN(expectedLength) && expectedLength > 0 && downloaded !== expectedLength) { + throw new Error( + `Download size mismatch for ${url}: expected ${expectedLength}, got ${downloaded}` + ); + } +} + +function fetchSha256(url, redirectsRemaining = MAX_REDIRECTS) { + const client = getHttpClient(url); + + return new Promise((resolve, reject) => { + const request = client.get( + url, + { headers: { 'User-Agent': '@covibes/zeroshot' } }, + (response) => { + const status = response.statusCode || 0; + if (isRedirectStatus(status)) { + if (redirectsRemaining <= 0) { + response.resume(); + reject(new Error(`Too many redirects while downloading ${url}`)); + return; + } + const location = response.headers.location; + if (!location) { + response.resume(); + reject(new Error(`Redirect without location while downloading ${url}`)); + return; + } + const nextUrl = resolveRedirectUrl(url, location); + response.resume(); + resolve(fetchSha256(nextUrl, redirectsRemaining - 1)); + return; + } + if (status === 404) { + response.resume(); + resolve(null); + return; + } + if (status !== 200) { + response.resume(); + reject(new Error(`Checksum download failed (${status}) for ${url}`)); + return; + } + let body = ''; + response.setEncoding('utf8'); + response.on('data', (chunk) => { + body += chunk; + }); + response.on('end', () => { + const match = body.trim().match(/^([a-fA-F0-9]{64})/); + if (!match) { + reject(new Error(`Invalid sha256 contents from ${url}`)); + return; + } + resolve(match[1].toLowerCase()); + }); + } + ); + + request.on('error', reject); + }); +} + +async function verifySha256(filePath, expected) { + const hash = crypto.createHash('sha256'); + const stream = fs.createReadStream(filePath); + await new Promise((resolve, reject) => { + stream.on('error', reject); + stream.on('data', (chunk) => hash.update(chunk)); + stream.on('end', resolve); + }); + const actual = hash.digest('hex'); + if (actual !== expected) { + throw new Error(`Checksum mismatch for ${filePath}: expected ${expected}, got ${actual}`); + } +} + +async function extractArchive(archivePath, destinationDir) { + const entries = await listArchiveEntries(archivePath); + const unsafeEntry = entries.find((entry) => !isSafeTarEntry(entry)); + if (unsafeEntry) { + throw new Error(`Unsafe archive entry detected: ${unsafeEntry}`); + } + await execFileAsync('tar', ['-xzf', archivePath, '-C', destinationDir]); +} + +async function listArchiveEntries(archivePath) { + const { stdout } = await execFileAsync('tar', ['-tf', archivePath]); + return stdout + .split('\n') + .map((line) => line.trim()) + .filter(Boolean); +} + +function isSafeTarEntry(entry) { + if (!entry || entry.includes('\0')) { + return false; + } + const normalized = path.posix.normalize(entry); + if (path.posix.isAbsolute(normalized)) { + return false; + } + if (normalized === '..' || normalized.startsWith('../')) { + return false; + } + return true; +} + +async function ensureBinaryInstalled(installPath, installDir) { + if (fs.existsSync(installPath)) { + return; + } + const defaultBinary = path.join(installDir, 'zeroshot-tui'); + if (fs.existsSync(defaultBinary)) { + await fs.promises.rename(defaultBinary, installPath); + return; + } + throw new Error(`Expected Rust TUI binary at ${installPath} after extraction`); +} + +main().catch((error) => { + console.error('Failed to install Rust TUI binary'); + console.error(error.stack || error.message); + process.exit(1); +}); diff --git a/scripts/validate-templates.js b/scripts/validate-templates.js index 5accc1b3..d47ce305 100755 --- a/scripts/validate-templates.js +++ b/scripts/validate-templates.js @@ -7,86 +7,52 @@ * Exit codes: 0 = all valid, 1 = validation errors found */ -const fs = require('fs'); const path = require('path'); -const { validateConfig } = require('../src/config-validator'); +const { validateTemplates } = require('../src/template-validation'); const TEMPLATES_DIR = path.join(__dirname, '../cluster-templates'); -function findJsonFiles(dir) { - const files = []; - if (!fs.existsSync(dir)) return files; - - for (const entry of fs.readdirSync(dir, { withFileTypes: true })) { - const fullPath = path.join(dir, entry.name); - if (entry.isDirectory()) { - files.push(...findJsonFiles(fullPath)); - } else if (entry.name.endsWith('.json')) { - files.push(fullPath); - } - } - return files; +function parseArgs(argv) { + const deep = + argv.includes('--deep') || + argv.includes('--sim=deep') || + process.env.ZEROSHOT_TEMPLATE_SIM === 'deep'; + return { deep }; } -function validateTemplate(filePath) { - const relativePath = path.relative(process.cwd(), filePath); +async function main() { + console.log('Validating cluster templates...\n'); - try { - const content = fs.readFileSync(filePath, 'utf-8'); - const config = JSON.parse(content); + const { deep } = parseArgs(process.argv.slice(2)); + const report = await validateTemplates({ templatesDir: TEMPLATES_DIR, deep }); - // Skip non-cluster configs (like package.json) - if (!config.agents && !config.name) { - return { valid: true, skipped: true }; - } + let hasErrors = false; - const result = validateConfig(config); + for (const { filePath, result } of report.results) { + const relativePath = path.relative(process.cwd(), filePath); if (!result.valid) { + hasErrors = true; console.error(`\nāŒ ${relativePath}`); for (const error of result.errors) { console.error(` ERROR: ${error}`); } - } else if (result.warnings.length > 0) { + continue; + } + + if (result.warnings.length > 0) { console.warn(`\nāš ļø ${relativePath}`); for (const warning of result.warnings) { console.warn(` WARN: ${warning}`); } - } else { - console.log(`āœ“ ${relativePath}`); + continue; } - return result; - } catch (err) { - console.error(`\nāŒ ${relativePath}`); - console.error(` PARSE ERROR: ${err.message}`); - return { valid: false, errors: [err.message], warnings: [] }; - } -} - -function main() { - console.log('Validating cluster templates...\n'); - - const templateFiles = [...findJsonFiles(TEMPLATES_DIR)]; - - let hasErrors = false; - let validated = 0; - let skipped = 0; - - for (const file of templateFiles) { - const result = validateTemplate(file); - if (result.skipped) { - skipped++; - } else { - validated++; - if (!result.valid) { - hasErrors = true; - } - } + console.log(`āœ“ ${relativePath}`); } console.log(`\n${'='.repeat(60)}`); - console.log(`Validated: ${validated} templates, Skipped: ${skipped} files`); + console.log(`Validated: ${report.validated} templates, Skipped: ${report.skipped} files`); if (hasErrors) { console.error('\nāŒ VALIDATION FAILED - Fix errors above before merging\n'); @@ -97,4 +63,7 @@ function main() { } } -main(); +main().catch((err) => { + console.error(`\nāŒ Template validation crashed: ${err.message}\n`); + process.exit(1); +}); diff --git a/src/agent-wrapper.js b/src/agent-wrapper.js index edef2765..7664b26f 100644 --- a/src/agent-wrapper.js +++ b/src/agent-wrapper.js @@ -12,12 +12,14 @@ const LogicEngine = require('./logic-engine'); const { validateAgentConfig } = require('./agent/agent-config'); -const { loadSettings, validateModelAgainstMax } = require('../lib/settings'); +const { loadSettings, validateModelAgainstMax, VALID_MODELS } = require('../lib/settings'); const { normalizeProviderName } = require('../lib/provider-names'); const { getProvider } = require('./providers'); const { buildContext } = require('./agent/agent-context-builder'); +const { collectQueuedGuidance } = require('./agent/guidance-queue'); const { findMatchingTrigger, evaluateTrigger } = require('./agent/agent-trigger-evaluator'); const { executeHook } = require('./agent/agent-hook-executor'); +const { injectInput: injectAgentInput } = require('./agent/agent-input-injector'); const { spawnClaudeTask, followClaudeTaskLogs, @@ -73,6 +75,8 @@ class AgentWrapper { this.lastTaskEndTime = null; // Track when last task completed (for context filtering) /** @type {number | null} */ this.lastAgentStartTime = null; // Track when agent last began executing (for context filtering) + /** @type {number | null} */ + this.lastGuidanceAppliedAt = null; // Track last queued guidance applied to prompt // LIVENESS DETECTION - Track output freshness to detect stuck agents /** @type {number | null} */ @@ -246,17 +250,17 @@ class AgentWrapper { /** * Select model based on current iteration and agent config - * Enforces legacy maxModel/minModel for Claude's haiku/sonnet/opus + * Enforces legacy maxModel/minModel aliases for Claude compatibility * @returns {string|null} * @private */ _selectModel() { const spec = this._resolveModelSpec(); const settings = loadSettings(); - const maxModel = settings.maxModel || 'sonnet'; + const maxModel = settings.maxModel; const minModel = settings.minModel || null; - if (spec.model && ['opus', 'sonnet', 'haiku'].includes(spec.model)) { + if (spec.model && maxModel && VALID_MODELS.includes(spec.model)) { return validateModelAgainstMax(spec.model, maxModel, minModel); } @@ -407,6 +411,12 @@ class AgentWrapper { */ _buildContext(triggeringMessage) { const previousAgentStart = this.lastAgentStartTime; + const queuedGuidance = collectQueuedGuidance({ + messageBus: this.messageBus, + clusterId: this.cluster.id, + agentId: this.id, + lastDeliveredAt: this.lastGuidanceAppliedAt, + }); const context = buildContext({ id: this.id, role: this.role, @@ -418,13 +428,22 @@ class AgentWrapper { lastAgentStartTime: previousAgentStart, triggeringMessage, selectedPrompt: this._selectPrompt(), + queuedGuidance: queuedGuidance.guidanceBlock, // Pass isolation state for conditional git restriction worktree: this.worktree, isolation: this.isolation, }); - // Record when this iteration started so future "since: last_agent_start" filters work - this.lastAgentStartTime = Date.now(); + // Record when this iteration started so future "since: last_agent_start" filters work. + const latestMessage = this.messageBus.findLast({ cluster_id: this.cluster.id }); + const latestTimestamp = latestMessage?.timestamp; + const now = Date.now(); + this.lastAgentStartTime = + typeof latestTimestamp === 'number' ? Math.max(now, latestTimestamp + 1) : now; + + if (queuedGuidance.latestTimestamp !== null) { + this.lastGuidanceAppliedAt = queuedGuidance.latestTimestamp; + } return context; } @@ -538,6 +557,16 @@ class AgentWrapper { await this._executeTask(triggeringMessage); } + /** + * Inject live input into a running agent task when possible + * @param {string} text + * @param {object} [options] + * @returns {Promise<{status: string, reason?: string|null, method?: string|null, taskId?: string|null}>} + */ + injectInput(text, options = {}) { + return injectAgentInput(this, text, options); + } + /** * Get current agent state */ diff --git a/src/agent/agent-config.js b/src/agent/agent-config.js index a2c150d5..489b1ed8 100644 --- a/src/agent/agent-config.js +++ b/src/agent/agent-config.js @@ -33,6 +33,13 @@ function applyOutputDefaults(config) { config.outputFormat = 'json'; } + // Map structuredOutput ā†’ jsonSchema (structuredOutput is the user-facing API name, + // jsonSchema is the internal key used by task executor and CLI args). + // Explicit jsonSchema takes precedence over structuredOutput. + if (config.structuredOutput && !config.jsonSchema) { + config.jsonSchema = config.structuredOutput; + } + // If outputFormat is json but no schema defined, use a minimal default schema if (config.outputFormat === 'json' && !config.jsonSchema) { config.jsonSchema = { @@ -205,7 +212,7 @@ function validateAgentConfig(config, options = {}) { // COST CEILING/FLOOR ENFORCEMENT: Validate model(s) against maxModel and minModel at config time // Catches violations EARLY (config load) instead of at runtime (iteration N) const settings = loadSettings(); - const maxModel = settings.maxModel || 'sonnet'; + const maxModel = settings.maxModel; const minModel = settings.minModel || null; // STRICT SCHEMA PROPAGATION: Issue #52 fix diff --git a/src/agent/agent-context-builder.js b/src/agent/agent-context-builder.js index 2be362f7..0be8b60a 100644 --- a/src/agent/agent-context-builder.js +++ b/src/agent/agent-context-builder.js @@ -5,13 +5,20 @@ * - Context assembly from multiple message sources * - Context strategy evaluation (topics, limits, since timestamps) * - Prompt injection and formatting - * - Token-based truncation + * - Token-budgeted context packs * - Defensive context overflow prevention */ // Defensive limit: 500,000 chars ā‰ˆ 125k tokens (safe buffer below 200k limit) // Prevents "Prompt is too long" errors that kill tasks const MAX_CONTEXT_CHARS = 500000; +const { + buildContextMetrics, + emitContextMetrics, + resolveLegacyMaxTokens, + updateTotalMetrics, +} = require('./context-metrics'); +const { buildContextPacks } = require('./context-pack-builder'); /** * Generate an example object from a JSON schema @@ -169,14 +176,16 @@ function resolveSourceSince(source, cluster, lastTaskEndTime, lastAgentStartTime return lastTaskEndTime || cluster.createdAt; } if (sinceValue === 'last_agent_start') { - return lastAgentStartTime || cluster.createdAt; + // Use strict "after" semantics to avoid timestamp collisions in the same millisecond + // (prevents stale context from leaking across agent restarts). + return lastAgentStartTime ? lastAgentStartTime + 1 : cluster.createdAt; } if (typeof sinceValue === 'string') { const parsed = Date.parse(sinceValue); if (Number.isNaN(parsed)) { throw new Error( - `Agent context source for topic ${source.topic} has invalid since value "${sinceValue}". ` + + `Unknown context source "since" value "${sinceValue}" for topic ${source.topic}. ` + 'Use cluster_start, last_task_end, last_agent_start, or an ISO timestamp.' ); } @@ -201,38 +210,108 @@ function formatSourceMessagesSection(source, messages) { return context; } -function buildSourcesSection({ - strategy, +function resolveSourceSelection(source, { compact = false } = {}) { + const baseAmount = source.amount ?? source.limit; + const baseStrategy = source.strategy ?? (baseAmount !== undefined ? 'latest' : 'all'); + + if (!compact) { + return { amount: baseAmount, strategy: baseStrategy }; + } + + const compactAmount = source.compactAmount ?? (baseAmount !== undefined ? 1 : 1); + const compactStrategy = + source.compactStrategy ?? (baseStrategy === 'all' ? 'latest' : baseStrategy); + + return { amount: compactAmount, strategy: compactStrategy }; +} + +function resolveSourceMessages({ + source, messageBus, cluster, lastTaskEndTime, lastAgentStartTime, + compact = false, }) { - let context = ''; - for (const source of strategy.sources) { - const sinceTimestamp = resolveSourceSince(source, cluster, lastTaskEndTime, lastAgentStartTime); - const messages = messageBus.query({ - cluster_id: cluster.id, - topic: source.topic, - sender: source.sender, - since: sinceTimestamp, - limit: source.limit, - }); + const sinceTimestamp = resolveSourceSince(source, cluster, lastTaskEndTime, lastAgentStartTime); + const { amount, strategy } = resolveSourceSelection(source, { compact }); + const order = strategy === 'latest' ? 'desc' : 'asc'; + const messages = messageBus.query({ + cluster_id: cluster.id, + topic: source.topic, + sender: source.sender, + since: sinceTimestamp, + limit: amount, + order, + }); - if (messages.length > 0) { - context += formatSourceMessagesSection(source, messages); - } + if (strategy !== 'latest' || messages.length <= 1) { + return messages; } - return context; + + return messages.slice().reverse(); } -function collectCannotValidateCriteria(prevValidations) { +function resolveSourcePriority(source) { + if (source.priority) { + return source.priority; + } + if (source.topic === 'STATE_SNAPSHOT') { + return 'required'; + } + if (source.topic === 'ISSUE_OPENED' || source.topic === 'PLAN_READY') { + return 'required'; + } + if (source.topic === 'VALIDATION_RESULT' || source.topic === 'IMPLEMENTATION_READY') { + return 'high'; + } + return 'medium'; +} + +function buildSourcePack({ + source, + index, + messageBus, + cluster, + lastTaskEndTime, + lastAgentStartTime, +}) { + const packId = `source:${source.topic}:${index}`; + const priority = resolveSourcePriority(source); + + const render = (compact) => { + const messages = resolveSourceMessages({ + source, + messageBus, + cluster, + lastTaskEndTime, + lastAgentStartTime, + compact, + }); + if (messages.length === 0) return ''; + return formatSourceMessagesSection(source, messages); + }; + + return { + id: packId, + section: 'sources', + priority, + render: () => render(false), + compact: () => render(true), + }; +} + +const { isPlatformMismatchReason } = require('./validation-platform'); + +function collectCannotValidateCriteria(prevValidations, options = {}) { const cannotValidateCriteria = []; + const ignoreReason = options.ignoreReason; for (const msg of prevValidations) { const criteriaResults = msg.content?.data?.criteriaResults; if (!Array.isArray(criteriaResults)) continue; for (const cr of criteriaResults) { if (cr.status !== 'CANNOT_VALIDATE' || !cr.id) continue; + if (ignoreReason && ignoreReason(cr.reason)) continue; if (cannotValidateCriteria.find((c) => c.id === cr.id)) continue; cannotValidateCriteria.push({ id: cr.id, @@ -257,7 +336,7 @@ function buildCannotValidateSection(cannotValidateCriteria) { return context; } -function buildValidatorSkipSection({ role, messageBus, cluster }) { +function buildValidatorSkipSection({ role, messageBus, cluster, isolation }) { if (role !== 'validator') return ''; const prevValidations = messageBus.query({ @@ -267,7 +346,8 @@ function buildValidatorSkipSection({ role, messageBus, cluster }) { limit: 50, }); - const cannotValidateCriteria = collectCannotValidateCriteria(prevValidations); + const ignoreReason = isolation?.enabled ? isPlatformMismatchReason : null; + const cannotValidateCriteria = collectCannotValidateCriteria(prevValidations, { ignoreReason }); return buildCannotValidateSection(cannotValidateCriteria); } @@ -281,110 +361,6 @@ function buildTriggeringMessageSection(triggeringMessage) { return context; } -function findContextSectionIndices(lines) { - let issueOpenedStart = -1; - let issueOpenedEnd = -1; - let triggeringStart = -1; - - for (let i = 0; i < lines.length; i++) { - if (lines[i].includes('## Messages from topic: ISSUE_OPENED')) { - issueOpenedStart = i; - } - if (issueOpenedStart !== -1 && issueOpenedEnd === -1 && lines[i].startsWith('## ')) { - issueOpenedEnd = i; - } - if (lines[i].includes('## Triggering Message')) { - triggeringStart = i; - break; - } - } - - return { issueOpenedStart, issueOpenedEnd, triggeringStart }; -} - -function collectRecentLines(middleLines, budgetForRecent) { - const recentLines = []; - let recentSize = 0; - - for (let i = middleLines.length - 1; i >= 0; i--) { - const line = middleLines[i]; - const lineSize = line.length + 1; - - if (recentSize + lineSize > budgetForRecent) { - break; - } - - recentLines.unshift(line); - recentSize += lineSize; - } - - return recentLines; -} - -function truncateContextIfNeeded(context) { - const originalLength = context.length; - if (originalLength <= MAX_CONTEXT_CHARS) { - return context; - } - - console.log( - `[Context] Context too large (${originalLength} chars), truncating to prevent overflow...` - ); - - const lines = context.split('\n'); - const { issueOpenedStart, issueOpenedEnd, triggeringStart } = findContextSectionIndices(lines); - - const headerEnd = issueOpenedStart !== -1 ? issueOpenedStart : triggeringStart; - const header = lines.slice(0, headerEnd).join('\n'); - - const issueOpened = - issueOpenedStart !== -1 && issueOpenedEnd !== -1 - ? lines.slice(issueOpenedStart, issueOpenedEnd).join('\n') - : ''; - - const triggeringMsg = lines.slice(triggeringStart).join('\n'); - - const fixedSize = header.length + issueOpened.length + triggeringMsg.length; - const budgetForRecent = MAX_CONTEXT_CHARS - fixedSize - 200; - - const middleStart = issueOpenedEnd !== -1 ? issueOpenedEnd : headerEnd; - const middleEnd = triggeringStart; - const middleLines = lines.slice(middleStart, middleEnd); - const recentLines = collectRecentLines(middleLines, budgetForRecent); - - const parts = [header]; - if (issueOpened) { - parts.push(issueOpened); - } - if (recentLines.length < middleLines.length) { - const truncatedCount = middleLines.length - recentLines.length; - parts.push( - `\n[...${truncatedCount} earlier context messages truncated to prevent overflow...]\n` - ); - } - if (recentLines.length > 0) { - parts.push(recentLines.join('\n')); - } - parts.push(triggeringMsg); - - const truncatedContext = parts.join('\n'); - const truncatedLength = truncatedContext.length; - console.log( - `[Context] Truncated from ${originalLength} to ${truncatedLength} chars (${Math.round((truncatedLength / originalLength) * 100)}% retained)` - ); - - return truncatedContext; -} - -function applyLegacyMaxTokens(context, strategy) { - const maxTokens = strategy.maxTokens || 100000; - const maxChars = maxTokens * 4; - if (context.length > maxChars) { - return context.slice(0, maxChars) + '\n\n[Context truncated...]'; - } - return context; -} - /** * Build execution context for an agent * @param {object} params - Context building parameters @@ -413,30 +389,84 @@ function buildContext({ lastAgentStartTime, triggeringMessage, selectedPrompt, + queuedGuidance, worktree, isolation, }) { const strategy = config.contextStrategy || { sources: [] }; const isIsolated = !!(worktree?.enabled || isolation?.enabled); - let context = buildHeaderContext({ id, role, iteration, isIsolated }); - context += buildInstructionsSection({ config, selectedPrompt, id }); - context += buildLegacyOutputSchemaSection(config); - context += buildJsonSchemaSection(config); - context += buildSourcesSection({ + const header = buildHeaderContext({ id, role, iteration, isIsolated }); + const instructions = buildInstructionsSection({ config, selectedPrompt, id }); + const legacyOutputSchema = buildLegacyOutputSchemaSection(config); + const queuedGuidanceSection = queuedGuidance || ''; + const jsonSchema = buildJsonSchemaSection(config); + const validatorSkip = buildValidatorSkipSection({ role, messageBus, cluster, isolation }); + const triggeringMessageSection = buildTriggeringMessageSection(triggeringMessage); + + const packs = []; + let order = 0; + + const pushStaticPack = (packId, section, text, options = {}) => { + if (!text) return; + packs.push({ + id: packId, + section, + priority: 'required', + order: order++, + preserve: options.preserve || false, + render: () => text, + }); + }; + + pushStaticPack('header', 'header', header); + pushStaticPack('instructions', 'instructions', instructions); + pushStaticPack('queuedGuidance', 'queuedGuidance', queuedGuidanceSection); + pushStaticPack('legacyOutputSchema', 'legacyOutputSchema', legacyOutputSchema); + pushStaticPack('jsonSchema', 'jsonSchema', jsonSchema); + + if (Array.isArray(strategy.sources)) { + strategy.sources.forEach((source, index) => { + const pack = buildSourcePack({ + source, + index, + messageBus, + cluster, + lastTaskEndTime, + lastAgentStartTime, + }); + packs.push({ ...pack, order: order++ }); + }); + } + + pushStaticPack('validatorSkip', 'validatorSkip', validatorSkip); + pushStaticPack('triggeringMessage', 'triggeringMessage', triggeringMessageSection, { + preserve: true, + }); + + const maxTokens = resolveLegacyMaxTokens(strategy); + const packResult = buildContextPacks({ + packs, + maxTokens, + maxChars: MAX_CONTEXT_CHARS, + }); + + const metrics = buildContextMetrics({ + clusterId: cluster.id, + agentId: id, + role, + iteration, + triggeringMessage, strategy, - messageBus, - cluster, - lastTaskEndTime, - lastAgentStartTime, + packs: packResult.packDecisions, + budget: packResult.budget, + truncation: packResult.truncation, }); - context += buildValidatorSkipSection({ role, messageBus, cluster }); - context += buildTriggeringMessageSection(triggeringMessage); - context = truncateContextIfNeeded(context); - context = applyLegacyMaxTokens(context, strategy); + updateTotalMetrics(metrics, packResult.context.length); + emitContextMetrics(metrics, { messageBus, clusterId: cluster.id, agentId: id }); - return context; + return packResult.context; } module.exports = { diff --git a/src/agent/agent-hook-executor.js b/src/agent/agent-hook-executor.js index 1cc4c7be..917906db 100644 --- a/src/agent/agent-hook-executor.js +++ b/src/agent/agent-hook-executor.js @@ -139,6 +139,127 @@ async function executeHook(params) { throw new Error('execute_system_command not implemented'); } + if (hook.action === 'verify_github_pr') { + const { extractJsonFromOutput } = require('./output-extraction'); + const structuredOutput = extractJsonFromOutput(result.output) || {}; + const claimedPrUrl = structuredOutput.pr_url || null; + const claimedPrNumber = structuredOutput.pr_number || null; + + // Skip actual gh CLI verification if explicitly disabled (for integration tests) + // Unit tests mock execSync, so they still test the verification logic + if (process.env.ZEROSHOT_SKIP_GH_VERIFY === '1') { + agent._log(`āœ… VERIFICATION SKIPPED (ZEROSHOT_SKIP_GH_VERIFY=1)`); + agent._publish({ + topic: 'CLUSTER_COMPLETE', + content: { + data: { + reason: 'git-pusher-complete-verified', + pr_number: claimedPrNumber, + pr_url: claimedPrUrl, + }, + }, + }); + return; + } + + // Use explicit PR number when available (deterministic). + // Branch-based resolution can fail after merge when branch is deleted/transitional. + const ghPrViewCmd = claimedPrNumber + ? `gh pr view ${claimedPrNumber} --json state,mergedAt,url,number` + : `gh pr view --json state,mergedAt,url,number`; + + // GitHub API is eventually consistent after `gh pr merge`. + // Merge state can take 5-30s to propagate. Poll with backoff before concluding agent lied. + // POSTMORTEM 2026-02-11: 3s retry window killed cluster gentle-hydra-56 ā€” PR was actually merged. + const MERGE_POLL_ATTEMPTS = 6; + const MERGE_POLL_INTERVAL_MS = 5000; + + let prData; + try { + prData = JSON.parse( + execSync(ghPrViewCmd, { + encoding: 'utf8', + cwd: agent.workingDirectory, + stdio: ['pipe', 'pipe', 'pipe'], + }) + ); + } catch (err) { + if ( + err.message.includes('Could not resolve to a PullRequest') || + err.message.toLowerCase().includes('no pull requests found') + ) { + throw new Error( + `VERIFICATION FAILED: Agent claimed a PR exists for this branch, ` + + `but GitHub says it DOES NOT EXIST. Agent HALLUCINATED.` + ); + } + throw err; + } + + if (claimedPrUrl && prData.url && claimedPrUrl !== prData.url) { + throw new Error( + `VERIFICATION FAILED: Agent claimed PR URL ${claimedPrUrl}, but GitHub CLI reports ${prData.url}.` + ); + } + + // Poll for merge propagation if not yet showing as merged + if (!prData.mergedAt) { + const prNumber = prData.number; + const pollCmd = `gh pr view ${prNumber} --json state,mergedAt,url,number`; + agent._log( + `ā³ PR #${prNumber} not yet showing as merged (state="${prData.state}"). ` + + `Polling for GitHub API propagation (up to ${MERGE_POLL_ATTEMPTS} attempts, ${MERGE_POLL_INTERVAL_MS / 1000}s apart)...` + ); + + for (let attempt = 1; attempt <= MERGE_POLL_ATTEMPTS; attempt++) { + await new Promise((resolve) => setTimeout(resolve, MERGE_POLL_INTERVAL_MS)); + try { + prData = JSON.parse( + execSync(pollCmd, { + encoding: 'utf8', + cwd: agent.workingDirectory, + stdio: ['pipe', 'pipe', 'pipe'], + }) + ); + } catch { + // gh CLI error during poll ā€” keep trying + continue; + } + if (prData.mergedAt) { + agent._log(`āœ… PR #${prNumber} merge confirmed on poll attempt ${attempt}`); + break; + } + agent._log( + `ā³ Poll ${attempt}/${MERGE_POLL_ATTEMPTS}: PR #${prNumber} still state="${prData.state}"` + ); + } + } + + if (!prData.mergedAt) { + throw new Error( + `VERIFICATION FAILED: Agent claimed PR is merged, ` + + `but GitHub says state="${prData.state}" after ${MERGE_POLL_ATTEMPTS} polls ` + + `over ${(MERGE_POLL_ATTEMPTS * MERGE_POLL_INTERVAL_MS) / 1000}s. Agent LIED.` + ); + } + + agent._log(`āœ… VERIFICATION PASSED: PR #${prData.number} actually merged`); + + // Publish CLUSTER_COMPLETE only after verification passes + agent._publish({ + topic: 'CLUSTER_COMPLETE', + content: { + data: { + reason: 'git-pusher-complete-verified', + pr_number: prData.number, + pr_url: prData.url, + }, + }, + }); + + return; + } + throw new Error(`Unknown hook action: ${hook.action}`); } @@ -214,15 +335,71 @@ async function parseTransformResultData({ context, agent, script, scriptUsesResu } function buildTransformSandbox({ resultData, context, agent }) { + const clusterId = agent.cluster?.id || context.cluster?.id || agent.cluster_id || 'unknown'; + const messageBus = agent.messageBus; + const cluster = context.cluster || agent.cluster || null; + + // Ledger API wrapper (auto-scoped to cluster) - mirrors logic-engine.js + const ledgerAPI = messageBus + ? { + query: (criteria) => { + return messageBus.query({ ...criteria, cluster_id: clusterId }); + }, + findLast: (criteria) => { + return messageBus.findLast({ ...criteria, cluster_id: clusterId }); + }, + count: (criteria) => { + return messageBus.count({ ...criteria, cluster_id: clusterId }); + }, + since: (timestamp) => { + return messageBus.since({ cluster_id: clusterId, timestamp }); + }, + } + : null; + + // Cluster API wrapper - mirrors logic-engine.js + const clusterAPI = { + id: clusterId, + getAgents: () => { + return cluster ? cluster.agents || [] : []; + }, + getAgentsByRole: (role) => { + return cluster ? (cluster.agents || []).filter((a) => a.role === role) : []; + }, + getAgent: (id) => { + return cluster ? (cluster.agents || []).find((a) => a.id === id) : null; + }, + }; + + // Helper functions - mirrors logic-engine.js const helpers = { getConfig: require('../config-router').getConfig, + allResponded: (agents, topic, since) => { + if (!ledgerAPI) return false; + const responses = ledgerAPI.query({ topic, since }); + const responders = new Set(responses.map((r) => r.sender)); + return agents.every((a) => responders.has(a.id || a)); + }, + hasConsensus: (topic, since) => { + if (!ledgerAPI) return false; + const responses = ledgerAPI.query({ topic, since }); + if (responses.length === 0) return false; + return responses.every((r) => r.content?.data?.approved === true); + }, }; return { result: resultData, triggeringMessage: context.triggeringMessage, + ledger: ledgerAPI, + cluster: clusterAPI, helpers, + // Safe built-ins JSON, + Set, + Map, + Array, + Object, console: { log: (...args) => agent._log('[transform]', ...args), error: (...args) => console.error('[transform]', ...args), @@ -530,6 +707,59 @@ function evaluateHookLogic(params) { throw new Error(`Unsupported hook logic engine: ${logic.engine}`); } + const clusterId = agent.cluster?.id || context.cluster?.id || agent.cluster_id || 'unknown'; + const messageBus = agent.messageBus; + const cluster = context.cluster || agent.cluster || null; + + // Ledger API wrapper (auto-scoped to cluster) - mirrors logic-engine.js + const ledgerAPI = messageBus + ? { + query: (criteria) => { + return messageBus.query({ ...criteria, cluster_id: clusterId }); + }, + findLast: (criteria) => { + return messageBus.findLast({ ...criteria, cluster_id: clusterId }); + }, + count: (criteria) => { + return messageBus.count({ ...criteria, cluster_id: clusterId }); + }, + since: (timestamp) => { + return messageBus.since({ cluster_id: clusterId, timestamp }); + }, + } + : null; + + // Cluster API wrapper - mirrors logic-engine.js + const clusterAPI = { + id: clusterId, + getAgents: () => { + return cluster ? cluster.agents || [] : []; + }, + getAgentsByRole: (role) => { + return cluster ? (cluster.agents || []).filter((a) => a.role === role) : []; + }, + getAgent: (id) => { + return cluster ? (cluster.agents || []).find((a) => a.id === id) : null; + }, + }; + + // Helper functions - mirrors logic-engine.js + const helpers = { + getConfig: require('../config-router').getConfig, + allResponded: (agents, topic, since) => { + if (!ledgerAPI) return false; + const responses = ledgerAPI.query({ topic, since }); + const responders = new Set(responses.map((r) => r.sender)); + return agents.every((a) => responders.has(a.id || a)); + }, + hasConsensus: (topic, since) => { + if (!ledgerAPI) return false; + const responses = ledgerAPI.query({ topic, since }); + if (responses.length === 0) return false; + return responses.every((r) => r.content?.data?.approved === true); + }, + }; + // Build sandbox context - similar to LogicEngine but focused on result data const sandbox = { // The parsed result from agent output - this is the main input @@ -545,6 +775,11 @@ function evaluateHookLogic(params) { // Triggering message (if available) message: context.triggeringMessage || null, + // APIs + ledger: ledgerAPI, + cluster: clusterAPI, + helpers, + // Safe built-ins Set, Map, diff --git a/src/agent/agent-input-injector.js b/src/agent/agent-input-injector.js new file mode 100644 index 00000000..064fa97f --- /dev/null +++ b/src/agent/agent-input-injector.js @@ -0,0 +1,141 @@ +/** + * AgentInputInjector - Live guidance injection for running agents + * + * Resolves the current task, validates attachable socket availability, + * and sends input via attach STDIN when possible. + */ + +const { getTask } = require('../../task-lib/store.js'); +const { sendInput } = require('../attach/send-input'); +const { isSocketAlive } = require('../attach/socket-discovery'); + +const DEFAULT_TIMEOUT_MS = 1500; + +function buildResult({ status, reason = null, method = null, taskId = null }) { + return { + status, + reason, + method, + taskId, + }; +} + +function ensureValidInputs(agent, text) { + if (!agent) { + throw new Error('AgentInputInjector: agent is required'); + } + if (typeof text !== 'string') { + throw new Error('AgentInputInjector: text must be a string'); + } + if (!text.trim()) { + throw new Error('AgentInputInjector: text cannot be empty'); + } +} + +function buildUnsupported(reason, taskId) { + return buildResult({ + status: 'unsupported', + reason, + taskId, + }); +} + +function getTaskId(agent) { + return agent.currentTaskId || null; +} + +function checkIsolation(agent, taskId) { + if (agent.isolation?.enabled) { + return buildUnsupported('isolation-enabled', taskId); + } + return null; +} + +function checkTaskId(taskId) { + if (!taskId) { + return buildUnsupported('no-current-task', null); + } + return null; +} + +function checkTaskInfo(taskInfo, taskId) { + if (!taskInfo) { + return buildUnsupported('task-not-found', taskId); + } + if (!taskInfo.socketPath) { + return buildUnsupported('no-socket', taskId); + } + if (!taskInfo.attachable) { + return buildUnsupported('task-not-attachable', taskId); + } + return null; +} + +async function checkSocketAlive(socketPath, taskId) { + const socketAlive = await isSocketAlive(socketPath); + if (!socketAlive) { + return buildUnsupported('socket-not-alive', taskId); + } + return null; +} + +function normalizePayload(text) { + return text.endsWith('\n') ? text : `${text}\n`; +} + +function resolveTimeout(options) { + return options.timeoutMs || DEFAULT_TIMEOUT_MS; +} + +function buildInjected(taskId) { + return buildResult({ + status: 'injected', + method: 'pty', + taskId, + }); +} + +function buildSendFailure(reason, taskId) { + return buildResult({ + status: 'unsupported', + reason: reason || 'send-failed', + method: 'pty', + taskId, + }); +} + +async function injectInput(agent, text, options = {}) { + ensureValidInputs(agent, text); + + const taskId = getTaskId(agent); + const isolationResult = checkIsolation(agent, taskId); + if (isolationResult) return isolationResult; + + const taskIdResult = checkTaskId(taskId); + if (taskIdResult) return taskIdResult; + + const taskInfo = getTask(taskId); + const taskInfoResult = checkTaskInfo(taskInfo, taskId); + if (taskInfoResult) return taskInfoResult; + + const socketResult = await checkSocketAlive(taskInfo.socketPath, taskId); + if (socketResult) return socketResult; + + const payload = normalizePayload(text); + const timeoutMs = resolveTimeout(options); + const result = await sendInput({ + socketPath: taskInfo.socketPath, + data: payload, + timeoutMs, + }); + + if (!result.ok) { + return buildSendFailure(result.error, taskId); + } + + return buildInjected(taskId); +} + +module.exports = { + injectInput, +}; diff --git a/src/agent/agent-lifecycle.js b/src/agent/agent-lifecycle.js index dcbbd128..727c976d 100644 --- a/src/agent/agent-lifecycle.js +++ b/src/agent/agent-lifecycle.js @@ -14,11 +14,148 @@ const { findMatchingTrigger, evaluateTrigger } = require('./agent-trigger-evaluator'); const { executeHook } = require('./agent-hook-executor'); +const IsolationManager = require('../isolation-manager'); +const crypto = require('crypto'); +const { bufferMessage, scheduleDrain, drainBufferedMessages } = require('../message-buffer'); const { analyzeProcessHealth, isPlatformSupported, STUCK_THRESHOLD, } = require('./agent-stuck-detector'); +const { normalizeProviderName } = require('../../lib/provider-names'); +const { loadSettings } = require('../../lib/settings'); +const { findPlatformMismatchReason } = require('./validation-platform'); +const { calculateRateLimitDelay, isRateLimitError } = require('./rate-limit-backoff'); + +const DEFAULT_VALIDATOR_IMAGE = 'zeroshot-cluster-base'; + +class HookExecutionError extends Error { + constructor(message, options) { + super(message); + this.name = 'HookExecutionError'; + this.hookFailure = true; + this.hookRetries = options?.hookRetries; + this.originalHookError = options?.originalHookError; + } +} + +function resolveValidatorIsolationConfig(agent) { + const config = agent.config?.isolation || {}; + if (config.type && config.type !== 'docker') { + return null; + } + + return { + image: config.image || DEFAULT_VALIDATOR_IMAGE, + mounts: config.mounts, + noMounts: config.noMounts, + containerHome: config.containerHome, + }; +} + +async function createValidatorIsolation(agent, isolationConfig) { + if (!IsolationManager.isDockerAvailable()) { + agent._log(`[${agent.id}] Docker not available - cannot retry validator in isolation`); + return null; + } + + const cluster = agent.cluster || {}; + const workDir = agent.config?.cwd || cluster.worktree?.path || cluster.cwd || process.cwd(); + const image = isolationConfig.image; + await IsolationManager.ensureImage(image); + + const manager = new IsolationManager({ image }); + const providerName = normalizeProviderName( + (agent._resolveProvider && agent._resolveProvider()) || + cluster.config?.forceProvider || + cluster.config?.defaultProvider || + loadSettings().defaultProvider || + 'claude' + ); + + const isolationClusterId = `${cluster.id}-validators`; + const containerId = await manager.createContainer(isolationClusterId, { + workDir, + image, + noMounts: isolationConfig.noMounts, + mounts: isolationConfig.mounts, + containerHome: isolationConfig.containerHome, + provider: providerName, + reuseExistingWorkspace: true, + }); + + const validatorIsolation = { + enabled: true, + manager, + clusterId: isolationClusterId, + containerId, + image, + workDir, + }; + + cluster.validatorIsolation = validatorIsolation; + return validatorIsolation; +} + +async function ensureValidatorIsolation(agent) { + const cluster = agent.cluster || {}; + + if (agent.isolation?.enabled) { + return agent.isolation; + } + + if (cluster.validatorIsolation?.enabled) { + agent.isolation = cluster.validatorIsolation; + return agent.isolation; + } + + if (cluster.validatorIsolationPromise) { + const isolation = await cluster.validatorIsolationPromise; + if (isolation?.enabled) { + agent.isolation = isolation; + } + return agent.isolation || null; + } + + const isolationConfig = resolveValidatorIsolationConfig(agent); + if (!isolationConfig) { + agent._log(`[${agent.id}] Validator isolation config is not docker - skipping fallback`); + return null; + } + + cluster.validatorIsolationPromise = createValidatorIsolation(agent, isolationConfig); + + try { + const isolation = await cluster.validatorIsolationPromise; + if (isolation?.enabled) { + agent.isolation = isolation; + return agent.isolation; + } + return null; + } finally { + cluster.validatorIsolationPromise = null; + } +} + +async function maybeRetryValidatorInDocker(agent, result) { + if (agent.role !== 'validator') return null; + if (agent.isolation?.enabled) return null; + if (agent._validatorIsolationAttemptedIteration === agent.iteration) { + return null; + } + + const reason = findPlatformMismatchReason(result?.result || {}); + if (!reason) return null; + + const isolation = await ensureValidatorIsolation(agent); + if (!isolation) { + return null; + } + + agent._validatorIsolationAttemptedIteration = agent.iteration; + agent._log(`[${agent.id}] Platform mismatch detected - retrying validator in Docker isolation`); + return reason; +} /** * Start the agent (begin listening for triggers) @@ -103,6 +240,10 @@ async function stop(agent) { * @param {Object} message - Incoming message */ async function handleMessage(agent, message) { + if (!agent._bufferedMessages) { + agent._bufferedMessages = []; + } + // Check if any trigger matches FIRST (before state check) const matchingTrigger = findMatchingTrigger({ triggers: agent.config.triggers, @@ -119,8 +260,16 @@ async function handleMessage(agent, message) { return; } if (agent.state !== 'idle') { + // IMPORTANT: Never drop a message that matches a trigger. + // Dropping validation/coordinator signals can wedge clusters in "running" state. + bufferMessage(agent, message); console.warn( - `[${agent.id}] āš ļø DROPPING message (busy, state=${agent.state}): ${message.topic}` + `[${agent.id}] āøļø BUFFERING message (busy, state=${agent.state}): ${message.topic}` + ); + scheduleDrain( + agent, + () => drainBufferedMessages(agent, (next) => handleMessage(agent, next), { label: 'Agent' }), + { label: 'Agent' } ); return; } @@ -194,8 +343,10 @@ async function executeTriggerAction(agent, trigger, message) { } /** - * Execute claude-zeroshots with built context - * Retries disabled by default. Set agent config `maxRetries` to enable (e.g., 3). + * Execute task with built context + * Default: uses settings.maxRetries (default 3) for exponential backoff retries. + * Rate limit errors (429, capacity exhausted) use longer delays (30s base). + * Override via agent config `maxRetries` to change retry behavior. * @param {AgentWrapper} agent - Agent instance * @param {Object} triggeringMessage - Message that triggered execution */ @@ -250,7 +401,7 @@ async function applyValidatorJitter(agent) { return; } - const jitterMs = Math.floor(Math.random() * 15000); // 0-15 seconds + const jitterMs = crypto.randomInt(0, 15000); // 0-15 seconds if (!agent.quiet) { agent._log( `[Agent ${agent.id}] Adding ${Math.round(jitterMs / 1000)}s jitter to prevent lock contention` @@ -350,11 +501,13 @@ ${'='.repeat(80)}`); } else { console.error(`${'='.repeat(80)} `); - // All hook retries exhausted - throw to trigger task-level handling - throw new Error( + // All hook retries exhausted - FAIL THE CLUSTER (do NOT rerun the whole task). + // Retrying the task wastes tokens and cannot fix a deterministic hook/config bug. + throw new HookExecutionError( `Hook execution failed after ${hookMaxRetries} attempts. ` + `Task completed successfully but hook could not publish result. ` + - `Original error: ${hookError.message}` + `Original error: ${hookError.message}`, + { hookRetries: hookMaxRetries, originalHookError: hookError.message } ); } } @@ -401,6 +554,13 @@ async function runTaskAttempt(agent, triggeringMessage) { throw new Error(result.error || 'Task execution failed'); } + const fallbackReason = await maybeRetryValidatorInDocker(agent, result); + if (fallbackReason) { + throw new Error( + `Validator platform mismatch detected (${fallbackReason}). Retrying in Docker isolation.` + ); + } + // Set state to idle BEFORE publishing lifecycle event // (so lifecycle message includes correct state) agent.state = 'idle'; @@ -424,7 +584,7 @@ ${'='.repeat(80)}`); async function handleLockContention() { // Lock contention - add significant jittered delay - const lockDelay = 10000 + Math.floor(Math.random() * 20000); // 10-30 seconds + const lockDelay = 10000 + crypto.randomInt(0, 20000); // 10-30 seconds console.error( `āš ļø Lock contention detected - waiting ${Math.round(lockDelay / 1000)}s before retry` ); @@ -488,6 +648,25 @@ ${'='.repeat(80)}`); // Non-validator agents: publish error and stop agent.state = 'error'; + // Hook failure: fail the whole cluster so it gets stopped + persisted (prevents deadlocked "running" clusters). + if (error?.hookFailure) { + agent._publish({ + topic: 'CLUSTER_FAILED', + receiver: 'broadcast', + content: { + text: `Cluster failed: onComplete hook failed for ${agent.id} - ${error.message}`, + data: { + reason: 'on_complete_hook_failed', + agentId: agent.id, + role: agent.role, + hookRetries: error.hookRetries ?? null, + originalHookError: error.originalHookError ?? null, + error: error.message, + }, + }, + }); + } + // Save failure info to cluster for resume capability agent.cluster.failureInfo = { agentId: agent.id, @@ -507,6 +686,9 @@ ${'='.repeat(80)}`); data: { error: error.message, stack: error.stack, + hookFailure: error?.hookFailure === true, + hookRetries: error?.hookRetries ?? undefined, + originalHookError: error?.originalHookError ?? undefined, agent: agent.id, role: agent.role, iteration: agent.iteration, @@ -541,19 +723,26 @@ ${'='.repeat(80)}`); agent.state = 'idle'; } -async function scheduleRetry(agent, error, attempt, maxRetries, baseDelay) { - const delay = baseDelay * Math.pow(2, attempt - 1); // 2s, 4s, 8s +async function scheduleRetry(agent, error, attempt, maxRetries, _baseDelay) { + // Use rate-limit-aware backoff (30s+ for 429s, 2s for others) + const settings = loadSettings(); + const delay = calculateRateLimitDelay(error, attempt, settings); + const isRateLimit = isRateLimitError(error); agent._publishLifecycle('RETRY_SCHEDULED', { attempt, maxRetries, delayMs: delay, error: error.message, + isRateLimitError: isRateLimit, }); - agent._log(`[${agent.id}] āš ļø Retrying in ${delay}ms... (${attempt + 1}/${maxRetries})`); + const prefix = isRateLimit ? 'šŸ”„ Rate limit - ' : 'āš ļø '; + agent._log( + `[${agent.id}] ${prefix}Retrying in ${Math.round(delay / 1000)}s... (${attempt + 1}/${maxRetries})` + ); - // Exponential backoff + // Rate-limit-aware backoff await new Promise((resolve) => setTimeout(resolve, delay)); agent._log(`[${agent.id}] šŸ”„ Starting retry attempt ${attempt + 1}/${maxRetries}`); @@ -590,9 +779,33 @@ async function handleTaskAttemptFailure({ return false; } +function maybeExtendMaxRetries({ + error, + attempt, + maxRetries, + sigtermRetryGranted, + noMessagesRetryGranted, +}) { + const message = error?.message || ''; + if (!message || attempt < maxRetries) { + return { maxRetries, sigtermRetryGranted, noMessagesRetryGranted }; + } + + if (message.includes('SIGTERM') && !sigtermRetryGranted) { + return { maxRetries: maxRetries + 1, sigtermRetryGranted: true, noMessagesRetryGranted }; + } + + if (message.toLowerCase().includes('no messages returned') && !noMessagesRetryGranted) { + return { maxRetries: maxRetries + 1, sigtermRetryGranted, noMessagesRetryGranted: true }; + } + + return { maxRetries, sigtermRetryGranted, noMessagesRetryGranted }; +} + /** * Execute claude-zeroshots with built context - * Retries disabled by default. Set agent config `maxRetries` to enable (e.g., 3). + * Default: uses settings.maxRetries (default 3) for exponential backoff retries. + * Override via agent config `maxRetries` to change retry behavior. * @param {AgentWrapper} agent - Agent instance * @param {Object} triggeringMessage - Message that triggered execution */ @@ -602,10 +815,13 @@ async function executeTask(agent, triggeringMessage) { return; } - // Default: no retries (maxRetries=1 means 1 attempt only) - // Set agent config `maxRetries: 3` to enable exponential backoff retries - const maxRetries = agent.config.maxRetries ?? 1; - const baseDelay = 2000; // 2 seconds + // Default: uses settings.maxRetries (default 3) + // Override via agent config `maxRetries` to change retry behavior + const settings = loadSettings(); + let maxRetries = agent.config.maxRetries ?? settings.maxRetries ?? 3; + const baseDelay = settings.backoffBaseMs ?? 2000; + let sigtermRetryGranted = false; + let noMessagesRetryGranted = false; for (let attempt = 1; attempt <= maxRetries; attempt++) { // Check if agent was stopped between retries @@ -617,6 +833,21 @@ async function executeTask(agent, triggeringMessage) { await runTaskAttempt(agent, triggeringMessage); return; } catch (error) { + if (error instanceof HookExecutionError) { + // Hook failures are deterministic; do not waste tokens retrying the provider task. + await handleFinalFailure(agent, triggeringMessage, error, 1); + return; + } + const updated = maybeExtendMaxRetries({ + error, + attempt, + maxRetries, + sigtermRetryGranted, + noMessagesRetryGranted, + }); + maxRetries = updated.maxRetries; + sigtermRetryGranted = updated.sigtermRetryGranted; + noMessagesRetryGranted = updated.noMessagesRetryGranted; const shouldStop = await handleTaskAttemptFailure({ agent, triggeringMessage, diff --git a/src/agent/agent-task-executor.js b/src/agent/agent-task-executor.js index 92ad8412..5bc4cd77 100644 --- a/src/agent/agent-task-executor.js +++ b/src/agent/agent-task-executor.js @@ -92,6 +92,95 @@ function sanitizeErrorMessage(error) { return error; } +function safeTail(text, maxChars) { + if (!text) return ''; + if (text.length <= maxChars) return text; + return text.slice(-maxChars); +} + +function getClaudeConfigDir() { + return process.env.CLAUDE_CONFIG_DIR || path.join(os.homedir(), '.claude'); +} + +function findLatestClaudeDebugFile(configDir) { + try { + const debugDir = path.join(configDir, 'debug'); + const latestLink = path.join(debugDir, 'latest'); + if (fs.existsSync(latestLink)) { + const resolved = fs.realpathSync(latestLink); + const stats = fs.statSync(resolved); + return { path: resolved, mtimeMs: stats.mtimeMs }; + } + + const entries = fs.readdirSync(debugDir); + let newest = null; + for (const entry of entries) { + const fullPath = path.join(debugDir, entry); + const stats = fs.statSync(fullPath); + if (!stats.isFile()) continue; + if (!newest || stats.mtimeMs > newest.mtimeMs) { + newest = { path: fullPath, mtimeMs: stats.mtimeMs }; + } + } + return newest; + } catch (error) { + return { error: error.message }; + } +} + +function readFileTail(filePath, maxBytes) { + try { + const fd = fs.openSync(filePath, 'r'); + try { + const size = fs.fstatSync(fd).size; + const start = Math.max(0, size - maxBytes); + const length = size - start; + if (length <= 0) return ''; + const buffer = Buffer.alloc(length); + fs.readSync(fd, buffer, 0, length, start); + return buffer.toString('utf8'); + } finally { + fs.closeSync(fd); + } + } catch { + return ''; + } +} + +function logNoMessagesReturned({ taskId, output, statusOutput, debug }) { + const claudeConfigDir = getClaudeConfigDir(); + const latestDebug = findLatestClaudeDebugFile(claudeConfigDir); + const latestDebugPath = latestDebug?.path || null; + const latestDebugTail = + latestDebugPath && typeof latestDebugPath === 'string' + ? safeTail(readFileTail(latestDebugPath, 4000), 4000) + : ''; + + const payload = { + event: 'NO_MESSAGES_RETURNED', + timestamp: new Date().toISOString(), + taskId, + agentId: debug?.agentId || null, + provider: debug?.providerName || null, + pid: debug?.pid || null, + cwd: debug?.cwd || null, + worktreePath: debug?.worktreePath || null, + isolation: debug?.isolation || false, + clusterId: debug?.clusterId || null, + logFilePath: debug?.logFilePath || null, + outputLen: output ? output.length : 0, + outputTail: safeTail(output || '', 1000), + statusOutputLen: statusOutput ? statusOutput.length : 0, + statusOutputTail: safeTail(statusOutput || '', 1000), + claudeConfigDir, + claudeDebugLatest: latestDebugPath, + claudeDebugLatestMtimeMs: latestDebug?.mtimeMs || null, + claudeDebugLatestTail: latestDebugTail, + }; + + console.error('[AgentTaskExecutor] Claude CLI returned no messages', payload); +} + /** * Extract error context from task output. * Shared by both isolated and non-isolated modes. @@ -101,9 +190,10 @@ function sanitizeErrorMessage(error) { * @param {string} [params.statusOutput] - Status command output (non-isolated only) * @param {string} params.taskId - Task ID for error messages * @param {boolean} [params.isNotFound=false] - True if task was not found + * @param {Object} [params.debug] - Additional debug context for logging * @returns {string|null} Sanitized error context or null if extraction failed */ -function extractErrorContext({ output, statusOutput, taskId, isNotFound = false }) { +function extractErrorContext({ output, statusOutput, taskId, isNotFound = false, debug }) { // Task not found - explicit error if (isNotFound) { return sanitizeErrorMessage(`Task ${taskId} not found (may have crashed or been killed)`); @@ -138,6 +228,14 @@ function extractErrorContext({ output, statusOutput, taskId, isNotFound = false ); } + // Claude CLI transient failure: no messages returned + if (fullOutput.includes('No messages returned')) { + logNoMessagesReturned({ taskId, output: fullOutput, statusOutput, debug }); + return sanitizeErrorMessage( + `Claude CLI returned no messages. This is usually transient; retry the task or resume the cluster.` + ); + } + // NEVER TRUNCATE OUTPUT - truncation corrupts structured JSON and causes false "crash" status // If output is too verbose, that's a prompt problem - fix the prompts, not the data const trimmedOutput = (output || '').trim(); @@ -203,6 +301,8 @@ let dangerousGitHookInstalled = false; /** * Extract token usage from NDJSON output. * Looks for the 'result' event line which contains usage data. + * Falls back to summing 'turn.completed' events for cache metrics + * when the result event doesn't include them. * * @param {string} output - Full NDJSON output from Claude CLI * @returns {Object|null} Token usage data or null if not found @@ -218,11 +318,34 @@ function extractTokenUsage(output, providerName = 'claude') { return null; } + let cacheReadInputTokens = resultEvent.cacheReadInputTokens || 0; + let cacheCreationInputTokens = resultEvent.cacheCreationInputTokens || 0; + + // Fallback: if result event has no cache data, extract from raw turn.completed events. + // Claude CLI emits turn.completed with cached_input_tokens but the result event may omit them. + if (cacheReadInputTokens === 0) { + const lines = output.split('\n'); + for (const line of lines) { + const trimmed = line.trim(); + if (!trimmed) continue; + try { + const raw = JSON.parse(trimmed); + if (raw.type === 'turn.completed' && raw.usage) { + const usage = raw.usage; + cacheReadInputTokens += usage.cached_input_tokens || usage.cache_read_input_tokens || 0; + cacheCreationInputTokens += usage.cache_creation_input_tokens || 0; + } + } catch { + // skip non-JSON lines + } + } + } + return { inputTokens: resultEvent.inputTokens || 0, outputTokens: resultEvent.outputTokens || 0, - cacheReadInputTokens: resultEvent.cacheReadInputTokens || 0, - cacheCreationInputTokens: resultEvent.cacheCreationInputTokens || 0, + cacheReadInputTokens, + cacheCreationInputTokens, totalCostUsd: resultEvent.cost || null, durationMs: resultEvent.duration || null, modelUsage: resultEvent.modelUsage || null, @@ -904,6 +1027,54 @@ function handleStatusExecError({ agent, state, ctPath, taskId, error, stderr, re return false; } + // CRITICAL: "ID not found" means task completed or was removed - FAIL-SAFE by restarting + // We have zero confidence about what happened: + // - Task may have completed successfully + // - Task may have failed and been cleaned up + // - Task may have been manually killed + // - Zeroshot storage may be corrupted + // With zero confidence ā†’ restart is safer than assuming success + const errorMessage = error.message || ''; + const stderrMessage = stderr || ''; + const isNotFound = + errorMessage.includes('ID not found') || + errorMessage.includes('Not found in tasks') || + stderrMessage.includes('ID not found') || + stderrMessage.includes('Not found in tasks'); + + if (isNotFound) { + console.warn( + `[Agent ${agent.id}] āš ļø Task ${taskId} not found - will restart to ensure completion` + ); + + if (!state.resolved) { + state.resolved = true; + finalizeLogFollow(agent, state); + + agent._publish({ + topic: 'AGENT_ERROR', + receiver: 'broadcast', + content: { + text: `Task ${taskId} not found - restarting for safety`, + data: { + taskId, + error: 'task_not_found', + role: agent.role, + iteration: agent.iteration, + }, + }, + }); + + resolve({ + success: false, + output: state.output, + error: `Task not found - restarting for safety`, + }); + } + + return true; + } + state.consecutiveExecFailures++; if (state.consecutiveExecFailures < MAX_STATUS_FAILURES) { return true; @@ -976,7 +1147,20 @@ function handleStatusCompletion({ finalizeLogFollow(agent, state); const errorContext = !success - ? extractErrorContext({ output: state.output, statusOutput: stdout, taskId }) + ? extractErrorContext({ + output: state.output, + statusOutput: stdout, + taskId, + debug: { + agentId: agent.id, + providerName, + pid: agent.processPid, + cwd: agent.config.cwd || process.cwd(), + worktreePath: agent.worktree?.path || null, + isolation: !!agent.isolation?.enabled, + logFilePath: state.logFilePath || null, + }, + }) : null; resolve({ @@ -1072,13 +1256,34 @@ function followClaudeTaskLogs(agent, taskId) { return createLogFollower({ agent, taskId, fsModule, ctPath, providerName }); } +// Cache zeroshot path at module load time (when PATH is correct) +let _cachedZeroshotPath = null; +function _resolveZeroshotPath() { + if (_cachedZeroshotPath) return _cachedZeroshotPath; + + try { + // Use safe execSync (already imported at top) with explicit PATH + const fullPath = execSync('which zeroshot', { + encoding: 'utf8', + env: { ...process.env }, // Pass current process's PATH + }).trim(); + if (fullPath) { + _cachedZeroshotPath = fullPath; + return fullPath; + } + } catch { + // which failed, fall back to bare command + } + _cachedZeroshotPath = 'zeroshot'; + return 'zeroshot'; +} + /** * Get path to claude-zeroshots executable * @returns {String} Path to zeroshot command */ function getClaudeTasksPath() { - // Use zeroshot command (unified CLI) - return 'zeroshot'; // Assumes zeroshot is installed globally + return _resolveZeroshotPath(); } /** @@ -1419,7 +1624,21 @@ async function checkIsolatedStatus({ const success = isSuccess && !isError; const errorContext = !success - ? extractErrorContext({ output: state.fullOutput, taskId, isNotFound }) + ? extractErrorContext({ + output: state.fullOutput, + taskId, + isNotFound, + debug: { + agentId: agent.id, + providerName, + pid: agent.processPid, + cwd: agent.config.cwd || process.cwd(), + worktreePath: agent.worktree?.path || null, + isolation: true, + clusterId, + logFilePath, + }, + }) : null; const parsedResult = await agent._parseResultOutput(state.fullOutput); @@ -1547,13 +1766,23 @@ function followClaudeTaskLogsIsolated(agent, taskId) { * @returns {Promise} Parsed result data */ async function parseResultOutput(agent, output) { - // Empty or error outputs = FAIL - if (!output || output.includes('Task not found') || output.includes('Process terminated')) { + // Empty outputs = FAIL + if (!output || !output.trim()) { throw new Error('Task execution failed - no output'); } const providerName = agent._resolveProvider ? agent._resolveProvider() : 'claude'; - const { extractJsonFromOutput } = require('./output-extraction'); + const { + extractJsonFromOutput, + extractCliError, + hasFatalStandaloneOutput, + } = require('./output-extraction'); + + // Check for CLI errors FIRST - surface the actual error message + const cliError = extractCliError(output); + if (cliError) { + throw new Error(`CLI error (${cliError.provider}): ${cliError.error}`); + } // Use clean extraction pipeline let parsed = extractJsonFromOutput(output, providerName); @@ -1584,6 +1813,9 @@ async function parseResultOutput(agent, output) { } if (!parsed) { + if (hasFatalStandaloneOutput(output)) { + throw new Error('Task execution failed - no output'); + } const trimmedOutput = output.trim(); console.error(`\n${'='.repeat(80)}`); console.error(`šŸ”“ AGENT OUTPUT MISSING REQUIRED JSON BLOCK`); diff --git a/src/agent/context-metrics.js b/src/agent/context-metrics.js new file mode 100644 index 00000000..256cec68 --- /dev/null +++ b/src/agent/context-metrics.js @@ -0,0 +1,160 @@ +const TOKENS_PER_CHAR_ESTIMATE = 4; + +function estimateTokensFromChars(chars) { + if (!Number.isFinite(chars) || chars <= 0) { + return 0; + } + + return Math.ceil(chars / TOKENS_PER_CHAR_ESTIMATE); +} + +function buildSectionMetrics(sections) { + const sectionMetrics = {}; + let totalChars = 0; + + for (const [sectionName, text] of Object.entries(sections)) { + const safeText = typeof text === 'string' ? text : ''; + const chars = safeText.length; + const estimatedTokens = estimateTokensFromChars(chars); + sectionMetrics[sectionName] = { chars, estimatedTokens }; + totalChars += chars; + } + + return { sectionMetrics, totalChars }; +} + +function buildSectionMetricsFromPacks(packs) { + const sectionMetrics = {}; + let totalChars = 0; + + for (const pack of packs) { + if (pack.status !== 'included') continue; + const sectionName = pack.section || pack.id || 'unknown'; + const chars = Number.isFinite(pack.chars) ? pack.chars : 0; + if (!sectionMetrics[sectionName]) { + sectionMetrics[sectionName] = { chars: 0, estimatedTokens: 0 }; + } + sectionMetrics[sectionName].chars += chars; + totalChars += chars; + } + + for (const section of Object.values(sectionMetrics)) { + section.estimatedTokens = estimateTokensFromChars(section.chars); + } + + return { sectionMetrics, totalChars }; +} + +function resolveLegacyMaxTokens(strategy) { + if (!strategy) { + return 100000; + } + + return strategy.maxTokens || 100000; +} + +function buildContextMetrics({ + clusterId, + agentId, + role, + iteration, + triggeringMessage, + strategy, + sections, + packs, + budget, + truncation, +}) { + const maxTokens = resolveLegacyMaxTokens(strategy); + const sourcesCount = Array.isArray(strategy?.sources) ? strategy.sources.length : 0; + const packMetrics = Array.isArray(packs) ? packs : []; + + let sectionMetrics = {}; + let totalChars = 0; + if (packMetrics.length > 0) { + const packTotals = buildSectionMetricsFromPacks(packMetrics); + sectionMetrics = packTotals.sectionMetrics; + totalChars = packTotals.totalChars; + } else if (sections) { + const sectionTotals = buildSectionMetrics(sections); + sectionMetrics = sectionTotals.sectionMetrics; + totalChars = sectionTotals.totalChars; + } + + return { + clusterId, + agentId, + role, + iteration, + triggeredBy: triggeringMessage?.topic || null, + triggerFrom: triggeringMessage?.sender || null, + strategy: { + maxTokens, + sourcesCount, + }, + budget: { + maxTokens: budget?.maxTokens ?? maxTokens, + remainingTokens: budget?.remainingTokens === undefined ? null : budget?.remainingTokens, + overBudgetTokens: budget?.overBudgetTokens ?? 0, + finalTokens: budget?.finalTokens ?? estimateTokensFromChars(totalChars), + }, + packs: packMetrics, + sections: sectionMetrics, + total: { + chars: totalChars, + estimatedTokens: estimateTokensFromChars(totalChars), + }, + truncation: { + maxContextChars: truncation?.maxContextChars || { + applied: false, + beforeChars: totalChars, + afterChars: totalChars, + }, + }, + }; +} + +function updateTotalMetrics(metrics, chars) { + if (!metrics || !Number.isFinite(chars)) { + return; + } + + metrics.total = { + chars, + estimatedTokens: estimateTokensFromChars(chars), + }; + + if (metrics.budget) { + metrics.budget.finalTokens = estimateTokensFromChars(chars); + } + + if (metrics.truncation?.maxContextChars) { + metrics.truncation.maxContextChars.afterChars = chars; + } +} + +function emitContextMetrics(metrics, { messageBus, clusterId, agentId }) { + if (process.env.ZEROSHOT_CONTEXT_METRICS === '1') { + console.log('[ContextMetrics]', JSON.stringify(metrics)); + } + + if (process.env.ZEROSHOT_CONTEXT_METRICS_LEDGER === '1' && messageBus?.publish) { + messageBus.publish({ + cluster_id: clusterId, + topic: 'CONTEXT_METRICS', + sender: agentId, + receiver: 'system', + content: { + data: metrics, + }, + }); + } +} + +module.exports = { + estimateTokensFromChars, + resolveLegacyMaxTokens, + buildContextMetrics, + updateTotalMetrics, + emitContextMetrics, +}; diff --git a/src/agent/context-pack-builder.js b/src/agent/context-pack-builder.js new file mode 100644 index 00000000..6a13fae6 --- /dev/null +++ b/src/agent/context-pack-builder.js @@ -0,0 +1,367 @@ +const { estimateTokensFromChars } = require('./context-metrics'); + +const PRIORITY_RANK = { + required: 0, + high: 1, + medium: 2, + low: 3, +}; + +const DEFAULT_PRIORITY = 'medium'; +const TRUNCATION_SUFFIX = '\n\n[Context truncated to fit limit]\n'; + +function normalizePriority(priority, required) { + if (required) return 'required'; + if (priority && PRIORITY_RANK[priority] !== undefined) return priority; + return DEFAULT_PRIORITY; +} + +function normalizePack(pack, index) { + const priority = normalizePriority(pack.priority, pack.required); + return { + ...pack, + priority, + required: pack.required || priority === 'required', + order: pack.order ?? index, + }; +} + +function renderVariant(pack, variant, cache) { + const cacheKey = `${pack.id}:${variant}`; + if (cache.has(cacheKey)) return cache.get(cacheKey); + + let text = ''; + if (variant === 'full') { + text = typeof pack.render === 'function' ? pack.render() : ''; + } else if (variant === 'compact') { + text = typeof pack.compact === 'function' ? pack.compact() : ''; + } + + if (typeof text !== 'string') { + text = ''; + } + + const chars = text.length; + const estimatedTokens = estimateTokensFromChars(chars); + const rendered = { text, chars, estimatedTokens }; + cache.set(cacheKey, rendered); + return rendered; +} + +function sortByPriorityThenOrder(a, b) { + const priorityDelta = PRIORITY_RANK[a.priority] - PRIORITY_RANK[b.priority]; + if (priorityDelta !== 0) return priorityDelta; + return a.order - b.order; +} + +function sortByOrder(a, b) { + return a.order - b.order; +} + +function sortByPriorityDescThenOrderDesc(a, b) { + const priorityDelta = PRIORITY_RANK[b.priority] - PRIORITY_RANK[a.priority]; + if (priorityDelta !== 0) return priorityDelta; + return b.order - a.order; +} + +function selectVariant(pack, remainingTokens, cache) { + const full = renderVariant(pack, 'full', cache); + const compact = pack.compact ? renderVariant(pack, 'compact', cache) : null; + + const hasFull = full.chars > 0; + const hasCompact = compact && compact.chars > 0; + + if (!hasFull && !hasCompact) { + return { + status: 'skipped', + variant: null, + reason: 'empty', + chars: 0, + estimatedTokens: 0, + }; + } + + if (pack.required) { + let chosen = full; + let variant = 'full'; + + if (!hasFull && hasCompact) { + chosen = compact; + variant = 'compact'; + } else if ( + Number.isFinite(remainingTokens) && + hasCompact && + full.estimatedTokens > remainingTokens + ) { + if ( + compact.estimatedTokens <= remainingTokens || + compact.estimatedTokens < full.estimatedTokens + ) { + chosen = compact; + variant = 'compact'; + } + } + + return { + status: 'included', + variant, + chars: chosen.chars, + estimatedTokens: chosen.estimatedTokens, + text: chosen.text, + }; + } + + if (!Number.isFinite(remainingTokens) || full.estimatedTokens <= remainingTokens) { + return { + status: 'included', + variant: 'full', + chars: full.chars, + estimatedTokens: full.estimatedTokens, + text: full.text, + }; + } + + if (hasCompact && compact.estimatedTokens <= remainingTokens) { + return { + status: 'included', + variant: 'compact', + chars: compact.chars, + estimatedTokens: compact.estimatedTokens, + text: compact.text, + }; + } + + return { + status: 'skipped', + variant: null, + reason: 'budget', + chars: 0, + estimatedTokens: 0, + }; +} + +function truncateText(text, targetChars) { + if (text.length <= targetChars) { + return { text, truncated: false }; + } + + if (targetChars <= 0) { + return { text: '', truncated: true }; + } + + if (targetChars <= TRUNCATION_SUFFIX.length) { + return { text: text.slice(0, targetChars), truncated: true }; + } + + const sliceLength = targetChars - TRUNCATION_SUFFIX.length; + return { text: text.slice(0, sliceLength) + TRUNCATION_SUFFIX, truncated: true }; +} + +function applyMaxCharsGuard({ packs, selected, decisions, cache, maxChars, totalChars }) { + let currentChars = totalChars; + if (!Number.isFinite(maxChars) || currentChars <= maxChars) { + return { applied: false, beforeChars: totalChars, afterChars: totalChars }; + } + + const includedOptional = packs + .filter((pack) => selected.has(pack.id) && !pack.required) + .sort(sortByPriorityDescThenOrderDesc); + + for (const pack of includedOptional) { + if (currentChars <= maxChars) break; + const decision = decisions.get(pack.id); + if (!decision || decision.variant === 'compact' || !pack.compact) continue; + + const compact = renderVariant(pack, 'compact', cache); + if (compact.chars === 0 || compact.chars >= decision.chars) continue; + + const previousChars = decision.chars; + selected.set(pack.id, { ...compact, variant: 'compact' }); + decision.variant = 'compact'; + decision.chars = compact.chars; + decision.estimatedTokens = compact.estimatedTokens; + decision.reason = decision.reason || 'max_chars'; + + currentChars -= previousChars - compact.chars; + currentChars = Math.max(0, currentChars); + } + + for (const pack of includedOptional) { + if (currentChars <= maxChars) break; + if (!selected.has(pack.id)) continue; + + const decision = decisions.get(pack.id); + currentChars -= decision?.chars || 0; + selected.delete(pack.id); + if (decision) { + decision.status = 'skipped'; + decision.reason = decision.reason || 'max_chars'; + decision.chars = 0; + decision.estimatedTokens = 0; + } + } + + if (currentChars > maxChars) { + let overage = currentChars - maxChars; + const requiredCandidates = packs + .filter((pack) => selected.has(pack.id) && pack.required) + .sort((a, b) => { + const preserveDelta = (a.preserve ? 1 : 0) - (b.preserve ? 1 : 0); + if (preserveDelta !== 0) return preserveDelta; + const sizeDelta = (selected.get(b.id)?.chars || 0) - (selected.get(a.id)?.chars || 0); + if (sizeDelta !== 0) return sizeDelta; + return b.order - a.order; + }); + + for (const pack of requiredCandidates) { + if (overage <= 0) break; + const decision = decisions.get(pack.id); + const selectedPack = selected.get(pack.id); + if (!decision || !selectedPack) continue; + + const targetChars = Math.max(0, selectedPack.chars - overage); + const truncated = truncateText(selectedPack.text, targetChars); + if (truncated.text.length === selectedPack.chars) continue; + + const newChars = truncated.text.length; + const reduced = selectedPack.chars - newChars; + overage -= reduced; + + selected.set(pack.id, { + text: truncated.text, + chars: newChars, + estimatedTokens: estimateTokensFromChars(newChars), + variant: selectedPack.variant, + }); + + decision.chars = newChars; + decision.estimatedTokens = estimateTokensFromChars(newChars); + decision.truncated = true; + decision.reason = decision.reason || 'max_chars'; + } + } + + const afterChars = Array.from(selected.values()).reduce((sum, item) => sum + item.chars, 0); + return { applied: true, beforeChars: totalChars, afterChars }; +} + +function buildContextPacks({ packs, maxTokens, maxChars }) { + const normalized = packs.map(normalizePack); + const selectionOrder = normalized.slice().sort(sortByPriorityThenOrder); + const renderCache = new Map(); + const decisions = new Map(); + const selected = new Map(); + + let remainingTokens = Number.isFinite(maxTokens) ? maxTokens : Infinity; + let overBudgetTokens = 0; + + for (const pack of selectionOrder) { + const selection = selectVariant(pack, remainingTokens, renderCache); + const decision = { + id: pack.id, + section: pack.section || null, + priority: pack.priority, + required: pack.required, + status: selection.status, + variant: selection.variant, + chars: selection.chars, + estimatedTokens: selection.estimatedTokens, + order: pack.order, + reason: selection.reason || null, + }; + + decisions.set(pack.id, decision); + + if (selection.status !== 'included') { + continue; + } + + selected.set(pack.id, { + text: selection.text, + chars: selection.chars, + estimatedTokens: selection.estimatedTokens, + variant: selection.variant, + }); + + if (!Number.isFinite(remainingTokens)) { + continue; + } + + if (selection.estimatedTokens > remainingTokens) { + overBudgetTokens += selection.estimatedTokens - remainingTokens; + remainingTokens = 0; + } else { + remainingTokens -= selection.estimatedTokens; + } + } + + const ordered = normalized.slice().sort(sortByOrder); + let context = ''; + for (const pack of ordered) { + const selectedPack = selected.get(pack.id); + if (selectedPack) { + context += selectedPack.text; + } + } + + const totalChars = context.length; + const truncation = applyMaxCharsGuard({ + packs: ordered, + selected, + decisions, + cache: renderCache, + maxChars, + totalChars, + }); + + if (truncation.applied) { + context = ''; + for (const pack of ordered) { + const selectedPack = selected.get(pack.id); + if (selectedPack) { + context += selectedPack.text; + } + } + } + + const finalChars = context.length; + const finalTokens = estimateTokensFromChars(finalChars); + const packDecisions = ordered.map((pack) => { + const decision = decisions.get(pack.id); + return { + id: decision.id, + section: decision.section, + priority: decision.priority, + required: decision.required, + status: decision.status, + variant: decision.variant, + chars: decision.chars, + estimatedTokens: decision.estimatedTokens, + order: decision.order, + reason: decision.reason, + truncated: decision.truncated || false, + }; + }); + + return { + context, + packDecisions, + budget: { + maxTokens, + remainingTokens: Number.isFinite(remainingTokens) ? remainingTokens : null, + overBudgetTokens, + finalTokens, + }, + truncation: { + maxContextChars: { + applied: truncation.applied, + beforeChars: truncation.beforeChars, + afterChars: truncation.afterChars, + }, + }, + }; +} + +module.exports = { + buildContextPacks, +}; diff --git a/src/agent/guidance-queue.js b/src/agent/guidance-queue.js new file mode 100644 index 00000000..5ced3638 --- /dev/null +++ b/src/agent/guidance-queue.js @@ -0,0 +1,77 @@ +const GUIDANCE_BLOCK_START = '<>'; +const GUIDANCE_BLOCK_END = '<>'; + +function formatGuidanceMessage(message) { + const timestamp = Number.isFinite(message.timestamp) + ? new Date(message.timestamp).toISOString() + : new Date().toISOString(); + const sender = message.sender || 'unknown'; + const topic = message.topic || 'GUIDANCE'; + const target = message.receiver || message.target_agent_id; + const targetSuffix = target ? ` -> ${target}` : ''; + + let formatted = `[${timestamp}] ${sender} (${topic}${targetSuffix})\n`; + if (message.content?.text) { + formatted += `${message.content.text}\n`; + } + if (message.content?.data) { + formatted += `${JSON.stringify(message.content.data, null, 2)}\n`; + } + + return formatted.trimEnd(); +} + +function formatGuidanceBlock(messages) { + if (!Array.isArray(messages) || messages.length === 0) return ''; + + const ordered = messages.slice().sort((a, b) => (a.timestamp || 0) - (b.timestamp || 0)); + + let block = '## Guidance (Queued)\n\n'; + block += `${GUIDANCE_BLOCK_START}\n`; + + ordered.forEach((message, index) => { + block += `${formatGuidanceMessage(message)}\n`; + if (index < ordered.length - 1) { + block += '\n'; + } + }); + + block += `\n${GUIDANCE_BLOCK_END}\n\n`; + return block; +} + +function collectQueuedGuidance({ messageBus, clusterId, agentId, lastDeliveredAt, limit }) { + if (!messageBus) { + throw new Error('collectQueuedGuidance: messageBus is required'); + } + if (!clusterId) { + throw new Error('collectQueuedGuidance: clusterId is required'); + } + if (!agentId) { + throw new Error('collectQueuedGuidance: agentId is required'); + } + + const messages = messageBus.queryGuidanceMailbox({ + cluster_id: clusterId, + target_agent_id: agentId, + lastDeliveredAt, + limit, + }); + + if (!messages.length) { + return { messages: [], latestTimestamp: null, guidanceBlock: '' }; + } + + const ordered = messages.slice().sort((a, b) => (a.timestamp || 0) - (b.timestamp || 0)); + const latestTimestamp = ordered[ordered.length - 1].timestamp; + const guidanceBlock = formatGuidanceBlock(ordered); + + return { messages: ordered, latestTimestamp, guidanceBlock }; +} + +module.exports = { + GUIDANCE_BLOCK_START, + GUIDANCE_BLOCK_END, + formatGuidanceBlock, + collectQueuedGuidance, +}; diff --git a/src/agent/output-extraction.js b/src/agent/output-extraction.js index 82312b38..416bf404 100644 --- a/src/agent/output-extraction.js +++ b/src/agent/output-extraction.js @@ -105,18 +105,40 @@ function extractResultContent(obj) { */ function extractFromTextEvents(output, providerName) { const provider = getProvider(providerName); - const events = parseChunkWithProvider(provider, output); + const normalized = output + .split('\n') + .map((line) => stripTimestamp(line)) + .filter(Boolean) + .join('\n'); + const events = parseChunkWithProvider(provider, normalized); + + // Fast-path: many providers eventually emit the full JSON as a single text event. + // Scan from the end to find the last parseable JSON snippet without requiring + // the entire concatenated stream to be valid JSON. + for (let i = events.length - 1; i >= 0; i--) { + const e = events[i]; + if (e.type !== 'text' || typeof e.text !== 'string') continue; + const direct = extractDirectJson(e.text) || extractFromMarkdown(e.text); + if (direct) return direct; + } // Accumulate all text events - const textContent = events - .filter((e) => e.type === 'text') - .map((e) => e.text) - .join(''); + const textEvents = events.filter((e) => e.type === 'text').map((e) => e.text); + const textContent = textEvents.join(''); if (!textContent.trim()) return null; // Try parsing accumulated text as JSON - return extractDirectJson(textContent) || extractFromMarkdown(textContent); + const combined = extractDirectJson(textContent) || extractFromMarkdown(textContent); + if (combined) return combined; + + for (let i = textEvents.length - 1; i >= 0; i--) { + const candidate = textEvents[i]; + const parsed = extractDirectJson(candidate) || extractFromMarkdown(candidate); + if (parsed) return parsed; + } + + return null; } /** @@ -145,10 +167,49 @@ function extractFromMarkdown(text) { return null; } +/** + * CLI metadata fields that indicate raw provider output (not agent content). + * These objects should be rejected - they're wrapper metadata, not actual output. + */ +const CLI_METADATA_FIELDS = new Set([ + 'duration_ms', + 'duration_api_ms', + 'total_cost_usd', + 'session_id', + 'num_turns', + 'permission_denials', + 'modelUsage', +]); + +/** + * Check if an object looks like CLI metadata rather than agent output. + * CLI metadata has specific fields like duration_ms, session_id, etc. + * + * @param {object} obj - Parsed JSON object + * @returns {boolean} True if this looks like CLI metadata + */ +function isCliMetadata(obj) { + if (!obj || typeof obj !== 'object') return false; + + // If it has type:result, it's definitely CLI wrapper (should have been handled by extractFromResultWrapper) + if (obj.type === 'result') return true; + + // Check for CLI-specific metadata fields + const keys = Object.keys(obj); + const metadataFieldCount = keys.filter((k) => CLI_METADATA_FIELDS.has(k)).length; + + // If 2+ CLI metadata fields present, reject as CLI output + return metadataFieldCount >= 2; +} + /** * Strategy 4: Direct JSON parse * Handles raw JSON output (single-line or multi-line) * + * IMPORTANT: Rejects CLI metadata objects to prevent schema validation + * against wrong data structure (e.g., validating {duration_ms, session_id} + * against agent schema expecting {summary, completionStatus}). + * * @param {string} text - Text to parse * @returns {object|null} Parsed JSON or null */ @@ -161,6 +222,10 @@ function extractDirectJson(text) { try { const parsed = JSON.parse(trimmed); if (typeof parsed === 'object' && parsed !== null && !Array.isArray(parsed)) { + // Reject CLI metadata - this is wrapper output, not agent content + if (isCliMetadata(parsed)) { + return null; + } return parsed; } } catch { @@ -170,6 +235,91 @@ function extractDirectJson(text) { return null; } +/** + * Extract CLI error from provider output (all providers). + * Returns the error message if the CLI reported an error, null otherwise. + * + * Provider error formats: + * - Claude: {type:"result", is_error:true, errors:["msg"]} or {type:"result", subtype:"error"} + * - Codex: {type:"turn.failed", error:{message:"msg"}} + * - Gemini: {type:"result", success:false, error:"msg"} + * - Opencode: {type:"session.error", error:{message:"msg"}} + * + * @param {string} output - Raw CLI output + * @returns {{error: string, provider: string}|null} Error info or null + */ +function extractCliError(output) { + if (!output || typeof output !== 'string') return null; + + const lines = output.split('\n'); + + for (const line of lines) { + const content = stripTimestamp(line); + if (!content.startsWith('{')) continue; + + let obj; + try { + obj = JSON.parse(content); + } catch { + continue; + } + + // Claude: {type:"result", is_error:true, errors:[...]} + if (obj.type === 'result' && obj.is_error === true) { + const errorMsg = Array.isArray(obj.errors) + ? obj.errors.join('; ') + : obj.error || obj.result || 'Unknown CLI error'; + return { error: errorMsg, provider: 'claude' }; + } + + // Claude: {type:"result", subtype:"error"} + if (obj.type === 'result' && obj.subtype === 'error') { + const errorMsg = obj.error || obj.result || 'CLI returned error'; + return { error: errorMsg, provider: 'claude' }; + } + + // Codex: {type:"turn.failed", error:{message:"..."}} + if (obj.type === 'turn.failed') { + const errorMsg = obj.error?.message || obj.error || 'Turn failed'; + return { error: errorMsg, provider: 'codex' }; + } + + // Gemini: {type:"result", success:false, error:"..."} + if (obj.type === 'result' && obj.success === false && obj.error) { + return { error: obj.error, provider: 'gemini' }; + } + + // Opencode: {type:"session.error", error:{...}} + if (obj.type === 'session.error') { + const errorMsg = + obj.error?.data?.message || obj.error?.message || obj.error?.name || 'Session error'; + return { error: errorMsg, provider: 'opencode' }; + } + } + + return null; +} + +/** + * Detects fatal standalone output lines that indicate no task output was produced. + * Only matches when the line itself is the fatal message (not when it appears inside JSON). + * + * @param {string} output - Raw output text + * @returns {boolean} True if a standalone fatal line is present + */ +function hasFatalStandaloneOutput(output) { + if (!output || typeof output !== 'string') return false; + const lines = output.split('\n'); + for (const line of lines) { + const stripped = stripTimestamp(line).trim(); + if (!stripped) continue; + if (/^(task not found|process terminated)\b/i.test(stripped)) { + return true; + } + } + return false; +} + /** * Main extraction function - tries all strategies in priority order * @@ -183,11 +333,6 @@ function extractJsonFromOutput(output, providerName = 'claude') { const trimmedOutput = output.trim(); if (!trimmedOutput) return null; - // Check for fatal error indicators - if (trimmedOutput.includes('Task not found') || trimmedOutput.includes('Process terminated')) { - return null; - } - // Strategy 1: Result wrapper (Claude format) const fromWrapper = extractFromResultWrapper(trimmedOutput); if (fromWrapper) return fromWrapper; @@ -204,14 +349,20 @@ function extractJsonFromOutput(output, providerName = 'claude') { const fromDirect = extractDirectJson(trimmedOutput); if (fromDirect) return fromDirect; + if (hasFatalStandaloneOutput(trimmedOutput)) { + return null; + } + return null; } module.exports = { extractJsonFromOutput, + extractCliError, extractFromResultWrapper, extractFromTextEvents, extractFromMarkdown, extractDirectJson, stripTimestamp, + hasFatalStandaloneOutput, }; diff --git a/src/agent/rate-limit-backoff.js b/src/agent/rate-limit-backoff.js new file mode 100644 index 00000000..460a9af6 --- /dev/null +++ b/src/agent/rate-limit-backoff.js @@ -0,0 +1,82 @@ +/** + * Rate-limit-aware backoff for API retries + * + * Rate limit errors (429, capacity exhausted, quota exceeded) need LONGER delays + * than transient errors (timeouts, network issues). + * + * - Regular errors: 2s base, exponential backoff up to 30s + * - Rate limits: 30s base, exponential backoff up to 5 minutes + * - Retry-After header: Honored if present (capped at 5 min) + */ + +/** + * Check if error is a rate limit error + * @param {Error|string} error - Error object or message + * @returns {boolean} True if this is a rate limit error + */ +function isRateLimitError(error) { + const msg = error?.message || String(error); + return /\b429\b|rate.?limit|too many requests|no capacity|quota.?exceeded|resource.?exhausted/i.test( + msg + ); +} + +/** + * Parse Retry-After from error message + * Looks for patterns like "Retry-After: 120" or "retry after 120 seconds" + * @param {Error} error - Error object + * @returns {number|null} Seconds to wait, or null if not found + */ +function parseRetryAfter(error) { + const msg = error?.message || ''; + // Match "Retry-After: 120" or "retry after 120" (seconds) + const match = msg.match(/retry.?after[:\s]+(\d+)/i); + return match ? parseInt(match[1], 10) : null; +} + +/** + * Calculate delay for retry with rate-limit awareness + * + * Rate limit errors get 30s base delay instead of 2s. + * Regular errors use exponential backoff from 2s base. + * + * @param {Error} error - The error that occurred + * @param {number} attempt - Current attempt number (1-based) + * @param {Object} settings - Settings with backoff config + * @param {number} [settings.backoffBaseMs=2000] - Base delay for regular errors + * @param {number} [settings.backoffMaxMs=30000] - Max delay for regular errors + * @param {number} [settings.jitterFactor=0.2] - Jitter factor (Ā±20%) + * @returns {number} Delay in milliseconds + */ +function calculateRateLimitDelay(error, attempt, settings = {}) { + const baseDelay = settings.backoffBaseMs ?? 2000; + const maxDelay = settings.backoffMaxMs ?? 30000; + const jitter = settings.jitterFactor ?? 0.2; + + // Check for Retry-After header in error message + const retryAfter = parseRetryAfter(error); + if (retryAfter) { + // Honor Retry-After but cap at 5 minutes + return Math.min(retryAfter * 1000, 300000); + } + + // Rate limits get 30s base, others get normal base + const isRateLimit = isRateLimitError(error); + const effectiveBase = isRateLimit ? 30000 : baseDelay; + + // Exponential: base * 2^(attempt-1) + let delay = effectiveBase * Math.pow(2, attempt - 1); + + // Cap at appropriate max (5 min for rate limits, settings max for others) + delay = Math.min(delay, isRateLimit ? 300000 : maxDelay); + + // Add jitter (Ā±jitterFactor) + const jitterAmount = delay * jitter * (Math.random() * 2 - 1); + return Math.round(delay + jitterAmount); +} + +module.exports = { + calculateRateLimitDelay, + isRateLimitError, + parseRetryAfter, +}; diff --git a/src/agent/validation-platform.js b/src/agent/validation-platform.js new file mode 100644 index 00000000..b41c7c7b --- /dev/null +++ b/src/agent/validation-platform.js @@ -0,0 +1,35 @@ +const PLATFORM_MISMATCH_REGEX = + /EBADPLATFORM|Unsupported platform|darwin-arm64|linux-x64|@esbuild\/linux-x64/i; + +function isPlatformMismatchReason(reason) { + if (!reason) return false; + return PLATFORM_MISMATCH_REGEX.test(String(reason)); +} + +function findPlatformMismatchReason(result = {}) { + const criteriaResults = result.criteriaResults; + if (Array.isArray(criteriaResults)) { + for (const criteria of criteriaResults) { + if (criteria?.status !== 'CANNOT_VALIDATE') continue; + if (isPlatformMismatchReason(criteria.reason)) { + return String(criteria.reason); + } + } + } + + const errors = result.errors; + if (Array.isArray(errors)) { + for (const error of errors) { + if (isPlatformMismatchReason(error)) { + return String(error); + } + } + } + + return null; +} + +module.exports = { + isPlatformMismatchReason, + findPlatformMismatchReason, +}; diff --git a/src/agents/git-pusher-template.js b/src/agents/git-pusher-template.js index 072eb89b..de81f546 100644 --- a/src/agents/git-pusher-template.js +++ b/src/agents/git-pusher-template.js @@ -18,59 +18,198 @@ const SHARED_TRIGGER_SCRIPT = `const validators = cluster.getAgentsByRole('valid const lastPush = ledger.findLast({ topic: 'IMPLEMENTATION_READY' }); if (!lastPush) return false; if (validators.length === 0) return true; + const results = ledger.query({ topic: 'VALIDATION_RESULT', since: lastPush.timestamp }); -if (results.length < validators.length) return false; -const allApproved = results.every(r => r.content?.data?.approved === 'true' || r.content?.data?.approved === true); -if (!allApproved) return false; -const hasRealEvidence = results.every(r => { - const criteria = r.content?.data?.criteriaResults || []; - return criteria.every(c => { - return c.evidence?.command && typeof c.evidence?.exitCode === 'number' && c.evidence?.output?.length > 10; +if (results.length === 0) return false; + +const validatorIds = new Set(validators.map((v) => v.id)); +const validatorResults = results.filter((r) => validatorIds.has(r.sender)); + +// Two supported patterns: +// 1) Per-validator VALIDATION_RESULT (sender is a validator) ā†’ require all validators approve. +// 2) Consensus-only VALIDATION_RESULT (sender is coordinator) ā†’ treat latest result as final. +if (validatorResults.length === 0) { + let latest = null; + for (const msg of results) { + if (!latest || (typeof msg.timestamp === 'number' && msg.timestamp > latest.timestamp)) { + latest = msg; + } + } + const approved = latest?.content?.data?.approved; + return approved === true || approved === 'true'; +} + +const latestByValidator = new Map(); +for (const msg of validatorResults) { + latestByValidator.set(msg.sender, msg); +} +if (latestByValidator.size < validators.length) return false; + +for (const validator of validators) { + const msg = latestByValidator.get(validator.id); + const approved = msg?.content?.data?.approved; + if (!(approved === true || approved === 'true')) return false; +} + +const hasSufficientEvidence = Array.from(latestByValidator.values()).every((r) => { + const criteria = r.content?.data?.criteriaResults; + if (!Array.isArray(criteria) || criteria.length === 0) return true; + return criteria.every((c) => { + const status = String(c.status || '').toUpperCase(); + if (status === 'CANNOT_VALIDATE') return true; + if (status === 'SKIPPED') return true; + if (status === 'CANNOT_VALIDATE_YET') return false; + const evidence = c.evidence || {}; + const hasCommand = typeof evidence.command === 'string' && evidence.command.trim().length > 0; + const exitCode = evidence.exitCode; + const hasExitCode = + typeof exitCode === 'number' || + (typeof exitCode === 'string' && exitCode.trim() !== '' && Number.isFinite(Number(exitCode))); + const hasOutput = evidence.output === undefined || typeof evidence.output === 'string'; + return hasCommand && hasExitCode && hasOutput; }); }); -return hasRealEvidence;`; + +return hasSufficientEvidence;`; + +const { readRepoSettings } = require('../../lib/repo-settings'); + +function getSafeBranchName(value) { + if (typeof value !== 'string') { + return null; + } + + const trimmed = value.trim(); + if (trimmed.length === 0) { + return null; + } + + // Conservative allowlist to avoid shell injection in generated CLI commands. + if (!/^[A-Za-z0-9._/-]+$/.test(trimmed)) { + return null; + } + + return trimmed; +} + +function parseBool(value) { + if (typeof value === 'boolean') return value; + if (typeof value !== 'string') return null; + const trimmed = value.trim().toLowerCase(); + if (trimmed === '1' || trimmed === 'true' || trimmed === 'yes') return true; + if (trimmed === '0' || trimmed === 'false' || trimmed === 'no') return false; + return null; +} + +function normalizeCloseIssueMode(value) { + if (typeof value !== 'string') return null; + const trimmed = value.trim().toLowerCase(); + if (trimmed === 'auto') return 'auto'; + if (trimmed === 'always') return 'always'; + if (trimmed === 'never') return 'never'; + return null; +} /** - * Platform-specific CLI commands and terminology + * Resolve GitHub configuration from CLI options and repo settings. + * Priority: CLI options > repo settings (.zeroshot/settings.json) > defaults + * + * @param {Object} options - CLI options + * @param {string} [options.prBase] - Target branch for PRs + * @param {boolean} [options.mergeQueue] - Use GitHub merge queue + * @param {string} [options.closeIssue] - When to close issue: auto|always|never + * @returns {Object} Resolved configuration */ -const PLATFORM_CONFIGS = { - github: { - prName: 'PR', - prNameLower: 'pull request', - createCmd: 'gh pr create --title "feat: {{issue_title}}" --body "Closes #{{issue_number}}"', - mergeCmd: 'gh pr merge --merge --auto', - mergeFallbackCmd: 'gh pr merge --merge', - prUrlExample: 'https://github.com/owner/repo/pull/123', - outputFields: { urlField: 'pr_url', numberField: 'pr_number', mergedField: 'merged' }, - }, - gitlab: { - prName: 'MR', - prNameLower: 'merge request', - createCmd: - 'glab mr create --title "feat: {{issue_title}}" --description "Closes #{{issue_number}}"', - mergeCmd: 'glab mr merge --auto-merge', - mergeFallbackCmd: 'glab mr merge', - prUrlExample: 'https://gitlab.com/owner/repo/-/merge_requests/123', - outputFields: { urlField: 'mr_url', numberField: 'mr_number', mergedField: 'merged' }, - }, - 'azure-devops': { - prName: 'PR', - prNameLower: 'pull request', - createCmd: - 'az repos pr create --title "feat: {{issue_title}}" --description "Closes #{{issue_number}}"', - mergeCmd: 'az repos pr update --id --auto-complete true', - mergeFallbackCmd: 'az repos pr update --id --status completed', - prUrlExample: 'https://dev.azure.com/org/project/_git/repo/pullrequest/123', - outputFields: { - urlField: 'pr_url', - numberField: 'pr_number', - mergedField: 'merged', - autoCompleteField: 'auto_complete', +function resolveGitHubConfig(options = {}) { + const repoSettingsResult = readRepoSettings(process.cwd()); + const repoSettings = repoSettingsResult.settings || {}; + const repoGithub = repoSettings.github || {}; + + // CLI options override repo settings + const prBase = getSafeBranchName(options.prBase) || getSafeBranchName(repoGithub.prBase); + + const useMergeQueue = + options.mergeQueue === true || + (options.mergeQueue !== false && parseBool(repoGithub.useMergeQueue) === true); + + const closeIssueMode = + normalizeCloseIssueMode(options.closeIssue) || + normalizeCloseIssueMode(repoGithub.closeIssue) || + (parseBool(repoGithub.closeIssue) === true ? 'always' : null) || + 'never'; + + return { prBase, useMergeQueue, closeIssueMode }; +} + +/** + * Generate platform-specific configuration based on resolved GitHub config. + * + * @param {string} platform - Platform ID ('github', 'gitlab', 'azure-devops') + * @param {Object} config - Resolved GitHub config from resolveGitHubConfig() + * @returns {Object|null} Platform configuration or null if unsupported + */ +function getPlatformConfig(platform, config = {}) { + const { prBase, useMergeQueue, closeIssueMode } = config; + + const PLATFORM_CONFIGS = { + github: { + prName: 'PR', + prNameLower: 'pull request', + createCmd: `gh pr create${prBase ? ` --base ${prBase}` : ''} --title "feat: {{issue_title}}" --body "Closes #{{issue_number}}"`, + mergeCmd: useMergeQueue + ? `PR_ID="$(gh pr view --json id --jq .id)" +gh api graphql -f query='mutation($id:ID!){enqueuePullRequest(input:{pullRequestId:$id}){mergeQueueEntry{state}}}' -f id="$PR_ID" +echo "Waiting for merge..." +until gh pr view --json mergedAt --jq .mergedAt | grep -q .; do + sleep 20 +done` + : 'gh pr merge --merge --auto', + mergeFallbackCmd: useMergeQueue ? 'gh pr merge --merge --auto' : 'gh pr merge --merge', + prUrlExample: 'https://github.com/owner/repo/pull/123', + outputFields: { urlField: 'pr_url', numberField: 'pr_number', mergedField: 'merged' }, + rebaseBranch: prBase || 'main', + usesMergeQueue: useMergeQueue, + closeIssueMode: closeIssueMode || 'never', }, - // Azure requires extracting PR ID from create output - requiresPrIdExtraction: true, - }, -}; + gitlab: { + prName: 'MR', + prNameLower: 'merge request', + createCmd: + 'glab mr create --title "feat: {{issue_title}}" --description "Closes #{{issue_number}}"', + mergeCmd: 'glab mr merge --auto-merge', + mergeFallbackCmd: 'glab mr merge', + prUrlExample: 'https://gitlab.com/owner/repo/-/merge_requests/123', + outputFields: { urlField: 'mr_url', numberField: 'mr_number', mergedField: 'merged' }, + closeIssueMode: closeIssueMode || 'never', + }, + 'azure-devops': { + prName: 'PR', + prNameLower: 'pull request', + createCmd: + 'az repos pr create --title "feat: {{issue_title}}" --description "Closes #{{issue_number}}"', + mergeCmd: 'az repos pr update --id --auto-complete true', + mergeFallbackCmd: 'az repos pr update --id --status completed', + prUrlExample: 'https://dev.azure.com/org/project/_git/repo/pullrequest/123', + outputFields: { + urlField: 'pr_url', + numberField: 'pr_number', + mergedField: 'merged', + autoCompleteField: 'auto_complete', + }, + // Azure requires extracting PR ID from create output + requiresPrIdExtraction: true, + closeIssueMode: closeIssueMode || 'never', + }, + }; + + return PLATFORM_CONFIGS[platform] || null; +} + +/** + * Get list of supported platforms for git-pusher + * @returns {string[]} Array of platform IDs + */ +const SUPPORTED_PLATFORMS = ['github', 'gitlab', 'azure-devops']; /** * Generate the prompt for a specific platform @@ -87,6 +226,9 @@ function generatePrompt(config) { prUrlExample, outputFields, requiresPrIdExtraction, + rebaseBranch, + usesMergeQueue, + closeIssueMode, } = config; // Azure-specific instructions for PR ID extraction @@ -99,14 +241,20 @@ Save the PR ID to a variable for step 6.` // Azure uses different merge terminology const mergeDescription = requiresPrIdExtraction ? 'SET AUTO-COMPLETE (MANDATORY - THIS IS NOT OPTIONAL)' - : `MERGE THE ${prName} (MANDATORY - THIS IS NOT OPTIONAL)`; + : usesMergeQueue + ? `ENQUEUE INTO MERGE QUEUE AND WAIT UNTIL THE ${prName} IS MERGED (MANDATORY - THIS IS NOT OPTIONAL)` + : `MERGE THE ${prName} (MANDATORY - THIS IS NOT OPTIONAL)`; const mergeExplanation = requiresPrIdExtraction ? `Replace with the actual PR number from step 5. This enables auto-complete (auto-merge when CI passes). If auto-complete is not available or you need to merge immediately:` - : `This sets auto-merge. If it fails (e.g., no auto-merge enabled), try:`; + : usesMergeQueue + ? `This enqueues the ${prName} into GitHub's merge queue and waits until it is merged. + +If enqueue fails (merge queue not enabled, missing permissions, etc.), fall back to auto-merge:` + : `This sets auto-merge. If it fails (e.g., no auto-merge enabled), try:`; const postMergeStatus = requiresPrIdExtraction ? 'PR IS CREATED AND AUTO-COMPLETE IS SET' @@ -182,10 +330,10 @@ ${mergeFallbackCmd} \`\`\` šŸšØ IF MERGE FAILS DUE TO CONFLICTS - YOU MUST RESOLVE THEM: -a) Pull latest main and rebase: +a) Pull latest ${rebaseBranch || 'main'} and rebase: \`\`\`bash - git fetch origin main - git rebase origin/main + git fetch origin ${rebaseBranch || 'main'} + git rebase origin/${rebaseBranch || 'main'} \`\`\` b) If conflicts appear - RESOLVE THEM IMMEDIATELY: - Read the conflicting files @@ -199,12 +347,45 @@ c) Force push the resolved branch: \`\`\` d) Retry merge: \`\`\`bash - ${mergeFallbackCmd} - \`\`\` +${mergeFallbackCmd} +\`\`\` REPEAT UNTIL MERGED. DO NOT GIVE UP. DO NOT SKIP. THE ${prName} MUST BE ${requiresPrIdExtraction ? 'SET TO AUTO-COMPLETE' : 'MERGED'}. If merge is blocked by CI, wait and retry. ${requiresPrIdExtraction ? 'The auto-complete will merge when CI passes.' : 'If blocked by reviews, set auto-merge.'} +${ + closeIssueMode !== 'never' + ? `### STEP 7: Close the issue (MANDATORY) +\`\`\`bash +if [ "{{issue_number}}" != "unknown" ]; then + ISSUE_STATE="$(gh issue view {{issue_number}} --json state --jq .state 2>/dev/null || true)" + if [ "$ISSUE_STATE" = "OPEN" ]; then + BASE_BRANCH="${rebaseBranch || 'main'}" + DEFAULT_BRANCH="$(gh repo view --json defaultBranchRef --jq .defaultBranchRef.name 2>/dev/null || true)" + SHOULD_CLOSE="0" + if [ "${closeIssueMode}" = "always" ]; then + SHOULD_CLOSE="1" + elif [ "${closeIssueMode}" = "auto" ]; then + if [ -z "$DEFAULT_BRANCH" ] || [ "$BASE_BRANCH" != "$DEFAULT_BRANCH" ]; then + SHOULD_CLOSE="1" + fi + fi + + if [ "$SHOULD_CLOSE" = "1" ]; then + PR_URL="$(gh pr view --json url --jq .url 2>/dev/null || true)" + if [ -n "$PR_URL" ]; then + gh issue close {{issue_number}} --comment "Implemented in $PR_URL" + else + gh issue close {{issue_number}} --comment "Implemented" + fi + fi + fi +fi +\`\`\` +Only do this AFTER the ${prName} is merged.` + : '' +} + ## CRITICAL RULES - Execute EVERY step in order (1, 2, 3, 4, 5, 6) - Do NOT skip git add -A @@ -225,14 +406,20 @@ ${finalOutputNote}`; * Generate a git-pusher agent configuration for a specific platform * * @param {string} platform - Platform ID ('github', 'gitlab', 'azure-devops') + * @param {Object} [options] - CLI options for GitHub configuration + * @param {string} [options.prBase] - Target branch for PRs + * @param {boolean} [options.mergeQueue] - Use GitHub merge queue + * @param {string} [options.closeIssue] - When to close issue: auto|always|never * @returns {Object} Agent configuration object * @throws {Error} If platform is not supported */ -function generateGitPusherAgent(platform) { - const config = PLATFORM_CONFIGS[platform]; +function generateGitPusherAgent(platform, options = {}) { + // Resolve config from CLI options and repo settings + const resolvedConfig = resolveGitHubConfig(options); + const platformConfig = getPlatformConfig(platform, resolvedConfig); - if (!config) { - const supported = Object.keys(PLATFORM_CONFIGS).join(', '); + if (!platformConfig) { + const supported = SUPPORTED_PLATFORMS.join(', '); throw new Error(`Unsupported platform '${platform}'. Supported: ${supported}`); } @@ -250,11 +437,34 @@ function generateGitPusherAgent(platform) { action: 'execute_task', }, ], - prompt: generatePrompt(config), + prompt: generatePrompt(platformConfig), + hooks: { + onComplete: { + action: 'verify_github_pr', + // No config needed - verification reads from result.structured_output + // and publishes CLUSTER_COMPLETE only if verification passes + }, + }, output: { topic: 'PR_CREATED', publishAfter: 'CLUSTER_COMPLETE', }, + structuredOutput: { + type: 'object', + properties: { + pr_number: { + type: 'number', + description: 'MUST extract from gh pr create output - NOT from git push link', + }, + pr_url: { type: 'string' }, + merged: { type: 'boolean' }, + merge_commit_sha: { + type: 'string', + description: 'MUST extract from gh pr merge output', + }, + }, + required: ['pr_number', 'pr_url', 'merged', 'merge_commit_sha'], + }, }; } @@ -263,7 +473,7 @@ function generateGitPusherAgent(platform) { * @returns {string[]} Array of platform IDs */ function getSupportedPlatforms() { - return Object.keys(PLATFORM_CONFIGS); + return SUPPORTED_PLATFORMS; } /** @@ -272,7 +482,7 @@ function getSupportedPlatforms() { * @returns {boolean} */ function isPlatformSupported(platform) { - return platform in PLATFORM_CONFIGS; + return SUPPORTED_PLATFORMS.includes(platform); } module.exports = { @@ -281,5 +491,7 @@ module.exports = { isPlatformSupported, // Export for testing SHARED_TRIGGER_SCRIPT, - PLATFORM_CONFIGS, + SUPPORTED_PLATFORMS, + resolveGitHubConfig, + getPlatformConfig, }; diff --git a/src/attach/index.js b/src/attach/index.js index ccecfc5e..657846f7 100644 --- a/src/attach/index.js +++ b/src/attach/index.js @@ -25,6 +25,7 @@ const AttachClient = require('./attach-client'); const RingBuffer = require('./ring-buffer'); const protocol = require('./protocol'); const socketDiscovery = require('./socket-discovery'); +const { sendInput } = require('./send-input'); module.exports = { AttachServer, @@ -32,4 +33,5 @@ module.exports = { RingBuffer, protocol, socketDiscovery, + sendInput, }; diff --git a/src/attach/send-input.js b/src/attach/send-input.js new file mode 100644 index 00000000..481fe53b --- /dev/null +++ b/src/attach/send-input.js @@ -0,0 +1,88 @@ +/** + * sendInput - write stdin data to a live attach socket + * + * Uses the attach protocol's STDIN message to forward input to the PTY. + * Returns { ok, error } instead of throwing on transport failures. + */ + +const net = require('net'); +const fs = require('fs'); +const protocol = require('./protocol'); + +const DEFAULT_TIMEOUT_MS = 1500; + +/** + * Send input to an attach socket via STDIN message. + * @param {object} options + * @param {string} options.socketPath - Unix socket path + * @param {Buffer|string} options.data - Data to send + * @param {number} [options.timeoutMs=1500] - Timeout in ms + * @returns {Promise<{ok: boolean, error: string|null}>} + */ +function sendInput(options = {}) { + const { socketPath, data, timeoutMs = DEFAULT_TIMEOUT_MS } = options; + + if (!socketPath) { + throw new Error('sendInput: socketPath is required'); + } + + if (data === undefined || data === null) { + throw new Error('sendInput: data is required'); + } + + if (!Number.isFinite(timeoutMs) || timeoutMs <= 0) { + throw new Error(`sendInput: timeoutMs must be positive (got ${timeoutMs})`); + } + + if (!fs.existsSync(socketPath)) { + return { ok: false, error: `Socket not found: ${socketPath}` }; + } + + return new Promise((resolve) => { + let settled = false; + let timeout; + const socket = net.createConnection(socketPath); + + const finish = (result) => { + if (settled) return; + settled = true; + if (timeout) { + clearTimeout(timeout); + } + try { + socket.end(); + socket.destroy(); + } catch (cleanupError) { + console.warn('[sendInput] socket cleanup failed:', cleanupError); + } + resolve(result); + }; + + timeout = setTimeout(() => { + finish({ ok: false, error: 'Timeout waiting for socket connection' }); + }, timeoutMs); + + socket.on('connect', () => { + try { + const encoded = protocol.encode(protocol.createStdinMessage(data)); + socket.write(encoded, (err) => { + if (err) { + finish({ ok: false, error: err.message }); + } else { + finish({ ok: true, error: null }); + } + }); + } catch (err) { + finish({ ok: false, error: err.message }); + } + }); + + socket.on('error', (err) => { + finish({ ok: false, error: err.message }); + }); + }); +} + +module.exports = { + sendInput, +}; diff --git a/src/config-router.js b/src/config-router.js index f6618a1b..f269edae 100644 --- a/src/config-router.js +++ b/src/config-router.js @@ -37,7 +37,9 @@ function getConfig(complexity, taskType) { if (complexity === 'TRIVIAL') return 0; if (complexity === 'SIMPLE') return 1; if (complexity === 'STANDARD') return 2; - if (complexity === 'CRITICAL') return 4; + // CRITICAL uses two-stage validation pipeline (validators loaded dynamically via meta-coordinator) + // Setting validator_count=0 signals that inline validators should be skipped + if (complexity === 'CRITICAL') return 0; return 1; }; diff --git a/src/config-validator.js b/src/config-validator.js index 03ad2ae4..659fc470 100644 --- a/src/config-validator.js +++ b/src/config-validator.js @@ -15,6 +15,7 @@ const { loadSettings } = require('../lib/settings'); const { VALID_PROVIDERS, normalizeProviderName } = require('../lib/provider-names'); const { getProvider } = require('./providers'); const { CAPABILITIES } = require('./providers/capabilities'); +const { GUIDANCE_TOPICS } = require('./guidance-topics'); /** * Check if config is a conductor-bootstrap style config @@ -330,9 +331,11 @@ function buildMessageFlowGraph(config) { }; } -function reportMissingBootstrap(topicConsumers, errors) { +function reportMissingBootstrap(topicConsumers, errors, config) { const issueOpenedConsumers = topicConsumers.get('ISSUE_OPENED') || []; - if (issueOpenedConsumers.length === 0) { + const isSubTemplate = config.params && Object.keys(config.params).length > 0; + + if (issueOpenedConsumers.length === 0 && !isSubTemplate) { errors.push( 'No agent triggers on ISSUE_OPENED. Cluster will never start. ' + 'Add a trigger: { "topic": "ISSUE_OPENED", "action": "execute_task" }' @@ -382,9 +385,18 @@ function reportOrphanTopics(topicProducers, topicConsumers, warnings) { } } -function reportUnproducedTopics(topicConsumers, topicProducers, errors) { +function reportUnproducedTopics(topicConsumers, topicProducers, errors, config) { + const EXTERNAL_TOPICS = [ + 'ISSUE_OPENED', + 'CLUSTER_RESUMED', + 'QUICK_VALIDATION_PASSED', + 'IMPLEMENTATION_READY', + ...GUIDANCE_TOPICS, + ]; + const isSubTemplate = config.params && Object.keys(config.params).length > 0; + for (const [topic, consumers] of topicConsumers) { - if (topic === 'ISSUE_OPENED' || topic === 'CLUSTER_RESUMED') { + if (EXTERNAL_TOPICS.includes(topic)) { continue; } if (topic.endsWith('*')) { @@ -393,6 +405,9 @@ function reportUnproducedTopics(topicConsumers, topicProducers, errors) { const producers = topicProducers.get(topic) || []; if (producers.length === 0) { + if (isSubTemplate) { + continue; + } errors.push( `Topic '${topic}' consumed by [${consumers.join(', ')}] but never produced. ` + 'These agents will never trigger.' @@ -506,10 +521,10 @@ function analyzeMessageFlow(config) { const { topicProducers, topicConsumers, agentOutputTopics, agentInputTopics } = buildMessageFlowGraph(config); - reportMissingBootstrap(topicConsumers, errors); + reportMissingBootstrap(topicConsumers, errors, config); reportCompletionHandlers(config, errors, warnings); reportOrphanTopics(topicProducers, topicConsumers, warnings); - reportUnproducedTopics(topicConsumers, topicProducers, errors); + reportUnproducedTopics(topicConsumers, topicProducers, errors, config); reportSelfTriggeringAgents(config, agentInputTopics, agentOutputTopics, errors); reportTwoAgentCycles(config, agentInputTopics, agentOutputTopics, warnings); reportMissingValidationTriggers(config, errors); @@ -981,7 +996,7 @@ function validateHookAction(hook, prefix, errors) { if (!hook.action) { errors.push( `[Gap 1] ${prefix}: Missing 'action' field. ` + - `Fix: Add "action": "publish_message" or "action": "execute_system_command"` + `Fix: Add "action": "publish_message", "action": "execute_system_command", or "action": "verify_github_pr"` ); } } @@ -1203,7 +1218,7 @@ function validateRuleCoverage(config) { agent, 4, 'Model rules', - 'Add catch-all rule { "iterations": "all", "model": "sonnet" } or extend existing ranges.', + 'Add catch-all rule { "iterations": "all", "modelLevel": "level2" } or extend existing ranges.', uncoveredIterations ); } @@ -1459,9 +1474,14 @@ function validateJsonSchema(prefix, agent, errors) { function validateContextSource(prefix, source, topicProducers, errors, warnings) { const topic = source.topic; - if (topic === 'ISSUE_OPENED' || topic === 'CLUSTER_RESUMED') return; + if (topic === 'ISSUE_OPENED' || topic === 'CLUSTER_RESUMED' || topic === 'STATE_SNAPSHOT') { + return; + } if (topic.endsWith('*')) return; + const resolvedAmount = source.amount ?? source.limit; + const resolvedStrategy = source.strategy ?? (resolvedAmount !== undefined ? 'latest' : 'all'); + const producers = topicProducers.get(topic) || []; if (producers.length === 0) { warnings.push( @@ -1470,7 +1490,7 @@ function validateContextSource(prefix, source, topicProducers, errors, warnings) ); } - if (source.amount === undefined) { + if (resolvedAmount === undefined && resolvedStrategy !== 'all') { warnings.push( `[Gap 14] ${prefix}: Context source for topic '${topic}' missing 'amount' field. ` + `Defaults may not be what you expect.` @@ -1483,6 +1503,29 @@ function validateContextSource(prefix, source, topicProducers, errors, warnings) `Fix: Use 'latest', 'all', or 'oldest'.` ); } + + if (source.priority && !['required', 'high', 'medium', 'low'].includes(source.priority)) { + errors.push( + `[Gap 14] ${prefix}: Context source priority '${source.priority}' is invalid. ` + + `Fix: Use 'required', 'high', 'medium', or 'low'.` + ); + } + + if (source.compactStrategy && !['latest', 'all', 'oldest'].includes(source.compactStrategy)) { + errors.push( + `[Gap 14] ${prefix}: Context source compactStrategy '${source.compactStrategy}' is invalid. ` + + `Fix: Use 'latest', 'all', or 'oldest'.` + ); + } + + if (source.compactAmount !== undefined) { + if (!Number.isFinite(source.compactAmount) || source.compactAmount <= 0) { + errors.push( + `[Gap 14] ${prefix}: Context source compactAmount must be a positive number, got ${source.compactAmount}. ` + + `Fix: Use a positive integer like 1 or 3.` + ); + } + } } function validateContextSources(prefix, agent, topicProducers, errors, warnings) { diff --git a/src/guidance-topics.js b/src/guidance-topics.js new file mode 100644 index 00000000..476f99ba --- /dev/null +++ b/src/guidance-topics.js @@ -0,0 +1,10 @@ +const USER_GUIDANCE_CLUSTER = 'USER_GUIDANCE_CLUSTER'; +const USER_GUIDANCE_AGENT = 'USER_GUIDANCE_AGENT'; + +const GUIDANCE_TOPICS = [USER_GUIDANCE_CLUSTER, USER_GUIDANCE_AGENT]; + +module.exports = { + USER_GUIDANCE_CLUSTER, + USER_GUIDANCE_AGENT, + GUIDANCE_TOPICS, +}; diff --git a/src/isolation-manager.js b/src/isolation-manager.js index f978dbdd..6fe203d4 100644 --- a/src/isolation-manager.js +++ b/src/isolation-manager.js @@ -20,6 +20,7 @@ const { CLAUDE_AUTH_ENV_VARS, resolveClaudeAuth } = require('../lib/settings/cla const { normalizeProviderName } = require('../lib/provider-names'); const { resolveMounts, resolveEnvs, expandEnvPatterns } = require('../lib/docker-config'); const { getProvider } = require('./providers'); +const { readRepoSettings } = require('../lib/repo-settings'); /** * Escape a string for safe use in shell commands @@ -55,6 +56,7 @@ class IsolationManager { this.isolatedDirs = new Map(); // clusterId -> { path, originalDir } this.clusterConfigDirs = new Map(); // clusterId -> configDirPath this.worktrees = new Map(); // clusterId -> { path, branch, repoRoot } + this._exitWatchers = new Map(); // clusterId -> ChildProcess } /** @@ -125,7 +127,44 @@ class IsolationManager { args.push('-w', '/workspace', image, 'tail', '-f', '/dev/null'); - return this._spawnContainer(clusterId, args, workDir); + const containerId = await this._spawnContainer(clusterId, args, workDir); + this._watchContainerExit(clusterId, containerId, config.onExit); + return containerId; + } + + _watchContainerExit(clusterId, containerId, onExit) { + if (typeof onExit !== 'function') { + return; + } + + const existing = this._exitWatchers.get(clusterId); + if (existing) { + try { + existing.kill('SIGKILL'); + } catch { + // Ignore + } + this._exitWatchers.delete(clusterId); + } + + const proc = spawn('docker', ['wait', containerId], { stdio: ['ignore', 'pipe', 'ignore'] }); + this._exitWatchers.set(clusterId, proc); + + let stdout = ''; + proc.stdout.on('data', (data) => { + stdout += data.toString(); + }); + + const finalize = () => { + if (this._exitWatchers.get(clusterId) === proc) { + this._exitWatchers.delete(clusterId); + } + const code = parseInt(stdout.trim(), 10); + onExit({ clusterId, containerId, exitCode: Number.isFinite(code) ? code : null }); + }; + + proc.on('close', finalize); + proc.on('error', finalize); } _getRunningContainerId(clusterId) { @@ -1256,19 +1295,19 @@ class IsolationManager { /** * Create worktree-based isolation for a cluster (lightweight alternative to Docker) - * Creates a git worktree at {os.tmpdir()}/zeroshot-worktrees/{clusterId} + * Creates a git worktree at ~/.zeroshot/worktrees/{clusterId} * @param {string} clusterId - Cluster ID * @param {string} workDir - Original working directory (must be a git repo) * @returns {{ path: string, branch: string, repoRoot: string }} */ - createWorktreeIsolation(clusterId, workDir) { + createWorktreeIsolation(clusterId, workDir, options = {}) { if (!this._isGitRepo(workDir)) { throw new Error( `Worktree isolation requires a git repository. ${workDir} is not a git repo.` ); } - const worktreeInfo = this.createWorktree(clusterId, workDir); + const worktreeInfo = this.createWorktree(clusterId, workDir, options); this.worktrees.set(clusterId, worktreeInfo); console.log(`[IsolationManager] Created worktree isolation at ${worktreeInfo.path}`); @@ -1301,18 +1340,49 @@ class IsolationManager { * @param {string} workDir - Original working directory * @returns {{ path: string, branch: string, repoRoot: string }} */ - createWorktree(clusterId, workDir) { + createWorktree(clusterId, workDir, options = {}) { const repoRoot = this._getGitRoot(workDir); if (!repoRoot) { throw new Error(`Cannot find git root for ${workDir}`); } + // Priority: 1) options.baseRef, 2) repo settings, 3) HEAD (default) + let worktreeBaseRef = options.baseRef || null; + try { + const repoSettingsResult = readRepoSettings(repoRoot); + const repoSettings = repoSettingsResult.settings || {}; + const candidate = repoSettings.worktree?.baseRef; + if ( + !worktreeBaseRef && + typeof candidate === 'string' && + /^[A-Za-z0-9._/-]+$/.test(candidate.trim()) + ) { + worktreeBaseRef = candidate.trim(); + } + } catch { + // ignore + } + + // Best-effort ensure origin/ exists locally if requested. + if (worktreeBaseRef && worktreeBaseRef.startsWith('origin/')) { + const branch = worktreeBaseRef.slice('origin/'.length); + try { + execSync(`git fetch origin ${escapeShell(branch)}`, { + cwd: repoRoot, + encoding: 'utf8', + stdio: 'pipe', + }); + } catch { + // ignore + } + } + // Create branch name from cluster ID (e.g., cluster-cosmic-meteor-87 -> zeroshot/cosmic-meteor-87) const baseBranchName = `zeroshot/${clusterId.replace(/^cluster-/, '')}`; let branchName = baseBranchName; - // Worktree path in tmp - const worktreePath = path.join(os.tmpdir(), 'zeroshot-worktrees', clusterId); + // Worktree path in persistent location (survives reboots) + const worktreePath = path.join(os.homedir(), '.zeroshot', 'worktrees', clusterId); // Ensure parent directory exists const parentDir = path.dirname(worktreePath); @@ -1343,7 +1413,9 @@ class IsolationManager { // ignore } - // Create worktree with new branch based on HEAD (retry on branch collision/in-use) + const baseRef = worktreeBaseRef || 'HEAD'; + + // Create worktree with new branch based on baseRef (retry on branch collision/in-use) for (let attempt = 0; attempt < 10; attempt++) { // Best-effort delete if branch exists and is not in use by another worktree. try { @@ -1358,7 +1430,7 @@ class IsolationManager { try { execSync( - `git worktree add -b ${escapeShell(branchName)} ${escapeShell(worktreePath)} HEAD`, + `git worktree add -b ${escapeShell(branchName)} ${escapeShell(worktreePath)} ${escapeShell(baseRef)}`, { cwd: repoRoot, encoding: 'utf8', @@ -1402,6 +1474,23 @@ class IsolationManager { * @param {boolean} [options.deleteBranch=false] - Also delete the branch */ removeWorktree(worktreeInfo, _options = {}) { + // Tear down any Docker Compose services that may have been started in this worktree. + // Without this, containers keep running with host port mappings after the worktree is deleted, + // blocking port allocation for the main project or other worktrees. + const composePath = path.join(worktreeInfo.path, 'docker-compose.yml'); + if (fs.existsSync(composePath)) { + try { + execSync('docker compose down --remove-orphans --volumes --timeout 10', { + cwd: worktreeInfo.path, + encoding: 'utf8', + stdio: 'pipe', + timeout: 30000, + }); + } catch { + // Best-effort: compose project may not have been started, or Docker may not be running + } + } + // Remove the worktree (prefer git so metadata is cleaned up). try { execSync(`git worktree remove --force ${escapeShell(worktreeInfo.path)}`, { diff --git a/src/issue-providers/azure-devops-provider.js b/src/issue-providers/azure-devops-provider.js index 14e72008..16c45b0f 100644 --- a/src/issue-providers/azure-devops-provider.js +++ b/src/issue-providers/azure-devops-provider.js @@ -7,6 +7,8 @@ const IssueProvider = require('./base-provider'); const { execSync } = require('../lib/safe-exec'); const { detectGitContext } = require('../../lib/git-remote-utils'); +const AUTH_CHECK_TIMEOUT_MS = 2000; + class AzureDevOpsProvider extends IssueProvider { static id = 'azure-devops'; static displayName = 'Azure DevOps'; @@ -76,10 +78,33 @@ class AzureDevOpsProvider extends IssueProvider { static checkAuth() { try { // First check Azure login - execSync('az account show', { encoding: 'utf8', stdio: 'pipe' }); + execSync('az account show', { + encoding: 'utf8', + stdio: 'pipe', + timeout: AUTH_CHECK_TIMEOUT_MS, + }); } catch (err) { const stderr = err.stderr || err.message || ''; + if (err.code === 'ENOENT' || stderr.includes('command not found')) { + return { + authenticated: false, + error: 'Azure CLI not installed', + recovery: [ + 'Install Azure CLI: https://docs.microsoft.com/cli/azure/', + 'Then verify: az --version', + ], + }; + } + + if (stderr.includes('Command timed out')) { + return { + authenticated: false, + error: 'az account show timed out', + recovery: ['Retry: az account show', 'If it still hangs, run: az login'], + }; + } + if (stderr.includes('az login') || stderr.includes('not logged in')) { return { authenticated: false, @@ -104,6 +129,7 @@ class AzureDevOpsProvider extends IssueProvider { const output = execSync('az extension list --query "[?name==\'azure-devops\']" -o json', { encoding: 'utf8', stdio: 'pipe', + timeout: AUTH_CHECK_TIMEOUT_MS, }); const extensions = JSON.parse(output); if (extensions.length === 0) { @@ -116,7 +142,15 @@ class AzureDevOpsProvider extends IssueProvider { ], }; } - } catch { + } catch (err) { + const stderr = err?.stderr || err?.message || ''; + if (stderr.includes('Command timed out')) { + return { + authenticated: false, + error: 'az extension list timed out', + recovery: ['Retry: az extension list', 'Ensure az is configured and responsive'], + }; + } return { authenticated: false, error: 'Could not verify Azure DevOps extension', diff --git a/src/issue-providers/github-provider.js b/src/issue-providers/github-provider.js index 39f65952..8f55c370 100644 --- a/src/issue-providers/github-provider.js +++ b/src/issue-providers/github-provider.js @@ -5,6 +5,8 @@ const IssueProvider = require('./base-provider'); const { execSync } = require('../lib/safe-exec'); +const AUTH_CHECK_TIMEOUT_MS = 2000; + class GitHubProvider extends IssueProvider { static id = 'github'; static displayName = 'GitHub'; @@ -69,11 +71,31 @@ class GitHubProvider extends IssueProvider { */ static checkAuth() { try { - execSync('gh auth status', { encoding: 'utf8', stdio: 'pipe' }); + execSync('gh auth status', { + encoding: 'utf8', + stdio: 'pipe', + timeout: AUTH_CHECK_TIMEOUT_MS, + }); return { authenticated: true, error: null, recovery: [] }; } catch (err) { const stderr = err.stderr || err.message || ''; + if (err.code === 'ENOENT' || stderr.includes('command not found')) { + return { + authenticated: false, + error: 'gh CLI not installed', + recovery: ['Install gh: https://cli.github.com/', 'Then verify: gh --version'], + }; + } + + if (stderr.includes('Command timed out')) { + return { + authenticated: false, + error: 'gh auth status timed out', + recovery: ['Retry: gh auth status', 'If it still hangs, re-run: gh auth login'], + }; + } + if (stderr.includes('not logged in')) { return { authenticated: false, @@ -102,12 +124,13 @@ class GitHubProvider extends IssueProvider { return {}; } - fetchIssue(identifier, _settings) { + fetchIssue(identifier, _settings, gitContext = null) { try { - const issueNumber = this._extractIssueNumber(identifier); + const { repo, number } = this._parseIdentifier(identifier, gitContext); - // Fetch issue using gh CLI - const cmd = `gh issue view ${issueNumber} --json number,title,body,labels,assignees,comments,url`; + // ALWAYS use -R flag when repo is known - never rely on CWD git remote + const repoFlag = repo ? `-R ${repo}` : ''; + const cmd = `gh issue view ${number} ${repoFlag} --json number,title,body,labels,assignees,comments,url`; const output = execSync(cmd, { encoding: 'utf8' }); const issue = JSON.parse(output); @@ -118,24 +141,32 @@ class GitHubProvider extends IssueProvider { } /** - * Extract issue number from URL or return as-is + * Parse identifier into repo and issue number * @private + * @returns {{ repo: string|null, number: string }} */ - _extractIssueNumber(issueRef) { - // If it's a URL, extract the number - const urlMatch = issueRef.match(/\/issues\/(\d+)/); + _parseIdentifier(identifier, gitContext = null) { + // GitHub URL: https://github.com/org/repo/issues/123 + const urlMatch = identifier.match(/github\.com\/([^/]+\/[^/]+)\/issues\/(\d+)/); if (urlMatch) { - return urlMatch[1]; + return { repo: urlMatch[1], number: urlMatch[2] }; } // org/repo#123 format - const repoMatch = issueRef.match(/^[\w-]+\/[\w-]+#(\d+)$/); + const repoMatch = identifier.match(/^([\w.-]+\/[\w.-]+)#(\d+)$/); if (repoMatch) { - return repoMatch[1]; + return { repo: repoMatch[1], number: repoMatch[2] }; + } + + // Bare number - use gitContext if available + if (/^\d+$/.test(identifier)) { + const repo = + gitContext?.owner && gitContext?.repo ? `${gitContext.owner}/${gitContext.repo}` : null; + return { repo, number: identifier }; } - // Otherwise assume it's already a number - return issueRef; + // Fallback: assume it's a number, no repo + return { repo: null, number: identifier }; } /** diff --git a/src/issue-providers/gitlab-provider.js b/src/issue-providers/gitlab-provider.js index 1f534981..c9bddae5 100644 --- a/src/issue-providers/gitlab-provider.js +++ b/src/issue-providers/gitlab-provider.js @@ -6,6 +6,8 @@ const IssueProvider = require('./base-provider'); const { execSync } = require('../lib/safe-exec'); +const AUTH_CHECK_TIMEOUT_MS = 2000; + class GitLabProvider extends IssueProvider { static id = 'gitlab'; static displayName = 'GitLab'; @@ -89,11 +91,30 @@ class GitLabProvider extends IssueProvider { const cmd = hostname ? `glab auth status --hostname ${hostname}` : 'glab auth status'; try { - execSync(cmd, { encoding: 'utf8', stdio: 'pipe' }); + execSync(cmd, { encoding: 'utf8', stdio: 'pipe', timeout: AUTH_CHECK_TIMEOUT_MS }); return { authenticated: true, error: null, recovery: [] }; } catch (err) { const stderr = err.stderr || err.message || ''; + if (err.code === 'ENOENT' || stderr.includes('command not found')) { + return { + authenticated: false, + error: 'glab CLI not installed', + recovery: [ + 'Install glab: https://gitlab.com/gitlab-org/cli', + 'Then verify: glab --version', + ], + }; + } + + if (stderr.includes('Command timed out')) { + return { + authenticated: false, + error: 'glab auth status timed out', + recovery: ['Retry: glab auth status', 'If it still hangs, run: glab auth login'], + }; + } + const hostnameHint = hostname || 'your GitLab instance'; if ( stderr.includes('not logged in') || diff --git a/src/issue-providers/jira-provider.js b/src/issue-providers/jira-provider.js index f8798d34..b5de25b6 100644 --- a/src/issue-providers/jira-provider.js +++ b/src/issue-providers/jira-provider.js @@ -6,6 +6,8 @@ const IssueProvider = require('./base-provider'); const { execSync } = require('../lib/safe-exec'); +const AUTH_CHECK_TIMEOUT_MS = 2000; + class JiraProvider extends IssueProvider { static id = 'jira'; static displayName = 'Jira'; @@ -71,11 +73,30 @@ class JiraProvider extends IssueProvider { try { // go-jira uses 'jira session' to verify authentication // If not configured, it fails with endpoint/login errors - execSync('jira session', { encoding: 'utf8', stdio: 'pipe', timeout: 5000 }); + execSync('jira session', { encoding: 'utf8', stdio: 'pipe', timeout: AUTH_CHECK_TIMEOUT_MS }); return { authenticated: true, error: null, recovery: [] }; } catch (err) { const stderr = err.stderr || err.message || ''; + if (err.code === 'ENOENT' || stderr.includes('command not found')) { + return { + authenticated: false, + error: 'Jira CLI not installed', + recovery: [ + 'Install jira CLI: https://github.com/go-jira/jira', + 'Then verify: jira version', + ], + }; + } + + if (stderr.includes('Command timed out')) { + return { + authenticated: false, + error: 'jira session timed out', + recovery: ['Retry: jira session', 'Check ~/.jira.d/config.yml for endpoint/login'], + }; + } + // go-jira has various error patterns for auth issues if ( stderr.includes('endpoint') || diff --git a/src/ledger.js b/src/ledger.js index df64b6ef..bf25695a 100644 --- a/src/ledger.js +++ b/src/ledger.js @@ -11,6 +11,11 @@ const Database = require('better-sqlite3'); const EventEmitter = require('events'); const crypto = require('crypto'); +const { + GUIDANCE_TOPICS, + USER_GUIDANCE_AGENT, + USER_GUIDANCE_CLUSTER, +} = require('./guidance-topics'); class Ledger extends EventEmitter { constructor(dbPath = ':memory:') { @@ -114,12 +119,13 @@ class Ledger extends EventEmitter { const timestamp = requestedTimestamp !== null ? Math.max(requestedTimestamp, baseTimestamp) : baseTimestamp; + const receiver = message.receiver || message.target_agent_id || 'broadcast'; const record = { id, timestamp, topic: message.topic, sender: message.sender, - receiver: message.receiver || 'broadcast', + receiver, content_text: message.content?.text || null, content_data: message.content?.data ? JSON.stringify(message.content.data) : null, metadata: message.metadata ? JSON.stringify(message.metadata) : null, @@ -188,12 +194,13 @@ class Ledger extends EventEmitter { // Use incrementing timestamps to preserve order within batch const timestamp = baseTimestamp + i; + const receiver = message.receiver || message.target_agent_id || 'broadcast'; const record = { id, timestamp, topic: message.topic, sender: message.sender, - receiver: message.receiver || 'broadcast', + receiver, content_text: message.content?.text || null, content_data: message.content?.data ? JSON.stringify(message.content.data) : null, metadata: message.metadata ? JSON.stringify(message.metadata) : null, @@ -304,6 +311,60 @@ class Ledger extends EventEmitter { return rows.map((row) => this._deserializeMessage(row)); } + /** + * Query guidance mailbox for cluster-wide + agent-specific guidance + * @param {Object} criteria - { cluster_id, target_agent_id, lastDeliveredAt, limit } + * @returns {Array} Guidance messages ordered by timestamp ASC + */ + queryGuidanceMailbox(criteria) { + const { cluster_id, target_agent_id, lastDeliveredAt, limit } = criteria || {}; + + if (!cluster_id) { + throw new Error('cluster_id is required for guidance mailbox queries'); + } + + const guidanceTopics = new Set(GUIDANCE_TOPICS); + if (!guidanceTopics.has(USER_GUIDANCE_CLUSTER) || !guidanceTopics.has(USER_GUIDANCE_AGENT)) { + throw new Error('GUIDANCE_TOPICS must include USER_GUIDANCE_CLUSTER and USER_GUIDANCE_AGENT'); + } + + let sinceTimestamp = null; + if (lastDeliveredAt !== undefined && lastDeliveredAt !== null) { + const candidate = + typeof lastDeliveredAt === 'number' ? lastDeliveredAt : new Date(lastDeliveredAt).getTime(); + if (!Number.isFinite(candidate)) { + throw new Error('lastDeliveredAt must be a number or valid date'); + } + sinceTimestamp = candidate; + } + + const params = [cluster_id, USER_GUIDANCE_CLUSTER]; + let sql = 'SELECT * FROM messages WHERE cluster_id = ? AND (topic = ?'; + + if (target_agent_id) { + params.push(USER_GUIDANCE_AGENT, target_agent_id); + sql += ' OR (topic = ? AND receiver = ?)'; + } + + sql += ')'; + + if (sinceTimestamp !== null) { + params.push(sinceTimestamp); + sql += ' AND timestamp > ?'; + } + + sql += ' ORDER BY timestamp ASC'; + + if (limit) { + params.push(limit); + sql += ' LIMIT ?'; + } + + const stmt = this.db.prepare(sql); + const rows = stmt.all(...params); + return rows.map((row) => this._deserializeMessage(row)); + } + /** * Find the last message matching criteria * @param {Object} criteria - Query criteria diff --git a/src/message-buffer.js b/src/message-buffer.js new file mode 100644 index 00000000..626cf13f --- /dev/null +++ b/src/message-buffer.js @@ -0,0 +1,81 @@ +/** + * Message buffering helper + * + * Ensures trigger-matching messages are never dropped just because an agent/subcluster is busy. + * Dropped workflow signals (e.g. VALIDATION_RESULT) can wedge clusters in "running" state. + */ + +function bufferMessage(target, message, options = {}) { + const maxBuffered = options.maxBuffered ?? 200; + + if (!target._bufferedMessages) { + target._bufferedMessages = []; + } + + if (target._bufferedMessages.length >= maxBuffered) { + target._bufferedMessages.shift(); + } + + target._bufferedMessages.push(message); +} + +function scheduleDrain(target, drainFn, options = {}) { + if (target._bufferDrainScheduled) { + return; + } + + target._bufferDrainScheduled = true; + + const label = options.label || 'MessageBuffer'; + const id = target.id || 'unknown'; + + const run = () => { + target._bufferDrainScheduled = false; + drainFn().catch((error) => { + console.error(`\n${'='.repeat(80)}`); + console.error(`šŸ”“ FATAL: ${label} drain crashed (${id})`); + console.error(`${'='.repeat(80)}`); + console.error(`Error: ${error.message}`); + console.error(`Stack: ${error.stack}`); + console.error(`${'='.repeat(80)}\n`); + setImmediate(() => { + throw error; + }); + }); + }; + + const current = target._currentExecution; + if (current && typeof current.finally === 'function') { + current.finally(() => setImmediate(run)); + return; + } + + setImmediate(run); +} + +async function drainBufferedMessages(target, handleFn, options = {}) { + if (!target.running) { + return; + } + + const buffer = target._bufferedMessages; + if (!buffer || buffer.length === 0) { + return; + } + + if (target.state !== 'idle') { + scheduleDrain(target, () => drainBufferedMessages(target, handleFn, options), options); + return; + } + + while (target.running && target.state === 'idle' && buffer.length > 0) { + const next = buffer.shift(); + await handleFn(next); + } +} + +module.exports = { + bufferMessage, + scheduleDrain, + drainBufferedMessages, +}; diff --git a/src/message-bus-bridge.js b/src/message-bus-bridge.js index 7525377e..4360ec4e 100644 --- a/src/message-bus-bridge.js +++ b/src/message-bus-bridge.js @@ -16,6 +16,11 @@ class MessageBusBridge { this.parentBus = parentBus; this.childBus = childBus; this.config = config; + this.parentTopicNames = new Set( + (config.parentTopics || []) + .map((entry) => (typeof entry === 'string' ? entry : entry?.topic)) + .filter((topic) => typeof topic === 'string' && topic.length > 0) + ); this.parentUnsubscribe = null; this.childUnsubscribe = null; @@ -30,7 +35,7 @@ class MessageBusBridge { */ _setupBridge() { // Forward specified parent topics to child - if (this.config.parentTopics && this.config.parentTopics.length > 0) { + if (this.parentTopicNames.size > 0) { this.parentUnsubscribe = this.parentBus.subscribe((message) => { this._forwardParentToChild(message); }); @@ -55,7 +60,7 @@ class MessageBusBridge { } // Only forward topics specified in config - if (!this.config.parentTopics.includes(message.topic)) { + if (!this.parentTopicNames.has(message.topic)) { return; } diff --git a/src/message-bus.js b/src/message-bus.js index 61afec64..3f0dc104 100644 --- a/src/message-bus.js +++ b/src/message-bus.js @@ -103,6 +103,13 @@ class MessageBus extends EventEmitter { return this.ledger.query(criteria); } + /** + * Query guidance mailbox (passthrough to ledger) + */ + queryGuidanceMailbox(criteria) { + return this.ledger.queryGuidanceMailbox(criteria); + } + /** * Find last message (passthrough to ledger) */ diff --git a/src/orchestrator.js b/src/orchestrator.js index e56a26fe..f32fa6c0 100644 --- a/src/orchestrator.js +++ b/src/orchestrator.js @@ -38,6 +38,7 @@ const SubClusterWrapper = require('./sub-cluster-wrapper'); const MessageBus = require('./message-bus'); const Ledger = require('./ledger'); const InputHelpers = require('./input-helpers'); +const { USER_GUIDANCE_AGENT, USER_GUIDANCE_CLUSTER } = require('./guidance-topics'); const { detectProvider } = require('./issue-providers'); const IsolationManager = require('./isolation-manager'); const { generateName } = require('./name-generator'); @@ -45,6 +46,8 @@ const configValidator = require('./config-validator'); const TemplateResolver = require('./template-resolver'); const { loadSettings } = require('../lib/settings'); const { normalizeProviderName } = require('../lib/provider-names'); +const { getProvider } = require('./providers'); +const StateSnapshotter = require('./state-snapshotter'); const crypto = require('crypto'); function applyModelOverride(agentConfig, modelOverride) { @@ -131,6 +134,38 @@ class Orchestrator { } } + /** + * Resolve provider for a cluster config using the standard precedence. + * @param {Object} clusterConfig + * @param {Object} settings + * @returns {string} + * @private + */ + _resolveClusterProvider(clusterConfig = {}, settings = loadSettings()) { + const resolved = + clusterConfig.forceProvider || + clusterConfig.defaultProvider || + settings.defaultProvider || + 'claude'; + return normalizeProviderName(resolved) || 'claude'; + } + + /** + * Resolve the model level for internal completion agents. + * Uses provider minLevel when configured, otherwise provider default min level. + * @param {Object} clusterConfig + * @returns {string} + * @private + */ + _resolveCompletionDetectorLevel(clusterConfig = {}) { + const settings = loadSettings(); + const providerName = this._resolveClusterProvider(clusterConfig, settings); + const provider = getProvider(providerName); + const providerSettings = settings.providerSettings?.[providerName] || {}; + + return providerSettings.minLevel || provider.getDefaultMinLevel(); + } + /** * Get input source type for metadata * @param {Object} input - Input object @@ -270,29 +305,40 @@ class Orchestrator { // Restore isolation manager FIRST if cluster was running in isolation mode const { isolation, isolationManager } = this._restoreClusterIsolation(clusterId, clusterData); + // Create mutable cluster context for reload path (used by AgentWrapper/SubClusterWrapper) + const clusterContext = { + ...clusterData, + id: clusterId, + isolation, + }; + // Reconstruct agent metadata from config (processes are ephemeral) // CRITICAL: Pass isolation context to agents if cluster was running in isolation const agents = this._rebuildClusterAgents( - clusterId, - clusterData, + clusterContext, messageBus, isolation, isolationManager ); const cluster = { - ...clusterData, + ...clusterContext, ledger, messageBus, agents, isolation, autoPr: clusterData.autoPr || false, + prOptions: clusterData.prOptions || null, + issue: clusterData.issue || null, }; - this.clusters.set(clusterId, cluster); + Object.assign(clusterContext, cluster); + + this.clusters.set(clusterId, clusterContext); + this._startSnapshotter(clusterContext); this._log(`[Orchestrator] Loaded cluster: ${clusterId} with ${agents.length} agents`); - return cluster; + return clusterContext; } _restoreClusterIsolation(clusterId, clusterData) { @@ -346,12 +392,12 @@ class Orchestrator { return agentOptions; } - _instantiateAgent(agentConfig, messageBus, clusterId, agentOptions) { + _instantiateAgent(agentConfig, messageBus, clusterContext, agentOptions) { if (agentConfig.type === 'subcluster') { - return new SubClusterWrapper(agentConfig, messageBus, { id: clusterId }, agentOptions); + return new SubClusterWrapper(agentConfig, messageBus, clusterContext, agentOptions); } - return new AgentWrapper(agentConfig, messageBus, { id: clusterId }, agentOptions); + return new AgentWrapper(agentConfig, messageBus, clusterContext, agentOptions); } _restoreAgentState(agent, agentConfig, clusterData) { @@ -367,8 +413,10 @@ class Orchestrator { agent.processPid = savedState.processPid || null; } - _rebuildClusterAgents(clusterId, clusterData, messageBus, isolation, isolationManager) { + _rebuildClusterAgents(clusterContext, messageBus, isolation, isolationManager) { const agents = []; + const clusterId = clusterContext.id; + const clusterData = clusterContext; const agentCwd = this._resolveAgentCwd(clusterData); if (!clusterData.config?.agents) { @@ -391,7 +439,7 @@ class Orchestrator { isolation, isolationManager ); - const agent = this._instantiateAgent(agentConfig, messageBus, clusterId, agentOptions); + const agent = this._instantiateAgent(agentConfig, messageBus, clusterContext, agentOptions); this._restoreAgentState(agent, agentConfig, clusterData); agents.push(agent); } @@ -399,11 +447,36 @@ class Orchestrator { return agents; } + _startSnapshotter(cluster) { + if (cluster.snapshotter) { + cluster.snapshotter.start(); + return; + } + + const snapshotter = new StateSnapshotter({ + messageBus: cluster.messageBus, + clusterId: cluster.id, + }); + snapshotter.start(); + cluster.snapshotter = snapshotter; + } + /** * Ensure clusters file exists (required for file locking) * @private */ _ensureClustersFile() { + if (!fs.existsSync(this.storageDir)) { + try { + fs.mkdirSync(this.storageDir, { recursive: true }); + } catch (error) { + console.warn( + `[Orchestrator] Failed to create storage directory ${this.storageDir}: ${error.message}` + ); + return null; + } + } + const clustersFile = path.join(this.storageDir, 'clusters.json'); if (!fs.existsSync(clustersFile)) { fs.writeFileSync(clustersFile, '{}'); @@ -411,6 +484,51 @@ class Orchestrator { return clustersFile; } + /** + * Find active clusters for a given issue number + * Used to prevent duplicate runs on the same issue + * @param {number|string} issueNumber - Issue number to check + * @returns {Array<{id: string, state: string, createdAt: number}>} Active clusters for this issue + * @private + */ + _getActiveClustersForIssue(issueNumber, excludeClusterId = null) { + const activeClusters = []; + const issueNum = Number(issueNumber); + + for (const [clusterId, cluster] of this.clusters) { + if (excludeClusterId && clusterId === excludeClusterId) continue; + // Skip clusters without issue numbers + if (!cluster.issue) continue; + + // Check if same issue number + if (Number(cluster.issue) !== issueNum) continue; + + // Check if cluster is still active (not completed/failed/stopped) + const inactiveStates = ['completed', 'failed', 'stopped', 'corrupted']; + if (inactiveStates.includes(cluster.state)) continue; + + // Check if process is still running (zombie detection) + if (cluster.pid) { + try { + // process.kill with signal 0 checks if process exists + process.kill(cluster.pid, 0); + // Process exists - cluster is active + activeClusters.push({ + id: clusterId, + state: cluster.state, + createdAt: cluster.createdAt, + pid: cluster.pid, + }); + } catch { + // Process doesn't exist - cluster is a zombie (stale entry) + // Don't include in active list + } + } + } + + return activeClusters; + } + /** * Save clusters to persistent storage * Uses file locking to prevent race conditions with other processes @@ -423,6 +541,9 @@ class Orchestrator { } const clustersFile = this._ensureClustersFile(); + if (!clustersFile) { + return; + } const lockfilePath = path.join(this.storageDir, 'clusters.json.lock'); let release; @@ -483,8 +604,12 @@ class Orchestrator { failureInfo: cluster.failureInfo || null, // Persist PR mode for completion agent selection autoPr: cluster.autoPr || false, + // Persist PR options for resume + prOptions: cluster.prOptions || null, // Persist model override for consistent agent spawning on resume modelOverride: cluster.modelOverride || null, + // Persist issue number for heroshot/external tools + issue: cluster.issue || null, // Persist isolation info (excluding manager instance which can't be serialized) // CRITICAL: workDir is required for resume() to recreate container with same workspace isolation: cluster.isolation @@ -514,6 +639,14 @@ class Orchestrator { this._log( `[Orchestrator] Saved ${this.clusters.size} cluster(s), file now has ${Object.keys(existingClusters).length} total` ); + } catch (error) { + if (error.code === 'ENOENT') { + console.warn( + `[Orchestrator] Skipping cluster save; storage directory missing: ${this.storageDir}` + ); + return; + } + throw error; } finally { // Always release lock if (release) { @@ -632,17 +765,20 @@ class Orchestrator { * @returns {Object} Cluster object */ start(config, input = {}, options = {}) { + const testMode = options.testMode || !!this.taskRunner; + const autoPr = options.autoPr ?? (testMode ? false : process.env.ZEROSHOT_PR === '1'); return this._startInternal(config, input, { - testMode: false, + testMode, cwd: options.cwd || process.cwd(), // Target working directory for agents isolation: options.isolation || false, isolationImage: options.isolationImage, worktree: options.worktree || false, - autoPr: options.autoPr || process.env.ZEROSHOT_PR === '1', + autoPr, modelOverride: options.modelOverride, // Model override for all agents clusterId: options.clusterId, // Explicit ID from CLI/daemon parent settings: options.settings, // User settings for issue provider detection forceProvider: options.forceProvider, // Force specific issue provider + force: options.force || false, // Skip duplicate issue check }); } @@ -695,6 +831,15 @@ class Orchestrator { initCompletePromise, _resolveInitComplete: resolveInitComplete, autoPr: options.autoPr || false, + // PR configuration options (persisted for resume) + prOptions: + options.prBase || options.mergeQueue || options.closeIssue + ? { + prBase: options.prBase || null, + mergeQueue: options.mergeQueue || false, + closeIssue: options.closeIssue || null, + } + : null, // Model override for all agents (applied to dynamically added agents) modelOverride: options.modelOverride || null, // Issue provider tracking (where issue was fetched from) @@ -726,6 +871,14 @@ class Orchestrator { }; this.clusters.set(clusterId, cluster); + this._startSnapshotter(cluster); + + // Persist the cluster immediately so detached runs can't create "invisible" initializing clusters. + // Without this, an early startup failure (before ISSUE_OPENED / TASK_STARTED) may never be saved, + // and external supervisors (e.g., heroshot) can't detect/cleanup the stuck state. + await this._saveClusters().catch((err) => { + console.warn(`[Orchestrator] Failed to persist initial cluster state for ${clusterId}:`, err); + }); try { // Fetch input (issue from provider, file, or text) @@ -747,6 +900,31 @@ class Orchestrator { // Store issue provider for logging/debugging and cross-provider workflows cluster.issueProvider = ProviderClass.id; + // Store issue number for heroshot/external tools (avoids log parsing) + cluster.issue = inputData.number || null; + + // Persist issue number early so supervisors can associate this cluster with the issue even if start fails. + await this._saveClusters().catch((err) => { + console.warn(`[Orchestrator] Failed to persist issue number for ${clusterId}:`, err); + }); + + // Check for duplicate active runs on same issue (unless --force) + if (cluster.issue && !options.force) { + const activeClusters = this._getActiveClustersForIssue(cluster.issue, clusterId); + if (activeClusters.length > 0) { + const existing = activeClusters[0]; + const age = Math.round((Date.now() - existing.createdAt) / 1000 / 60); + throw new Error( + `Issue #${cluster.issue} already has an active cluster:\n` + + ` Cluster: ${existing.id} (state: ${existing.state}, running for ${age}min, pid: ${existing.pid})\n\n` + + `Options:\n` + + ` 1. Kill existing: zeroshot kill ${existing.id}\n` + + ` 2. Override check: zeroshot run ${input.issue} --force\n` + + ` 3. View status: zeroshot status ${existing.id}` + ); + } + } + // Log clickable issue link if (inputData.url) { this._log(`[Orchestrator] Issue (${ProviderClass.displayName}): ${inputData.url}`); @@ -879,6 +1057,58 @@ class Orchestrator { if (cluster._resolveInitComplete) { cluster._resolveInitComplete(); } + // Persist the failure state (prevents "invisible" failures that supervisors can't detect/cleanup). + await this._saveClusters().catch((err) => { + console.warn( + `[Orchestrator] Failed to persist failed cluster state for ${clusterId}:`, + err + ); + }); + + // Best-effort cleanup of partially initialized resources (prevents orphaned worktrees/containers). + try { + if (cluster.snapshotter) { + cluster.snapshotter.stop(); + } + } catch { + // ignore + } + try { + for (const agent of cluster.agents || []) { + // Agent start may have partially succeeded. + // Stop is best-effort and must not mask the original error. + // eslint-disable-next-line no-await-in-loop + await agent.stop(); + } + } catch { + // ignore + } + try { + if (cluster.isolation?.manager) { + // Preserve workspace for inspection; callers can `zeroshot kill` for full cleanup. + // eslint-disable-next-line no-await-in-loop + await cluster.isolation.manager.cleanup(clusterId, { preserveWorkspace: true }); + } + } catch { + // ignore + } + try { + if (cluster.worktree?.manager) { + cluster.worktree.manager.cleanupWorktreeIsolation(clusterId, { preserveBranch: true }); + } + } catch { + // ignore + } + try { + cluster.messageBus?.close?.(); + } catch { + // ignore + } + try { + cluster.ledger?.close?.(); + } catch { + // ignore + } console.error(`Cluster ${clusterId} failed to start:`, error); throw error; } @@ -980,16 +1210,27 @@ class Orchestrator { this._subscribeToClusterTopic(messageBus, clusterId, 'AGENT_ERROR', async (message) => { const agentRole = message.content?.data?.role; const attempts = message.content?.data?.attempts || 1; + const hookFailure = message.content?.data?.hookFailure === true; await this._saveClusters(); - if (agentRole === 'implementation' && attempts >= 3) { + const shouldStopForRole = + agentRole === 'implementation' || + agentRole === 'coordinator' || + message.sender === 'consensus-coordinator'; + const shouldStop = shouldStopForRole && (hookFailure || attempts >= 3); + + if (shouldStop) { this._log(`\n${'='.repeat(80)}`); - this._log(`āŒ WORKER AGENT FAILED: ${clusterId}`); + this._log(`āŒ CRITICAL AGENT FAILED: ${clusterId}`); this._log(`${'='.repeat(80)}`); - this._log(`Worker agent ${message.sender} failed after ${attempts} attempts`); + this._log( + `${message.sender} (${agentRole || 'unknown role'}) failed` + + (hookFailure ? ` (hookFailure=true)` : ``) + + ` after ${attempts} attempts` + ); this._log(`Error: ${message.content?.data?.error || 'unknown'}`); - this._log(`Stopping cluster - worker cannot continue`); + this._log(`Stopping cluster - critical agent cannot continue`); this._log(`${'='.repeat(80)}\n`); this.stop(clusterId).catch((err) => { @@ -1194,9 +1435,7 @@ class Orchestrator { this._log(`[Orchestrator] Starting cluster in isolation mode (image: ${image})`); const workDir = options.cwd || process.cwd(); - const providerName = normalizeProviderName( - config.forceProvider || config.defaultProvider || loadSettings().defaultProvider || 'claude' - ); + const providerName = this._resolveClusterProvider(config); containerId = await isolationManager.createContainer(clusterId, { workDir, image, @@ -1210,7 +1449,13 @@ class Orchestrator { const workDir = options.cwd || process.cwd(); isolationManager = new IsolationManager({}); - worktreeInfo = isolationManager.createWorktreeIsolation(clusterId, workDir); + // Use origin/${prBase} if prBase is set (ensures worktree is up-to-date with remote) + const worktreeOptions = {}; + if (options.prBase) { + worktreeOptions.baseRef = `origin/${options.prBase}`; + this._log(`[Orchestrator] Using remote base ref: origin/${options.prBase}`); + } + worktreeInfo = isolationManager.createWorktreeIsolation(clusterId, workDir, worktreeOptions); this._log(`[Orchestrator] Starting cluster in worktree isolation mode`); this._log(`[Orchestrator] Worktree: ${worktreeInfo.path}`); @@ -1271,7 +1516,11 @@ class Orchestrator { ); } - const gitPusherConfig = generateGitPusherAgent(platform); + const gitPusherConfig = generateGitPusherAgent(platform, { + prBase: options.prBase, + mergeQueue: options.mergeQueue, + closeIssue: options.closeIssue, + }); // Template replacement for issue context const issueRef = skipCloseIssue ? '' : `Closes #${inputData.number || 'unknown'}`; @@ -1359,6 +1608,114 @@ class Orchestrator { throw new Error('Failed to generate unique cluster ID after many attempts'); } + /** + * Wait for a process to exit + * @param {Number} pid - Process ID + * @param {Number} timeoutMs - Timeout in milliseconds + * @param {Number} intervalMs - Poll interval in milliseconds + * @returns {Promise} True if process exited + * @private + */ + async _waitForProcessExit(pid, timeoutMs, intervalMs = 100) { + const deadline = Date.now() + timeoutMs; + while (Date.now() < deadline) { + if (!this._isProcessRunning(pid)) { + return true; + } + await new Promise((resolve) => setTimeout(resolve, intervalMs)); + } + return !this._isProcessRunning(pid); + } + + /** + * Tear down Docker Compose services in a worktree directory to free host ports. + * Best-effort ā€” silently ignores failures (compose may not have been started, Docker may not be running). + * @param {string} worktreePath - Path to the worktree directory + * @private + */ + _teardownWorktreeCompose(worktreePath) { + const { execSync } = require('./lib/safe-exec'); + const composePath = path.join(worktreePath, 'docker-compose.yml'); + if (!fs.existsSync(composePath)) return; + + try { + this._log(`[Orchestrator] Tearing down Docker Compose services in ${worktreePath}...`); + execSync('docker compose down --remove-orphans --volumes --timeout 10', { + cwd: worktreePath, + encoding: 'utf8', + stdio: 'pipe', + timeout: 30000, + }); + this._log(`[Orchestrator] Docker Compose services torn down`); + } catch { + // Best-effort: compose project may not have been started + } + } + + /** + * Signal a remote daemon that owns the cluster and wait for exit + * @param {Object} cluster - Cluster object + * @param {Object} options - { action, timeoutMs, killTimeoutMs, signal } + * @returns {Promise} { handled, remotePid, exited, forced } + * @private + */ + async _signalRemoteCluster(cluster, options) { + const remotePid = cluster.pid; + if (!remotePid || remotePid === process.pid) { + return { handled: false }; + } + + if (!this._isProcessRunning(remotePid)) { + return { handled: false, remotePid, alreadyExited: true }; + } + + const action = options?.action || 'stop'; + const timeoutMs = options?.timeoutMs ?? 10000; + const killTimeoutMs = options?.killTimeoutMs ?? 5000; + const signal = options?.signal || 'SIGTERM'; + + this._log(`[Orchestrator] Sending ${signal} to daemon PID ${remotePid} for ${cluster.id}...`); + try { + process.kill(remotePid, signal); + } catch (error) { + if (error.code === 'ESRCH') { + return { handled: true, remotePid, exited: true }; + } + throw error; + } + + const exited = await this._waitForProcessExit(remotePid, timeoutMs); + if (exited) { + return { handled: true, remotePid, exited: true }; + } + + if (action !== 'kill') { + throw new Error( + `Timed out waiting for daemon PID ${remotePid} to stop cluster ${cluster.id}` + ); + } + + this._log( + `[Orchestrator] Daemon PID ${remotePid} still running after ${timeoutMs}ms, sending SIGKILL...` + ); + try { + process.kill(remotePid, 'SIGKILL'); + } catch (error) { + if (error.code !== 'ESRCH') { + throw error; + } + } + + const killed = await this._waitForProcessExit(remotePid, killTimeoutMs); + if (!killed) { + throw new Error( + `Failed to kill daemon PID ${remotePid} for cluster ${cluster.id} after SIGKILL` + ); + } + + return { handled: true, remotePid, exited: true, forced: true }; + } + /** * Stop a cluster * @param {String} clusterId - Cluster ID @@ -1369,6 +1726,8 @@ class Orchestrator { throw new Error(`Cluster ${clusterId} not found`); } + await this._signalRemoteCluster(cluster, { action: 'stop' }); + // CRITICAL: Wait for initialization to complete before stopping // This ensures ISSUE_OPENED is published, preventing 0-message clusters // Timeout after 30s to prevent infinite hang if init truly fails @@ -1387,6 +1746,10 @@ class Orchestrator { await agent.stop(); } + if (cluster.snapshotter) { + cluster.snapshotter.stop(); + } + // Clean up isolation container if enabled // CRITICAL: Preserve workspace for resume capability - only delete on kill() if (cluster.isolation?.manager) { @@ -1397,12 +1760,26 @@ class Orchestrator { this._log(`[Orchestrator] Container stopped, workspace preserved`); } + if (cluster.validatorIsolation?.manager) { + this._log(`[Orchestrator] Cleaning up validator isolation container for ${clusterId}...`); + await cluster.validatorIsolation.manager.cleanup(cluster.validatorIsolation.clusterId, { + preserveWorkspace: false, + }); + cluster.validatorIsolation = null; + } + // Worktree cleanup on stop: preserve for resume capability // Branch stays, worktree stays - can resume work later + // BUT: tear down Docker Compose services to free host ports if (cluster.worktree?.manager) { this._log(`[Orchestrator] Worktree preserved at ${cluster.worktree.path} for resume`); this._log(`[Orchestrator] Branch: ${cluster.worktree.branch}`); - // Don't cleanup worktree - it will be reused on resume + // Tear down Docker Compose services in the worktree to free host ports. + // Without this, stopped worktrees hold ports (5433, 6379, etc.) indefinitely. + if (cluster.worktree.path) { + this._teardownWorktreeCompose(cluster.worktree.path); + } + // Don't cleanup worktree itself - it will be reused on resume } cluster.state = 'stopped'; @@ -1423,6 +1800,8 @@ class Orchestrator { throw new Error(`Cluster ${clusterId} not found`); } + await this._signalRemoteCluster(cluster, { action: 'kill' }); + cluster.state = 'stopping'; // Force stop all agents @@ -1430,6 +1809,10 @@ class Orchestrator { await agent.stop(); } + if (cluster.snapshotter) { + cluster.snapshotter.stop(); + } + // Force remove isolation container AND workspace (full cleanup, no resume) if (cluster.isolation?.manager) { this._log( @@ -1439,6 +1822,14 @@ class Orchestrator { this._log(`[Orchestrator] Container and workspace removed`); } + if (cluster.validatorIsolation?.manager) { + this._log(`[Orchestrator] Force removing validator isolation container for ${clusterId}...`); + await cluster.validatorIsolation.manager.cleanup(cluster.validatorIsolation.clusterId, { + preserveWorkspace: false, + }); + cluster.validatorIsolation = null; + } + // Force remove worktree (full cleanup, no resume) // Note: Branch is preserved for potential PR creation / inspection if (cluster.worktree?.manager) { @@ -1533,6 +1924,7 @@ class Orchestrator { await this._ensureIsolationForResume(clusterId, cluster); this._ensureWorktreeForResume(clusterId, cluster); + this._startSnapshotter(cluster); await this._restartClusterAgents(cluster); const recentMessages = this._loadRecentMessages(cluster, clusterId, 50); @@ -1544,6 +1936,197 @@ class Orchestrator { return this._resumeCleanCluster(clusterId, cluster, recentMessages, prompt); } + _validateGuidanceAgentArgs(clusterId, agentId, text) { + if (!clusterId) { + throw new Error('sendGuidanceToAgent: clusterId is required'); + } + if (!agentId) { + throw new Error('sendGuidanceToAgent: agentId is required'); + } + if (typeof text !== 'string' || !text.trim()) { + throw new Error('sendGuidanceToAgent: text must be a non-empty string'); + } + } + + _validateGuidanceClusterArgs(clusterId, text) { + if (!clusterId) { + throw new Error('sendGuidanceToCluster: clusterId is required'); + } + if (typeof text !== 'string' || !text.trim()) { + throw new Error('sendGuidanceToCluster: text must be a non-empty string'); + } + } + + _getClusterOrThrow(clusterId, caller = 'sendGuidanceToAgent') { + const cluster = this.clusters.get(clusterId); + if (!cluster) { + throw new Error(`${caller}: cluster not found: ${clusterId}`); + } + return cluster; + } + + _getAgentOrThrow(cluster, agentId) { + const agent = cluster.agents.find((candidate) => candidate.id === agentId); + if (!agent) { + throw new Error(`sendGuidanceToAgent: agent not found: ${agentId}`); + } + return agent; + } + + _buildDefaultDelivery(agent) { + return { + status: 'unsupported', + reason: 'unknown', + method: null, + taskId: agent.currentTaskId || null, + }; + } + + async _attemptGuidanceInjection(agent, text, timeoutMs) { + try { + const result = await agent.injectInput(text, { timeoutMs }); + return { + status: result.status === 'injected' ? 'injected' : 'unsupported', + reason: result.status === 'injected' ? null : result.reason || 'unsupported', + method: result.status === 'injected' ? result.method || 'pty' : result.method || null, + taskId: result.taskId || agent.currentTaskId || null, + }; + } catch (error) { + return { + status: 'unsupported', + reason: error.message, + method: null, + taskId: agent.currentTaskId || null, + }; + } + } + + _buildGuidanceMetadata(options, delivery) { + return { + ...(options.metadata || {}), + delivery: { + status: delivery.status, + reason: delivery.reason, + method: delivery.method, + taskId: delivery.taskId, + timestamp: Date.now(), + }, + }; + } + + _summarizeGuidanceDeliveries(deliveries) { + const agentIds = Object.keys(deliveries); + const summary = { + injected: 0, + queued: 0, + total: agentIds.length, + }; + + for (const agentId of agentIds) { + const delivery = deliveries[agentId]; + if (delivery?.status === 'injected') { + summary.injected += 1; + } else { + summary.queued += 1; + } + } + + return summary; + } + + _buildClusterGuidanceMetadata(options, deliveries) { + return { + ...(options.metadata || {}), + delivery: { + summary: this._summarizeGuidanceDeliveries(deliveries), + agents: deliveries, + timestamp: Date.now(), + }, + }; + } + + _publishGuidance(cluster, clusterId, agentId, text, metadata, sender) { + cluster.messageBus.publish({ + cluster_id: clusterId, + topic: USER_GUIDANCE_AGENT, + sender, + target_agent_id: agentId, + content: { text }, + metadata, + }); + } + + _publishClusterGuidance(cluster, clusterId, text, metadata, sender) { + cluster.messageBus.publish({ + cluster_id: clusterId, + topic: USER_GUIDANCE_CLUSTER, + sender, + content: { text }, + metadata, + }); + } + + /** + * Send guidance to a specific agent with optional live injection. + * Always persists USER_GUIDANCE_AGENT in the ledger with delivery metadata. + * @param {string} clusterId + * @param {string} agentId + * @param {string} text + * @param {object} [options] + * @param {string} [options.sender='user'] + * @param {object} [options.metadata] + * @param {number} [options.timeoutMs] + * @returns {Promise<{status: string, reason: string|null, method: string|null, taskId: string|null}>} + */ + async sendGuidanceToAgent(clusterId, agentId, text, options = {}) { + this._validateGuidanceAgentArgs(clusterId, agentId, text); + + const cluster = this._getClusterOrThrow(clusterId, 'sendGuidanceToAgent'); + const agent = this._getAgentOrThrow(cluster, agentId); + const defaultDelivery = this._buildDefaultDelivery(agent); + const delivery = + (await this._attemptGuidanceInjection(agent, text, options.timeoutMs)) || defaultDelivery; + const metadata = this._buildGuidanceMetadata(options, delivery); + + this._publishGuidance(cluster, clusterId, agentId, text, metadata, options.sender || 'user'); + + return delivery; + } + + /** + * Send guidance to every agent in a cluster with optional live injection. + * Always persists a single USER_GUIDANCE_CLUSTER in the ledger with delivery metadata. + * @param {string} clusterId + * @param {string} text + * @param {object} [options] + * @param {string} [options.sender='user'] + * @param {object} [options.metadata] + * @param {number} [options.timeoutMs] + * @returns {Promise<{summary: {injected: number, queued: number, total: number}, agents: Record, timestamp: number}>} + */ + async sendGuidanceToCluster(clusterId, text, options = {}) { + this._validateGuidanceClusterArgs(clusterId, text); + + const cluster = this._getClusterOrThrow(clusterId, 'sendGuidanceToCluster'); + const agents = Array.isArray(cluster.agents) ? cluster.agents : []; + const deliveries = {}; + + await Promise.all( + agents.map(async (agent) => { + const defaultDelivery = this._buildDefaultDelivery(agent); + const delivery = + (await this._attemptGuidanceInjection(agent, text, options.timeoutMs)) || defaultDelivery; + deliveries[agent.id] = delivery; + }) + ); + + const metadata = this._buildClusterGuidanceMetadata(options, deliveries); + + this._publishClusterGuidance(cluster, clusterId, text, metadata, options.sender || 'user'); + + return metadata.delivery; + } + _resolveFailureInfo(cluster, clusterId) { if (cluster.failureInfo) { return cluster.failureInfo; @@ -1607,12 +2190,7 @@ class Orchestrator { ); } - const providerName = normalizeProviderName( - cluster.config?.forceProvider || - cluster.config?.defaultProvider || - loadSettings().defaultProvider || - 'claude' - ); + const providerName = this._resolveClusterProvider(cluster.config); const newContainerId = await cluster.isolation.manager.createContainer(clusterId, { workDir, image: cluster.isolation.image, @@ -2028,6 +2606,14 @@ Continue from where you left off. Review your previous output to understand what proposedAgentConfigs.push(agentConfig); } } + } else if (op.action === 'load_config' && op.config) { + const loadedAgentConfigs = this._resolveLoadConfigAgents(op.config); + for (const agentConfig of loadedAgentConfigs) { + const existingIdx = proposedAgentConfigs.findIndex((a) => a.id === agentConfig.id); + if (existingIdx === -1) { + proposedAgentConfigs.push(agentConfig); + } + } } else if (op.action === 'remove_agents' && op.agentIds) { for (const agentId of op.agentIds) { const idx = proposedAgentConfigs.findIndex((a) => a.id === agentId); @@ -2046,6 +2632,40 @@ Continue from where you left off. Review your previous output to understand what return proposedAgentConfigs; } + _resolveLoadConfigAgents(config) { + if (!config) { + throw new Error('load_config operation missing config'); + } + + const templatesDir = path.join(__dirname, '..', 'cluster-templates'); + let loadedConfig; + + // Parameterized template - resolve with TemplateResolver + if (typeof config === 'object' && config.base) { + const { base, params } = config; + const resolver = new TemplateResolver(templatesDir); + loadedConfig = resolver.resolve(base, params || {}); + } else if (typeof config === 'string') { + // Static config - load directly from file + const configPath = path.join(templatesDir, `${config}.json`); + if (!fs.existsSync(configPath)) { + throw new Error(`Config not found: ${config} (looked in ${configPath})`); + } + const configContent = fs.readFileSync(configPath, 'utf8'); + loadedConfig = JSON.parse(configContent); + } else { + throw new Error( + `Invalid config format: expected string or {base, params}, got ${typeof config}` + ); + } + + if (!loadedConfig.agents || !Array.isArray(loadedConfig.agents)) { + throw new Error(`Config has no agents array`); + } + + return loadedConfig.agents; + } + _validateProposedConfig(clusterId, cluster, proposedAgentConfigs, operations) { const mockConfig = { agents: proposedAgentConfigs }; const validation = configValidator.validateConfig(mockConfig); @@ -2135,11 +2755,33 @@ Continue from where you left off. Review your previous output to understand what throw new Error('Agent config missing required field: id'); } - // Check for duplicate agent ID - const existingAgent = cluster.agents.find((a) => a.id === agentConfig.id); - if (existingAgent) { - this._log(` āš ļø Agent ${agentConfig.id} already exists, skipping`); - continue; + // Check for duplicate agent ID - REPLACE agent entirely + // Previous behavior merged triggers but kept old hooks, causing bugs when + // loading templates with same agent IDs but different hooks (e.g., quick-validation + // and heavy-validation both have consensus-coordinator with different onComplete hooks) + const existingAgentIndex = cluster.agents.findIndex((a) => a.id === agentConfig.id); + if (existingAgentIndex !== -1) { + const existingAgent = cluster.agents[existingAgentIndex]; + this._log( + ` šŸ”„ Replacing agent ${agentConfig.id} (old role: ${existingAgent.config.role})` + ); + + // Stop the existing agent (cluster.agents contains AgentWrapper instances directly) + if (existingAgent.stop) { + existingAgent.stop(); + } + + // Remove from cluster.agents array + cluster.agents.splice(existingAgentIndex, 1); + + // Remove from cluster.config.agents if present + if (cluster.config.agents) { + const configIndex = cluster.config.agents.findIndex((a) => a.id === agentConfig.id); + if (configIndex !== -1) { + cluster.config.agents.splice(configIndex, 1); + } + } + // Continue to add the new agent below } // Add to config agents array (for persistence) @@ -2346,7 +2988,7 @@ Continue from where you left off. Review your previous output to understand what return; } - const isPrMode = cluster.autoPr || process.env.ZEROSHOT_PR === '1'; + const isPrMode = cluster.autoPr ?? process.env.ZEROSHOT_PR === '1'; if (isPrMode) { // Detect platform from stored cluster metadata OR git context @@ -2374,7 +3016,8 @@ Continue from where you left off. Review your previous output to understand what ); } - const gitPusherConfig = generateGitPusherAgent(platform); + // Use persisted PR options from cluster state (or empty for repo settings fallback) + const gitPusherConfig = generateGitPusherAgent(platform, cluster.prOptions || {}); // Get issue context from ledger const issueMsg = cluster.messageBus.ledger.findLast({ topic: 'ISSUE_OPENED' }); @@ -2390,39 +3033,18 @@ Continue from where you left off. Review your previous output to understand what this._log(` [--pr mode] Injected ${platform}-git-pusher agent`); } else { // Default completion-detector + const { SHARED_TRIGGER_SCRIPT } = require('./agents/git-pusher-template'); const completionDetector = { id: 'completion-detector', role: 'orchestrator', - model: 'haiku', + modelLevel: this._resolveCompletionDetectorLevel(cluster.config), timeout: 0, triggers: [ { topic: 'VALIDATION_RESULT', logic: { engine: 'javascript', - script: `const validators = cluster.getAgentsByRole('validator'); -const lastPush = ledger.findLast({ topic: 'IMPLEMENTATION_READY' }); -if (!lastPush) return false; -if (validators.length === 0) return true; - -const validatorIds = new Set(validators.map((v) => v.id)); -const results = ledger.query({ topic: 'VALIDATION_RESULT', since: lastPush.timestamp }); - -const latestByValidator = new Map(); -for (const msg of results) { - if (!validatorIds.has(msg.sender)) continue; - latestByValidator.set(msg.sender, msg); -} - -if (latestByValidator.size < validators.length) return false; - -for (const validator of validators) { - const msg = latestByValidator.get(validator.id); - const approved = msg?.content?.data?.approved; - if (!(approved === true || approved === 'true')) return false; -} - -return true;`, + script: SHARED_TRIGGER_SCRIPT, }, action: 'stop_cluster', }, @@ -2494,7 +3116,15 @@ return true;`, pid: cluster.pid || null, createdAt: cluster.createdAt, agents: cluster.agents.map((a) => a.getState()), - messageCount: cluster.messageBus.count({ cluster_id: clusterId }), + messageCount: (() => { + try { + return cluster.messageBus.count({ cluster_id: clusterId }); + } catch { + // Cluster may have closed its ledger during startup failure cleanup. + // Status/list should remain safe to call for visibility + supervisor cleanup. + return 0; + } + })(), }; } @@ -2521,8 +3151,17 @@ return true;`, id: cluster.id, state: state, createdAt: cluster.createdAt, + issue: cluster.issue || null, agentCount: cluster.agents.length, - messageCount: cluster.messageBus.getAll(cluster.id).length, + messageCount: (() => { + try { + return cluster.messageBus.count({ cluster_id: cluster.id }); + } catch { + // Cluster may have closed its ledger during startup failure cleanup. + // List should remain safe to call for cleanup routines. + return 0; + } + })(), }; }); } diff --git a/src/preflight.js b/src/preflight.js index b79461cb..cd14da6a 100644 --- a/src/preflight.js +++ b/src/preflight.js @@ -76,28 +76,50 @@ function getClaudeVersion(claudeCommand = 'claude') { const command = parts[0]; const extraArgs = parts.slice(1); + // Determine CLI presence without depending on `claude --version` working. + // Some environments can have Claude installed but a broken/unwritable config dir, + // which should NOT block preflight. + if (!commandExists(command)) { + return { + installed: false, + version: null, + error: `Command '${command}' not installed`, + }; + } + try { const versionArgs = [...extraArgs, '--version']; const versionCmd = [command, ...versionArgs].join(' '); - const output = execSync(versionCmd, { encoding: 'utf8', stdio: 'pipe' }); + // Claude Code may try to write debug output under CLAUDE_CONFIG_DIR even for `--version`. + // Preflight should detect installation, not fail due to an unwritable/invalid config dir. + const preflightConfigDir = path.join(os.tmpdir(), 'zeroshot-claude-preflight'); + try { + fs.mkdirSync(preflightConfigDir, { recursive: true }); + } catch { + // If we can't create it, fall back to current env. + } + const output = execSync(versionCmd, { + encoding: 'utf8', + stdio: 'pipe', + env: { + ...process.env, + CLAUDE_CONFIG_DIR: preflightConfigDir, + }, + }); const match = output.match(/(\d+\.\d+\.\d+)/); return { installed: true, version: match ? match[1] : 'unknown', error: null, }; - } catch (err) { - if (err.message.includes('command not found') || err.message.includes('not found')) { - return { - installed: false, - version: null, - error: `Command '${command}' not installed`, - }; - } + } catch { + // The CLI exists, but the version command can still fail due to local environment + // (e.g. config-dir permissions). Treat this as installed with unknown version so + // preflight doesn't block on non-essential metadata. return { - installed: false, - version: null, - error: err.message, + installed: true, + version: 'unknown', + error: null, }; } } @@ -220,19 +242,41 @@ function checkGhAuth() { }; } - // Check auth status + // Check auth status with timeout to prevent hangs + // gh auth status: exit 0 = authenticated, exit 1 = not authenticated + // Note: gh outputs to stderr even on success + const AUTH_CHECK_TIMEOUT_MS = 10000; try { - execSync('gh auth status', { encoding: 'utf8', stdio: 'pipe' }); + execSync('gh auth status', { + encoding: 'utf8', + stdio: 'pipe', + timeout: AUTH_CHECK_TIMEOUT_MS, + }); + // Exit code 0 = authenticated return { installed: true, authenticated: true, error: null, }; } catch (err) { - // gh auth status returns non-zero if not authenticated - const stderr = err.stderr || err.message || ''; + const stderr = err.stderr || ''; - if (stderr.includes('not logged in')) { + // Check if killed due to timeout + if (err.killed || err.signal === 'SIGTERM') { + return { + installed: true, + authenticated: false, + error: 'gh auth status timed out - try running: gh auth login', + }; + } + + // gh auth status returns non-zero if not authenticated + // Check stderr for common "not logged in" patterns + if ( + stderr.includes('not logged in') || + stderr.includes('not logged into') || + stderr.includes('You are not logged') + ) { return { installed: true, authenticated: false, @@ -240,10 +284,21 @@ function checkGhAuth() { }; } + // But if stderr contains "Logged in", trust that (some edge cases) + if (stderr.includes('Logged in')) { + return { + installed: true, + authenticated: true, + error: null, + }; + } + + // Provide more helpful error - include actual stderr for debugging + const details = stderr.trim() || `Exit code ${err.status || 'unknown'}`; return { installed: true, authenticated: false, - error: stderr.trim() || 'Unknown gh auth error', + error: `gh auth check failed: ${details}`, }; } } @@ -473,6 +528,24 @@ function runPreflight(options = {}) { const warnings = []; const settings = loadSettings(); + + if (process.platform === 'win32') { + return { + valid: false, + errors: [ + formatError( + 'Windows not supported', + 'Zeroshot currently supports Linux and macOS only; Windows (native or WSL) is deferred.', + [ + 'Use Linux or macOS to run Zeroshot', + 'Or run inside a Linux VM/container on Windows', + 'Check README for supported platforms and updates', + ] + ), + ], + warnings: [], + }; + } const providerName = normalizeProviderName( options.provider || settings.defaultProvider || 'claude' ); diff --git a/src/providers/anthropic/index.js b/src/providers/anthropic/index.js index bd496ec7..1c336da1 100644 --- a/src/providers/anthropic/index.js +++ b/src/providers/anthropic/index.js @@ -20,6 +20,19 @@ class AnthropicProvider extends BaseProvider { this._cliFeatures = null; } + getRetryableErrorPatterns() { + return [ + ...super.getRetryableErrorPatterns(), + /no messages returned/i, + /\boverloaded\b/i, + /\brate[_ -]?limit\b/i, + ]; + } + + getPermanentErrorPatterns() { + return [...super.getPermanentErrorPatterns(), /invalid_request_error/i, /model_not_available/i]; + } + // SDK not implemented - uses CLI only // See BaseProvider for SDK extension point documentation diff --git a/src/providers/anthropic/models.js b/src/providers/anthropic/models.js index e2a0dc53..68327c8d 100644 --- a/src/providers/anthropic/models.js +++ b/src/providers/anthropic/models.js @@ -2,12 +2,13 @@ const MODEL_CATALOG = { haiku: { rank: 1 }, sonnet: { rank: 2 }, opus: { rank: 3 }, + 'opus-4.6': { rank: 3 }, }; const LEVEL_MAPPING = { level1: { rank: 1, model: 'haiku' }, level2: { rank: 2, model: 'sonnet' }, - level3: { rank: 3, model: 'opus' }, + level3: { rank: 3, model: 'opus-4.6' }, }; const DEFAULT_LEVEL = 'level2'; diff --git a/src/providers/anthropic/output-parser.js b/src/providers/anthropic/output-parser.js index 083274ba..8cd69572 100644 --- a/src/providers/anthropic/output-parser.js +++ b/src/providers/anthropic/output-parser.js @@ -18,7 +18,8 @@ function parseResultEvent(event) { duration: event.duration_ms, inputTokens: usage.input_tokens || 0, outputTokens: usage.output_tokens || 0, - cacheReadInputTokens: usage.cache_read_input_tokens || 0, + // Claude CLI uses 'cached_input_tokens'; Anthropic API uses 'cache_read_input_tokens' + cacheReadInputTokens: usage.cache_read_input_tokens || usage.cached_input_tokens || 0, cacheCreationInputTokens: usage.cache_creation_input_tokens || 0, modelUsage: event.modelUsage || null, }; diff --git a/src/providers/base-provider.js b/src/providers/base-provider.js index 3a57cc49..d60b2edc 100644 --- a/src/providers/base-provider.js +++ b/src/providers/base-provider.js @@ -14,6 +14,103 @@ class BaseProvider { this.cliCommand = options.cliCommand || null; } + // ============================================================================ + // ERROR CLASSIFICATION (Retry vs Permanent) + // ============================================================================ + // + // Providers currently execute via CLI, so errors are often unstructured strings. + // These helpers are used by the agent retry loop to decide whether to retry. + // + + /** + * Patterns that indicate the error is retryable/transient. + * Override in provider implementations to add provider-specific patterns. + * @returns {Array} Retryable error patterns + */ + getRetryableErrorPatterns() { + return [ + /rate.?limit/i, + /\b429\b/i, + /too many requests/i, + /overloaded/i, + /temporar(?:y|ily)/i, + /unavailable/i, + /try again/i, + /timeout/i, + /timed out/i, + /deadline exceeded/i, + /connection (?:reset|refused)/i, + /\b(econnreset|econnrefused|etimedout|eai_again)\b/i, + /network/i, + ]; + } + + /** + * Patterns that indicate the error is permanent and should not be retried. + * Override in provider implementations to add provider-specific patterns. + * @returns {Array} Permanent error patterns + */ + getPermanentErrorPatterns() { + return [ + /invalid[_ -]?api[_ -]?key/i, + /api[_ -]?key.*invalid/i, + /unauthorized/i, + /forbidden/i, + /authentication/i, + /permission denied/i, + /invalid argument/i, + /unknown option/i, + /\busage:\b/i, + /command not found/i, + /not recognized as an internal or external command/i, + /model not found/i, + /context length exceeded/i, + /insufficient quota/i, + ]; + } + + /** + * Classify whether an error is retryable. + * Default behavior is conservative: unknown errors are treated as retryable + * to prevent stuck clusters from single transient failures. + * @param {any} err - Error object (often Error with message) + * @returns {boolean} True if retryable + */ + isRetryableError(err) { + const status = + err?.status ?? err?.statusCode ?? err?.response?.status ?? err?.response?.statusCode ?? null; + + if (typeof status === 'number') { + if (status === 429 || status >= 500) return true; + if (status >= 400 && status < 500) return false; + } + + const code = err?.code || null; + if ( + typeof code === 'string' && + /\b(econnreset|econnrefused|etimedout|eai_again)\b/i.test(code) + ) { + return true; + } + + const message = (err?.message || String(err) || '').trim(); + if (!message) { + return true; + } + + const permanent = this.getPermanentErrorPatterns(); + if (Array.isArray(permanent) && permanent.some((p) => p.test(message))) { + return false; + } + + const retryable = this.getRetryableErrorPatterns(); + if (Array.isArray(retryable) && retryable.some((p) => p.test(message))) { + return true; + } + + return true; + } + // ============================================================================ // SDK SUPPORT (Future Extension Point) // ============================================================================ @@ -177,7 +274,7 @@ class BaseProvider { * Resolve a model name to its CLI-compatible identifier. * Override in provider implementations that need model ID transformation * (e.g., Anthropic Bedrock mapping). - * @param {string} model - Model name (e.g., 'opus', 'sonnet') + * @param {string} model - Provider model identifier * @param {Object} _authEnv - Authentication environment variables * @returns {string} CLI-compatible model identifier */ diff --git a/src/providers/google/index.js b/src/providers/google/index.js index a8c1c69c..95baa66e 100644 --- a/src/providers/google/index.js +++ b/src/providers/google/index.js @@ -20,6 +20,26 @@ class GoogleProvider extends BaseProvider { this._parserState = { lastToolId: null }; } + getRetryableErrorPatterns() { + return [ + ...super.getRetryableErrorPatterns(), + /\bRESOURCE_EXHAUSTED\b/i, + /\bUNAVAILABLE\b/i, + /\bDEADLINE_EXCEEDED\b/i, + /No capacity available/i, // Gemini rate limit + /quota.?exceeded/i, // Quota exhaustion + ]; + } + + getPermanentErrorPatterns() { + return [ + ...super.getPermanentErrorPatterns(), + /\bINVALID_ARGUMENT\b/i, + /\bPERMISSION_DENIED\b/i, + /\bNOT_FOUND\b/i, + ]; + } + // SDK not implemented - uses CLI only // See BaseProvider for SDK extension point documentation diff --git a/src/providers/openai/index.js b/src/providers/openai/index.js index 5ed716e5..d242d539 100644 --- a/src/providers/openai/index.js +++ b/src/providers/openai/index.js @@ -19,6 +19,24 @@ class OpenAIProvider extends BaseProvider { this._unknownEventCounts = new Map(); } + getRetryableErrorPatterns() { + return [ + ...super.getRetryableErrorPatterns(), + /rate_limit_exceeded/i, + /\bserver_error\b/i, + /\bservice_unavailable\b/i, + ]; + } + + getPermanentErrorPatterns() { + return [ + ...super.getPermanentErrorPatterns(), + /\binsufficient_quota\b/i, + /\bmodel_not_found\b/i, + /\bcontext_length_exceeded\b/i, + ]; + } + // SDK not implemented - uses CLI only // See BaseProvider for SDK extension point documentation diff --git a/src/providers/openai/models.js b/src/providers/openai/models.js index 3283dea4..595daa76 100644 --- a/src/providers/openai/models.js +++ b/src/providers/openai/models.js @@ -1,11 +1,12 @@ -// Codex CLI - use null to let CLI pick its default model -// Levels vary by reasoning effort only -const MODEL_CATALOG = {}; +// Codex defaults to gpt-5.3-codex; levels vary by reasoning effort only. +const MODEL_CATALOG = { + 'gpt-5.3-codex': { rank: 2 }, +}; const LEVEL_MAPPING = { - level1: { rank: 1, model: null, reasoningEffort: 'low' }, - level2: { rank: 2, model: null, reasoningEffort: 'medium' }, - level3: { rank: 3, model: null, reasoningEffort: 'high' }, + level1: { rank: 1, model: 'gpt-5.3-codex', reasoningEffort: 'medium' }, + level2: { rank: 2, model: 'gpt-5.3-codex', reasoningEffort: 'high' }, + level3: { rank: 3, model: 'gpt-5.3-codex', reasoningEffort: 'xhigh' }, }; const DEFAULT_LEVEL = 'level2'; diff --git a/src/providers/openai/output-parser.js b/src/providers/openai/output-parser.js index da3257e9..c7098524 100644 --- a/src/providers/openai/output-parser.js +++ b/src/providers/openai/output-parser.js @@ -47,8 +47,47 @@ function parseFunctionCallOutput(item) { }; } -function parseItem(item) { +function parseCommandExecutionItem(item, phase) { + // Codex CLI (newer) emits `command_execution` items for bash-like tool runs. + // Map them into the shared schema expected by the logs renderer. + const toolId = item.id; + const command = item.command || item.cmd || item.input?.command || item.input?.cmd; + + if (phase === 'started') { + return { + type: 'tool_call', + toolName: 'Bash', + toolId, + input: command ? { command } : {}, + }; + } + + const output = item.aggregated_output ?? item.output ?? item.result ?? ''; + const exitCode = + typeof item.exit_code === 'number' + ? item.exit_code + : typeof item.exitCode === 'number' + ? item.exitCode + : null; + + return { + type: 'tool_result', + toolId, + content: output, + isError: exitCode !== null ? exitCode !== 0 : !!item.error, + }; +} + +function parseReasoningItem(item) { + const text = item.text || item.content || ''; + if (!text) return null; + return { type: 'thinking', text }; +} + +function parseItem(item, eventType) { const events = []; + const phase = + eventType === 'item.started' ? 'started' : eventType === 'item.completed' ? 'completed' : null; // Handle assistant messages (Claude-style: type=message, role=assistant) if (item.type === 'message' && item.role === 'assistant') { @@ -60,6 +99,15 @@ function parseItem(item) { events.push({ type: 'text', text: item.text }); } + if (item.type === 'reasoning') { + const reasoning = parseReasoningItem(item); + if (reasoning) events.push(reasoning); + } + + if (item.type === 'command_execution' && phase) { + events.push(parseCommandExecutionItem(item, phase)); + } + if (item.type === 'function_call') { events.push(parseFunctionCall(item)); } @@ -82,13 +130,25 @@ function parseEvent(line, options = {}) { } switch (event.type) { + case 'error': + return { + type: 'result', + success: false, + error: event.error?.message || event.message || event.error || 'Error', + }; + case 'thread.started': case 'turn.started': return null; + case 'item.started': + if (!event.item || event.item.type !== 'command_execution') return null; + return parseItem(event.item, event.type); + case 'item.created': case 'item.completed': - return parseItem(event.item); + if (!event.item) return null; + return parseItem(event.item, event.type); case 'turn.completed': { const usage = event.usage || event.response?.usage || {}; diff --git a/src/schemas/sub-cluster.js b/src/schemas/sub-cluster.js index 0e05b5ae..3c363980 100644 --- a/src/schemas/sub-cluster.js +++ b/src/schemas/sub-cluster.js @@ -132,13 +132,36 @@ function validateContextStrategy(agentConfig, errors) { return; } - // Validate each parent topic is a string - for (const topic of parentTopics) { - if (typeof topic !== 'string') { + // Validate each parent topic entry + for (const entry of parentTopics) { + if (typeof entry === 'string') { + continue; + } + + if (!entry || typeof entry !== 'object') { + errors.push( + `Sub-cluster '${agentConfig.id}' parentTopics must contain strings or objects, got ${typeof entry}` + ); + continue; + } + + if (typeof entry.topic !== 'string') { + errors.push(`Sub-cluster '${agentConfig.id}' parentTopics entry must include a string topic`); + } + + if (entry.strategy && !['latest', 'all', 'oldest'].includes(entry.strategy)) { errors.push( - `Sub-cluster '${agentConfig.id}' parentTopics must contain strings, got ${typeof topic}` + `Sub-cluster '${agentConfig.id}' parentTopics entry has invalid strategy '${entry.strategy}'` ); } + + if (entry.amount !== undefined && !Number.isFinite(entry.amount)) { + errors.push(`Sub-cluster '${agentConfig.id}' parentTopics entry amount must be a number`); + } + + if (entry.limit !== undefined && !Number.isFinite(entry.limit)) { + errors.push(`Sub-cluster '${agentConfig.id}' parentTopics entry limit must be a number`); + } } } diff --git a/src/state-snapshot.js b/src/state-snapshot.js new file mode 100644 index 00000000..11ecdd1a --- /dev/null +++ b/src/state-snapshot.js @@ -0,0 +1,398 @@ +const SNAPSHOT_VERSION = 1; + +const LIMITS = { + errors: 5, + criteriaResults: 10, + acceptanceCriteria: 10, + filesAffected: 20, + blockers: 5, + nextSteps: 10, + rootCauses: 5, +}; + +const TEXT_LIMITS = { + task: 2000, + plan: 2500, // Slightly increased for actionable plans with embedded patterns (was 2000) + fixPlan: 1200, + summary: 300, + listItem: 200, +}; + +function toTimestamp(message) { + if (message && Number.isFinite(message.timestamp)) { + return message.timestamp; + } + return Date.now(); +} + +function normalizeText(value, maxLength, singleLine = false) { + if (value === undefined || value === null) return undefined; + let text = String(value); + if (singleLine) { + text = text.replace(/\s+/g, ' ').trim(); + } else { + text = text.trim(); + } + if (!text) return undefined; + if (maxLength && text.length > maxLength) { + return `${text.slice(0, maxLength - 3)}...`; + } + return text; +} + +function normalizeStringList(list, maxItems) { + if (!Array.isArray(list)) return undefined; + const normalized = list + .map((item) => normalizeText(item, TEXT_LIMITS.listItem, true)) + .filter(Boolean); + if (normalized.length === 0) return undefined; + if (maxItems && normalized.length > maxItems) { + return normalized.slice(-maxItems); + } + return normalized; +} + +function normalizeAcceptanceCriteria(criteria) { + if (!Array.isArray(criteria)) return undefined; + const normalized = criteria + .map((item) => { + if (typeof item === 'string') { + return normalizeText(item, TEXT_LIMITS.listItem, true); + } + if (!item || typeof item !== 'object') return undefined; + const id = item.id ? String(item.id) : ''; + const priority = item.priority ? ` (${item.priority})` : ''; + const criterion = item.criterion || item.text || item.summary || ''; + const label = id ? `${id}${priority}: ` : ''; + const merged = `${label}${criterion}`.trim(); + if (!merged) return undefined; + return normalizeText(merged, TEXT_LIMITS.listItem, true); + }) + .filter(Boolean); + if (normalized.length === 0) return undefined; + if (normalized.length > LIMITS.acceptanceCriteria) { + return normalized.slice(-LIMITS.acceptanceCriteria); + } + return normalized; +} + +function normalizeCriteriaEvidence(evidence) { + if (!evidence || typeof evidence !== 'object') return undefined; + const normalized = {}; + if (evidence.command) { + const command = normalizeText(evidence.command, TEXT_LIMITS.listItem, true); + if (command) normalized.command = command; + } + if (Number.isFinite(evidence.exitCode)) { + normalized.exitCode = evidence.exitCode; + } + return Object.keys(normalized).length > 0 ? normalized : undefined; +} + +function normalizeCriteriaResult(item) { + if (!item || typeof item !== 'object') return undefined; + const entry = {}; + if (item.id) entry.id = String(item.id); + if (item.status) entry.status = String(item.status); + if (item.reason) { + const reason = normalizeText(item.reason, TEXT_LIMITS.listItem, true); + if (reason) entry.reason = reason; + } + const evidence = normalizeCriteriaEvidence(item.evidence); + if (evidence) entry.evidence = evidence; + return Object.keys(entry).length > 0 ? entry : undefined; +} + +function normalizeCriteriaResults(results) { + if (!Array.isArray(results)) return undefined; + const normalized = results.map(normalizeCriteriaResult).filter(Boolean); + if (normalized.length === 0) return undefined; + if (normalized.length > LIMITS.criteriaResults) { + return normalized.slice(-LIMITS.criteriaResults); + } + return normalized; +} + +function normalizeErrors(data) { + if (!data || typeof data !== 'object') return undefined; + if (Array.isArray(data.errors)) { + return normalizeStringList(data.errors, LIMITS.errors); + } + if (Array.isArray(data.issues)) { + const mapped = data.issues.map((issue) => { + if (typeof issue === 'string') return issue; + if (!issue || typeof issue !== 'object') return undefined; + return issue.bug || issue.message || issue.error || issue.summary || undefined; + }); + return normalizeStringList(mapped, LIMITS.errors); + } + return undefined; +} + +function normalizeRootCauses(rootCauses) { + if (!Array.isArray(rootCauses)) return undefined; + const normalized = rootCauses + .map((cause) => { + if (typeof cause === 'string') { + return normalizeText(cause, TEXT_LIMITS.listItem, true); + } + if (!cause || typeof cause !== 'object') return undefined; + return normalizeText( + cause.cause || cause.summary || cause.description, + TEXT_LIMITS.listItem, + true + ); + }) + .filter(Boolean); + if (normalized.length === 0) return undefined; + if (normalized.length > LIMITS.rootCauses) { + return normalized.slice(-LIMITS.rootCauses); + } + return normalized; +} + +function normalizeFilesAffected(filesAffected) { + return normalizeStringList(filesAffected, LIMITS.filesAffected); +} + +function normalizeBoolean(value) { + if (typeof value === 'boolean') return value; + if (value === 'true') return true; + if (value === 'false') return false; + return undefined; +} + +function normalizeProgressStatus(data) { + if (!data || typeof data !== 'object') return undefined; + if (data.completionStatus && typeof data.completionStatus === 'object') { + return data.completionStatus; + } + const hasProgressFields = + Object.prototype.hasOwnProperty.call(data, 'canValidate') || + Object.prototype.hasOwnProperty.call(data, 'percentComplete'); + return hasProgressFields ? data : undefined; +} + +function buildBaseState(state, message) { + return { + version: SNAPSHOT_VERSION, + updatedAt: toTimestamp(message), + clusterId: message?.cluster_id || state?.clusterId || null, + sourceMessageId: message?.id || state?.sourceMessageId || null, + task: state?.task, + plan: state?.plan, + progress: state?.progress, + validation: state?.validation, + debug: state?.debug, + }; +} + +function pruneEmpty(value) { + if (Array.isArray(value)) { + const next = value.map(pruneEmpty).filter((item) => item !== undefined); + return next.length > 0 ? next : undefined; + } + if (value && typeof value === 'object') { + const next = {}; + for (const [key, entry] of Object.entries(value)) { + const pruned = pruneEmpty(entry); + if (pruned !== undefined) { + next[key] = pruned; + } + } + return Object.keys(next).length > 0 ? next : undefined; + } + if (value === undefined || value === null) return undefined; + return value; +} + +function finalizeState(state) { + const meta = { + version: state.version ?? SNAPSHOT_VERSION, + updatedAt: state.updatedAt ?? Date.now(), + clusterId: state.clusterId ?? null, + sourceMessageId: state.sourceMessageId ?? null, + }; + const sections = pruneEmpty({ + task: state.task, + plan: state.plan, + progress: state.progress, + validation: state.validation, + debug: state.debug, + }); + return { + ...meta, + ...(sections || {}), + }; +} + +function initStateFromIssue(issueMessage) { + const content = issueMessage?.content || {}; + const data = content.data || {}; + const task = { + raw: normalizeText(content.text, TEXT_LIMITS.task), + title: normalizeText(data.title, TEXT_LIMITS.summary, true), + issueNumber: data.issue_number ?? data.issueNumber, + source: issueMessage?.metadata?.source, + }; + const base = buildBaseState(null, issueMessage); + base.task = pruneEmpty(task); + return finalizeState(base); +} + +function applyIssueOpened(state, message) { + const base = buildBaseState(state, message); + const content = message?.content || {}; + const data = content.data || {}; + const task = { + raw: normalizeText(content.text, TEXT_LIMITS.task), + title: normalizeText(data.title, TEXT_LIMITS.summary, true), + issueNumber: data.issue_number ?? data.issueNumber, + source: message?.metadata?.source, + }; + base.task = pruneEmpty(task); + return finalizeState(base); +} + +function applyPlanReady(state, message) { + const base = buildBaseState(state, message); + const content = message?.content || {}; + const data = content.data || {}; + const plan = { + text: normalizeText(content.text, TEXT_LIMITS.plan), + summary: normalizeText(data.summary, TEXT_LIMITS.summary, true), + acceptanceCriteria: normalizeAcceptanceCriteria(data.acceptanceCriteria), + filesAffected: normalizeFilesAffected(data.filesAffected), + updatedAt: toTimestamp(message), + }; + base.plan = pruneEmpty(plan); + return finalizeState(base); +} + +function applyWorkerProgress(state, message) { + const base = buildBaseState(state, message); + const content = message?.content || {}; + const status = normalizeProgressStatus(content.data || {}); + if (!status) { + return finalizeState(base); + } + const progress = { + canValidate: normalizeBoolean(status.canValidate), + percentComplete: Number.isFinite(status.percentComplete) ? status.percentComplete : undefined, + blockers: normalizeStringList(status.blockers, LIMITS.blockers), + nextSteps: normalizeStringList(status.nextSteps, LIMITS.nextSteps), + lastSummary: normalizeText(content.text || status.summary, TEXT_LIMITS.summary, true), + updatedAt: toTimestamp(message), + }; + base.progress = pruneEmpty(progress); + return finalizeState(base); +} + +function applyImplementationReady(state, message) { + return applyWorkerProgress(state, message); +} + +function applyValidationResult(state, message) { + const base = buildBaseState(state, message); + const content = message?.content || {}; + const data = content.data || {}; + const validation = { + approved: normalizeBoolean(data.approved), + errors: normalizeErrors(data), + criteriaResults: normalizeCriteriaResults(data.criteriaResults), + updatedAt: toTimestamp(message), + }; + base.validation = pruneEmpty(validation); + return finalizeState(base); +} + +function applyInvestigationComplete(state, message) { + const base = buildBaseState(state, message); + const content = message?.content || {}; + const data = content.data || {}; + const debug = { + fixPlan: normalizeText(content.text, TEXT_LIMITS.fixPlan), + successCriteria: normalizeText(data.successCriteria, TEXT_LIMITS.summary, true), + rootCauses: normalizeRootCauses(data.rootCauses), + updatedAt: toTimestamp(message), + }; + base.debug = pruneEmpty(debug); + return finalizeState(base); +} + +function buildTaskSummary(state) { + const taskTitle = normalizeText(state.task?.title || state.task?.raw, TEXT_LIMITS.summary, true); + return taskTitle ? `Task: ${taskTitle}` : undefined; +} + +function buildPlanSummary(state) { + const planSummary = normalizeText( + state.plan?.summary || state.plan?.text, + TEXT_LIMITS.summary, + true + ); + return planSummary ? `Plan: ${planSummary}` : undefined; +} + +function buildProgressSummary(state) { + if (!state.progress) return undefined; + const parts = []; + if (Number.isFinite(state.progress.percentComplete)) { + parts.push(`${state.progress.percentComplete}%`); + } + if (typeof state.progress.canValidate === 'boolean') { + parts.push(`canValidate=${state.progress.canValidate}`); + } + const nextStepText = normalizeText(state.progress.nextSteps?.[0], TEXT_LIMITS.listItem, true); + if (nextStepText) { + parts.push(`next: ${nextStepText}`); + } + return parts.length > 0 ? `Progress: ${parts.join(' | ')}` : undefined; +} + +function resolveValidationStatus(approved) { + if (approved === true) return 'approved'; + if (approved === false) return 'rejected'; + return 'pending'; +} + +function buildValidationSummary(state) { + if (!state.validation) return undefined; + const status = resolveValidationStatus(state.validation.approved); + const errorCount = state.validation.errors?.length || 0; + return `Validation: ${status}${errorCount ? ` (${errorCount} errors)` : ''}`; +} + +function buildDebugSummary(state) { + const debugSummary = normalizeText( + state.debug?.fixPlan || state.debug?.successCriteria, + TEXT_LIMITS.summary, + true + ); + return debugSummary ? `Debug: ${debugSummary}` : undefined; +} + +function renderStateSummary(state) { + if (!state || typeof state !== 'object') return ''; + const lines = [ + buildTaskSummary(state), + buildPlanSummary(state), + buildProgressSummary(state), + buildValidationSummary(state), + buildDebugSummary(state), + ].filter(Boolean); + + return lines.join('\n'); +} + +module.exports = { + SNAPSHOT_VERSION, + initStateFromIssue, + applyIssueOpened, + applyPlanReady, + applyWorkerProgress, + applyImplementationReady, + applyValidationResult, + applyInvestigationComplete, + renderStateSummary, +}; diff --git a/src/state-snapshotter.js b/src/state-snapshotter.js new file mode 100644 index 00000000..5a8eeb12 --- /dev/null +++ b/src/state-snapshotter.js @@ -0,0 +1,142 @@ +const crypto = require('crypto'); +const { + initStateFromIssue, + applyIssueOpened, + applyPlanReady, + applyWorkerProgress, + applyImplementationReady, + applyValidationResult, + applyInvestigationComplete, + renderStateSummary, +} = require('./state-snapshot'); + +const SNAPSHOT_TOPICS = [ + 'ISSUE_OPENED', + 'PLAN_READY', + 'WORKER_PROGRESS', + 'IMPLEMENTATION_READY', + 'VALIDATION_RESULT', + 'INVESTIGATION_COMPLETE', +]; + +class StateSnapshotter { + constructor({ messageBus, clusterId }) { + this.messageBus = messageBus; + this.clusterId = clusterId; + this.state = null; + this.lastHash = null; + this.unsubscribe = null; + } + + start() { + if (this.unsubscribe) { + return; + } + + this._bootstrapFromLedger(); + + this.unsubscribe = this.messageBus.subscribeTopics(SNAPSHOT_TOPICS, (message) => { + if (message.cluster_id !== this.clusterId) return; + this._handleMessage(message); + }); + } + + stop() { + if (!this.unsubscribe) return; + this.unsubscribe(); + this.unsubscribe = null; + } + + _bootstrapFromLedger() { + const existing = this.messageBus.findLast({ + cluster_id: this.clusterId, + topic: 'STATE_SNAPSHOT', + }); + + if (existing?.content?.data && typeof existing.content.data === 'object') { + this.state = existing.content.data; + this.lastHash = this._hashState(this.state); + return; + } + + const messages = SNAPSHOT_TOPICS.map((topic) => + this.messageBus.findLast({ cluster_id: this.clusterId, topic }) + ).filter(Boolean); + + if (messages.length === 0) { + return; + } + + messages.sort((a, b) => (a.timestamp || 0) - (b.timestamp || 0)); + + let state = null; + for (const message of messages) { + state = this._applyMessage(state, message); + } + + if (state) { + this.state = state; + this._publishSnapshot(state); + } + } + + _handleMessage(message) { + const nextState = this._applyMessage(this.state, message); + if (!nextState) return; + + this.state = nextState; + this._publishSnapshot(nextState); + } + + _applyMessage(state, message) { + switch (message.topic) { + case 'ISSUE_OPENED': + return state ? applyIssueOpened(state, message) : initStateFromIssue(message); + case 'PLAN_READY': + return applyPlanReady(state, message); + case 'WORKER_PROGRESS': + return applyWorkerProgress(state, message); + case 'IMPLEMENTATION_READY': + return applyImplementationReady(state, message); + case 'VALIDATION_RESULT': + return applyValidationResult(state, message); + case 'INVESTIGATION_COMPLETE': + return applyInvestigationComplete(state, message); + default: + return state; + } + } + + _publishSnapshot(state) { + const hash = this._hashState(state); + if (this._hashEquals(hash, this.lastHash)) { + return; + } + this.lastHash = hash; + + this.messageBus.publish({ + cluster_id: this.clusterId, + topic: 'STATE_SNAPSHOT', + sender: 'state-snapshotter', + receiver: 'broadcast', + content: { + text: renderStateSummary(state), + data: state, + }, + }); + } + + _hashState(state) { + return crypto.createHash('sha256').update(JSON.stringify(state)).digest('hex'); + } + + _hashEquals(left, right) { + if (!left || !right) return false; + const leftBuffer = Buffer.from(left, 'utf8'); + const rightBuffer = Buffer.from(right, 'utf8'); + if (leftBuffer.length !== rightBuffer.length) return false; + return crypto.timingSafeEqual(leftBuffer, rightBuffer); + } +} + +module.exports = StateSnapshotter; diff --git a/src/sub-cluster-wrapper.js b/src/sub-cluster-wrapper.js index 699111fa..ced36853 100644 --- a/src/sub-cluster-wrapper.js +++ b/src/sub-cluster-wrapper.js @@ -15,6 +15,39 @@ const LogicEngine = require('./logic-engine'); const MessageBusBridge = require('./message-bus-bridge'); const { DEFAULT_MAX_ITERATIONS } = require('./agent/agent-config'); +const { bufferMessage, scheduleDrain, drainBufferedMessages } = require('./message-buffer'); + +function normalizeParentTopicConfig(entry) { + if (typeof entry === 'string') { + return { topic: entry, amount: 10, strategy: 'latest' }; + } + if (!entry || typeof entry !== 'object') { + return null; + } + const amount = entry.amount ?? entry.limit; + const strategy = entry.strategy ?? (amount !== undefined ? 'latest' : 'all'); + return { ...entry, amount, strategy }; +} + +function selectParentTopicMessages(messageBus, clusterId, topicConfig) { + const { topic, sender, since, until, amount, strategy } = topicConfig; + const order = strategy === 'latest' ? 'desc' : 'asc'; + const messages = messageBus.query({ + cluster_id: clusterId, + topic, + sender, + since, + until, + limit: amount, + order, + }); + + if (strategy === 'latest' && messages.length > 1) { + return messages.slice().reverse(); + } + + return messages; +} class SubClusterWrapper { constructor(config, messageBus, parentCluster, options = {}) { @@ -135,6 +168,10 @@ class SubClusterWrapper { * @private */ async _handleMessage(message) { + if (!this._bufferedMessages) { + this._bufferedMessages = []; + } + // Check if any trigger matches const matchingTrigger = this._findMatchingTrigger(message); if (!matchingTrigger) { @@ -147,8 +184,17 @@ class SubClusterWrapper { return; } if (this.state !== 'idle') { + bufferMessage(this, message); console.warn( - `[${this.id}] āš ļø DROPPING message (busy, state=${this.state}): ${message.topic}` + `[${this.id}] āøļø BUFFERING message (busy, state=${this.state}): ${message.topic}` + ); + scheduleDrain( + this, + () => + drainBufferedMessages(this, (next) => this._handleMessage(next), { + label: 'SubCluster', + }), + { label: 'SubCluster' } ); return; } @@ -308,8 +354,12 @@ class SubClusterWrapper { lines.push('## Parent Cluster Messages', ''); - for (const topic of parentTopics) { - const topicLines = this._buildTopicContextLines(topic); + for (const entry of parentTopics) { + const topicConfig = normalizeParentTopicConfig(entry); + if (!topicConfig?.topic) { + continue; + } + const topicLines = this._buildTopicContextLines(topicConfig); if (topicLines.length === 0) { continue; } @@ -318,18 +368,14 @@ class SubClusterWrapper { } } - _buildTopicContextLines(topic) { - const messages = this.messageBus.query({ - cluster_id: this.parentCluster.id, - topic, - limit: 10, - }); + _buildTopicContextLines(topicConfig) { + const messages = selectParentTopicMessages(this.messageBus, this.parentCluster.id, topicConfig); if (messages.length === 0) { return []; } - const lines = [`### Topic: ${topic}`, '']; + const lines = [`### Topic: ${topicConfig.topic}`, '']; for (const message of messages) { lines.push(...this._buildMessageContextLines(message)); @@ -430,6 +476,29 @@ class SubClusterWrapper { if (message.topic === 'CLUSTER_COMPLETE' && message.cluster_id === childId) { this._onChildComplete(message).catch((err) => { console.error(`Failed to handle child completion: ${err.message}`); + + // CRITICAL: Hook failure = cluster failure + this._publishLifecycle('HOOK_FAILED', { + error: err.message, + hook: 'onComplete', + }); + + this.messageBus.publish({ + cluster_id: this.parentCluster.id, + topic: 'CLUSTER_FAILED', + sender: this.id, + content: { + text: `Hook failed: ${err.message}`, + data: { + reason: 'onComplete hook failed', + error: err.message, + stack: err.stack, + }, + }, + }); + + this.state = 'failed'; + throw err; }); } }); @@ -439,6 +508,29 @@ class SubClusterWrapper { if (message.topic === 'CLUSTER_FAILED' && message.cluster_id === childId) { this._onChildFailed(message).catch((err) => { console.error(`Failed to handle child failure: ${err.message}`); + + // CRITICAL: Hook failure = cluster failure + this._publishLifecycle('HOOK_FAILED', { + error: err.message, + hook: 'onFailed', + }); + + this.messageBus.publish({ + cluster_id: this.parentCluster.id, + topic: 'CLUSTER_FAILED', + sender: this.id, + content: { + text: `Hook failed: ${err.message}`, + data: { + reason: 'onFailed hook failed', + error: err.message, + stack: err.stack, + }, + }, + }); + + this.state = 'failed'; + throw err; }); } }); @@ -456,13 +548,13 @@ class SubClusterWrapper { iteration: this.iteration, }); - // Execute onComplete hook + // Execute onComplete hook - will throw if verification fails await this._executeHook('onComplete', { result: message, triggeringMessage: null, }); - // Clean up child cluster + // Only clean up and transition to idle if hook succeeded await this._stopChildCluster(); this.state = 'idle'; diff --git a/src/template-validation/index.js b/src/template-validation/index.js new file mode 100644 index 00000000..f3f02ac3 --- /dev/null +++ b/src/template-validation/index.js @@ -0,0 +1,89 @@ +const fs = require('node:fs'); +const path = require('node:path'); + +const { validateConfig } = require('../config-validator'); +const { simulateConsensusGates } = require('./simulate-consensus-gates'); +const { simulateTwoStageValidation } = require('./simulate-two-stage-validation'); + +function findJsonFiles(dir) { + const files = []; + if (!fs.existsSync(dir)) return files; + + for (const entry of fs.readdirSync(dir, { withFileTypes: true })) { + const fullPath = path.join(dir, entry.name); + if (entry.isDirectory()) { + files.push(...findJsonFiles(fullPath)); + } else if (entry.name.endsWith('.json')) { + files.push(fullPath); + } + } + return files; +} + +function inferTemplateIdFromPath(filePath) { + const base = path.basename(filePath, '.json'); + return base || 'unknown'; +} + +async function validateTemplateConfig({ config, templateId, deep }) { + const result = validateConfig(config); + + if (result.valid) { + const simErrors = []; + simErrors.push(...simulateConsensusGates(config)); + if (deep) { + simErrors.push(...(await simulateTwoStageValidation({ templateId, config }))); + } + if (simErrors.length > 0) { + result.valid = false; + result.errors.push(...simErrors); + } + } + + return result; +} + +async function validateTemplates({ templatesDir, deep = false }) { + const templateFiles = [...findJsonFiles(templatesDir)]; + + let hasErrors = false; + let validated = 0; + let skipped = 0; + const results = []; + + for (const filePath of templateFiles) { + try { + const content = fs.readFileSync(filePath, 'utf-8'); + const config = JSON.parse(content); + + // Skip non-cluster configs (like package.json) + if (!config.agents && !config.name) { + skipped++; + continue; + } + + const templateId = inferTemplateIdFromPath(filePath); + const result = await validateTemplateConfig({ config, templateId, deep }); + + results.push({ filePath, result }); + validated++; + if (!result.valid) hasErrors = true; + } catch (err) { + results.push({ filePath, result: { valid: false, errors: [err.message], warnings: [] } }); + validated++; + hasErrors = true; + } + } + + return { + valid: !hasErrors, + validated, + skipped, + results, + }; +} + +module.exports = { + validateTemplates, + validateTemplateConfig, +}; diff --git a/src/template-validation/simulate-consensus-gates.js b/src/template-validation/simulate-consensus-gates.js new file mode 100644 index 00000000..457085d6 --- /dev/null +++ b/src/template-validation/simulate-consensus-gates.js @@ -0,0 +1,159 @@ +const Ledger = require('../ledger'); +const MessageBus = require('../message-bus'); +const LogicEngine = require('../logic-engine'); + +function maybePublishStageStart({ messageBus, clusterId, logicScript }) { + const now = Date.now(); + + if (logicScript.includes('IMPLEMENTATION_READY')) { + messageBus.publish({ + cluster_id: clusterId, + topic: 'IMPLEMENTATION_READY', + sender: 'worker', + timestamp: now, + }); + } + + if (logicScript.includes('QUICK_VALIDATION_PASSED')) { + messageBus.publish({ + cluster_id: clusterId, + topic: 'QUICK_VALIDATION_PASSED', + sender: 'consensus-coordinator', + timestamp: now, + }); + } +} + +function collectTopicProducers(config) { + const producersByTopic = new Map(); + + for (const agent of config.agents || []) { + const hooks = agent?.hooks; + if (!hooks) continue; + + const onComplete = hooks.onComplete; + if (!onComplete) continue; + + if (onComplete.action === 'publish_message' && onComplete.config?.topic) { + const topic = String(onComplete.config.topic); + if (!producersByTopic.has(topic)) { + producersByTopic.set(topic, new Set()); + } + producersByTopic.get(topic).add(agent.id); + } + } + + return producersByTopic; +} + +/** + * Micro-sim: consensus-like trigger gates must not fire early due to duplicate publishes + * from the same producer (common in retries / double-publish bugs). + * + * Returns an array of error strings. + */ +function simulateConsensusGates(config) { + const agents = Array.isArray(config.agents) ? config.agents : []; + const producersByTopic = collectTopicProducers(config); + + const cluster = { + id: 'template-sim', + agents: agents.map((a) => ({ id: a.id, role: a.role })), + }; + + const failures = []; + + for (const agent of agents) { + const isConsensusLike = + agent?.role === 'coordinator' || + String(agent?.id || '').includes('consensus') || + String(agent?.id || '').includes('coordinator'); + + if (!isConsensusLike) continue; + + for (const trigger of agent.triggers || []) { + const topic = trigger?.topic; + const script = trigger?.logic?.script; + if (!topic || !script) continue; + + const producers = Array.from(producersByTopic.get(topic) || []); + if (producers.length < 2) continue; + + // Scenario A: Duplicate publishes from one producer MUST NOT satisfy the gate. + { + const ledger = new Ledger(':memory:'); + const messageBus = new MessageBus(ledger); + const logicEngine = new LogicEngine(messageBus, cluster); + + maybePublishStageStart({ messageBus, clusterId: cluster.id, logicScript: script }); + + messageBus.publish({ + cluster_id: cluster.id, + topic, + sender: producers[0], + content: { data: { approved: true } }, + }); + messageBus.publish({ + cluster_id: cluster.id, + topic, + sender: producers[0], + content: { data: { approved: true } }, + }); + + const shouldTriggerEarly = logicEngine.evaluate( + script, + { id: agent.id, cluster_id: cluster.id }, + { topic } + ); + ledger.close(); + + if (shouldTriggerEarly) { + failures.push( + `Agent "${agent.id}" trigger on "${topic}" fires early on duplicate sender (${producers[0]}). ` + + `Gate must require distinct producers: ${producers.join(', ')}` + ); + continue; + } + } + + // Scenario B: One publish from each producer SHOULD satisfy the gate. + { + const ledger = new Ledger(':memory:'); + const messageBus = new MessageBus(ledger); + const logicEngine = new LogicEngine(messageBus, cluster); + + maybePublishStageStart({ messageBus, clusterId: cluster.id, logicScript: script }); + + for (const producer of producers) { + messageBus.publish({ + cluster_id: cluster.id, + topic, + sender: producer, + content: { data: { approved: true } }, + }); + } + + const shouldTrigger = logicEngine.evaluate( + script, + { id: agent.id, cluster_id: cluster.id }, + { topic } + ); + ledger.close(); + + if (!shouldTrigger) { + failures.push( + `Agent "${agent.id}" trigger on "${topic}" did not fire after all producers published. ` + + `Expected producers: ${producers.join(', ')}` + ); + } + } + } + } + + return failures; +} + +module.exports = { + collectTopicProducers, + simulateConsensusGates, +}; diff --git a/src/template-validation/simulate-two-stage-validation.js b/src/template-validation/simulate-two-stage-validation.js new file mode 100644 index 00000000..b62f049b --- /dev/null +++ b/src/template-validation/simulate-two-stage-validation.js @@ -0,0 +1,270 @@ +const assert = require('node:assert'); + +const Ledger = require('../ledger'); +const MessageBus = require('../message-bus'); +const LogicEngine = require('../logic-engine'); +const { executeHook } = require('../agent/agent-hook-executor'); +const { parseResultOutput } = require('../agent/agent-task-executor'); + +function createSimAgent({ agentConfig, cluster, messageBus }) { + const simAgent = { + id: agentConfig.id, + role: agentConfig.role, + iteration: 1, + cluster, + messageBus, + config: agentConfig, + currentTaskId: 'sim-task', + workingDirectory: process.cwd(), + _log: () => {}, + _resolveProvider: () => 'claude', + _parseResultOutput: (output) => parseResultOutput(simAgent, output), + _publish: (message) => { + const receiver = message.receiver || 'broadcast'; + return messageBus.publish({ + ...message, + receiver, + cluster_id: cluster.id, + sender: simAgent.id, + }); + }, + }; + return simAgent; +} + +async function simulateQuickValidation({ config }) { + const cluster = { + id: 'quick-sim', + agents: config.agents.map((a) => ({ id: a.id, role: a.role })), + }; + + const coordinator = config.agents.find((a) => a.id === 'consensus-coordinator'); + assert.ok(coordinator, 'quick-validation: consensus-coordinator missing'); + + const trigger = coordinator.triggers.find((t) => t.topic === 'QUICK_VALIDATION_RESULT'); + assert.ok(trigger?.logic?.script, 'quick-validation: coordinator trigger logic missing'); + assert.ok(coordinator.hooks?.onComplete, 'quick-validation: coordinator onComplete missing'); + + const failures = []; + + const runScenario = async (allApproved) => { + const ledger = new Ledger(':memory:'); + const messageBus = new MessageBus(ledger); + const logicEngine = new LogicEngine(messageBus, cluster); + + // Stage start + const now = Date.now(); + messageBus.publish({ + cluster_id: cluster.id, + topic: 'IMPLEMENTATION_READY', + sender: 'worker', + timestamp: now, + }); + + // Validator outputs (stage 1) + messageBus.publish({ + cluster_id: cluster.id, + topic: 'QUICK_VALIDATION_RESULT', + sender: 'validator-requirements', + timestamp: now + 10, + content: { data: { approved: true, errors: ['req-error'] } }, + }); + messageBus.publish({ + cluster_id: cluster.id, + topic: 'QUICK_VALIDATION_RESULT', + sender: 'validator-code', + timestamp: now + 20, + content: { data: { approved: true, errors: ['code-error'] } }, + }); + + const gateOk = logicEngine.evaluate( + trigger.logic.script, + { id: 'consensus-coordinator', cluster_id: cluster.id }, + { topic: 'QUICK_VALIDATION_RESULT' } + ); + if (!gateOk) { + ledger.close(); + return { ok: false, error: 'quick-validation: gate did not open after both validators' }; + } + + const simAgent = createSimAgent({ agentConfig: coordinator, cluster, messageBus }); + const triggeringMessage = messageBus.findLast({ + cluster_id: cluster.id, + topic: 'QUICK_VALIDATION_RESULT', + }); + + try { + await executeHook({ + hook: coordinator.hooks.onComplete, + agent: simAgent, + message: triggeringMessage, + result: { + output: JSON.stringify({ allApproved, summary: allApproved ? 'ok' : 'nope' }), + success: true, + taskId: 'sim-task', + }, + messageBus, + cluster, + }); + } catch (err) { + ledger.close(); + return { ok: false, error: `quick-validation: onComplete failed: ${err.message}` }; + } + + const passed = messageBus.findLast({ + cluster_id: cluster.id, + topic: 'QUICK_VALIDATION_PASSED', + }); + const validationResult = messageBus.findLast({ + cluster_id: cluster.id, + topic: 'VALIDATION_RESULT', + }); + + ledger.close(); + + if (allApproved) { + if (!passed) + return { ok: false, error: 'quick-validation: expected QUICK_VALIDATION_PASSED' }; + return { ok: true }; + } + + if (!validationResult) { + return { ok: false, error: 'quick-validation: expected VALIDATION_RESULT on rejection' }; + } + + const errors = validationResult.content?.data?.errors || []; + if (!errors.includes('req-error') || !errors.includes('code-error')) { + return { ok: false, error: 'quick-validation: rejection did not aggregate validator errors' }; + } + + return { ok: true }; + }; + + for (const allApproved of [true, false]) { + const res = await runScenario(allApproved); + if (!res.ok) failures.push(res.error); + } + + return failures; +} + +async function simulateHeavyValidation({ config }) { + const cluster = { + id: 'heavy-sim', + agents: config.agents.map((a) => ({ id: a.id, role: a.role })), + }; + + const coordinator = config.agents.find((a) => a.id === 'consensus-coordinator'); + assert.ok(coordinator, 'heavy-validation: consensus-coordinator missing'); + + const trigger = coordinator.triggers.find((t) => t.topic === 'HEAVY_VALIDATION_RESULT'); + assert.ok(trigger?.logic?.script, 'heavy-validation: coordinator trigger logic missing'); + assert.ok(coordinator.hooks?.onComplete, 'heavy-validation: coordinator onComplete missing'); + + const failures = []; + + const runScenario = async (allApproved) => { + const ledger = new Ledger(':memory:'); + const messageBus = new MessageBus(ledger); + const logicEngine = new LogicEngine(messageBus, cluster); + + const now = Date.now(); + messageBus.publish({ + cluster_id: cluster.id, + topic: 'QUICK_VALIDATION_PASSED', + sender: 'consensus-coordinator', + timestamp: now, + }); + + messageBus.publish({ + cluster_id: cluster.id, + topic: 'HEAVY_VALIDATION_RESULT', + sender: 'validator-security', + timestamp: now + 10, + content: { data: { approved: true, errors: ['sec-error'] } }, + }); + messageBus.publish({ + cluster_id: cluster.id, + topic: 'HEAVY_VALIDATION_RESULT', + sender: 'validator-tester', + timestamp: now + 20, + content: { data: { approved: true, errors: ['test-error'] } }, + }); + + const gateOk = logicEngine.evaluate( + trigger.logic.script, + { id: 'consensus-coordinator', cluster_id: cluster.id }, + { topic: 'HEAVY_VALIDATION_RESULT' } + ); + if (!gateOk) { + ledger.close(); + return { ok: false, error: 'heavy-validation: gate did not open after both validators' }; + } + + const simAgent = createSimAgent({ agentConfig: coordinator, cluster, messageBus }); + const triggeringMessage = messageBus.findLast({ + cluster_id: cluster.id, + topic: 'HEAVY_VALIDATION_RESULT', + }); + + try { + await executeHook({ + hook: coordinator.hooks.onComplete, + agent: simAgent, + message: triggeringMessage, + result: { + output: JSON.stringify({ allApproved, summary: allApproved ? 'ok' : 'nope' }), + success: true, + taskId: 'sim-task', + }, + messageBus, + cluster, + }); + } catch (err) { + ledger.close(); + return { ok: false, error: `heavy-validation: onComplete failed: ${err.message}` }; + } + + const validationResult = messageBus.findLast({ + cluster_id: cluster.id, + topic: 'VALIDATION_RESULT', + }); + ledger.close(); + + if (!validationResult) { + return { ok: false, error: 'heavy-validation: expected VALIDATION_RESULT' }; + } + + const errors = validationResult.content?.data?.errors || []; + if (!errors.includes('sec-error') || !errors.includes('test-error')) { + return { ok: false, error: 'heavy-validation: did not aggregate validator errors' }; + } + + return { ok: true }; + }; + + for (const allApproved of [true, false]) { + const res = await runScenario(allApproved); + if (!res.ok) failures.push(res.error); + } + + return failures; +} + +/** + * Deep sim: run deterministic two-stage validation scenarios for base templates. + * Returns an array of error strings. + */ +function simulateTwoStageValidation({ templateId, config }) { + if (templateId === 'quick-validation') { + return simulateQuickValidation({ config }); + } + if (templateId === 'heavy-validation') { + return simulateHeavyValidation({ config }); + } + return Promise.resolve([]); +} + +module.exports = { + simulateTwoStageValidation, +}; diff --git a/src/tui-backend/index.ts b/src/tui-backend/index.ts new file mode 100644 index 00000000..5ca31305 --- /dev/null +++ b/src/tui-backend/index.ts @@ -0,0 +1,4 @@ +export * from './server'; +export * from './protocol'; +export * from './subscriptions'; +export * from './services'; diff --git a/src/tui-backend/protocol/constants.ts b/src/tui-backend/protocol/constants.ts new file mode 100644 index 00000000..db72c69a --- /dev/null +++ b/src/tui-backend/protocol/constants.ts @@ -0,0 +1,33 @@ +const PROTOCOL_VERSION = 1; +const MAX_FRAME_BYTES = 10 * 1024 * 1024; + +const RPC_ERROR_CODES = Object.freeze({ + PARSE_ERROR: -32700, + INVALID_REQUEST: -32600, + METHOD_NOT_FOUND: -32601, + INVALID_PARAMS: -32602, + INTERNAL_ERROR: -32603, + PROTOCOL_VERSION_MISMATCH: -32000, + ORCHESTRATOR_UNAVAILABLE: -32001, + CLUSTER_NOT_FOUND: -32002, + UNSUPPORTED_CAPABILITY: -32003, +}); + +const RPC_ERROR_MESSAGES = Object.freeze({ + [RPC_ERROR_CODES.PARSE_ERROR]: 'Parse error', + [RPC_ERROR_CODES.INVALID_REQUEST]: 'Invalid request', + [RPC_ERROR_CODES.METHOD_NOT_FOUND]: 'Method not found', + [RPC_ERROR_CODES.INVALID_PARAMS]: 'Invalid params', + [RPC_ERROR_CODES.INTERNAL_ERROR]: 'Internal error', + [RPC_ERROR_CODES.PROTOCOL_VERSION_MISMATCH]: 'Protocol version mismatch', + [RPC_ERROR_CODES.ORCHESTRATOR_UNAVAILABLE]: 'Orchestrator unavailable', + [RPC_ERROR_CODES.CLUSTER_NOT_FOUND]: 'Cluster not found', + [RPC_ERROR_CODES.UNSUPPORTED_CAPABILITY]: 'Unsupported capability', +}); + +module.exports = { + PROTOCOL_VERSION, + MAX_FRAME_BYTES, + RPC_ERROR_CODES, + RPC_ERROR_MESSAGES, +}; diff --git a/src/tui-backend/protocol/dispatcher.ts b/src/tui-backend/protocol/dispatcher.ts new file mode 100644 index 00000000..6d8362e0 --- /dev/null +++ b/src/tui-backend/protocol/dispatcher.ts @@ -0,0 +1,88 @@ +const { RPC_ERROR_CODES, RPC_ERROR_MESSAGES } = require('./constants'); + +const buildError = (code: number, message: string, detail?: string) => { + const error: any = { code, message }; + if (detail) { + error.data = { detail }; + } + return error; +}; + +const isRpcError = (error: any) => + error && + typeof error === 'object' && + typeof error.code === 'number' && + typeof error.message === 'string'; + +const createDispatcher = (options: any = {}) => { + const serverInfo = options.serverInfo || { name: 'zeroshot', version: '0.0.0' }; + const protocolVersion = typeof options.protocolVersion === 'number' ? options.protocolVersion : 1; + const baseHandlers = { + initialize: async () => ({ + protocolVersion, + server: serverInfo, + capabilities: { + methods: [], + notifications: [], + }, + }), + ping: async () => ({ ok: true }), + }; + const extraHandlers = + options.handlers && typeof options.handlers === 'object' ? options.handlers : {}; + const handlers = { ...baseHandlers, ...extraHandlers }; + const methods = Array.from(new Set(Object.keys(handlers))); + const notifications = Array.isArray(options.notifications) ? options.notifications : []; + handlers.initialize = async () => ({ + protocolVersion, + server: serverInfo, + capabilities: { + methods, + notifications, + }, + }); + + const dispatchRequest = async (message) => { + const handler = handlers[message.method]; + if (!handler) { + return { + ok: false, + error: buildError( + RPC_ERROR_CODES.METHOD_NOT_FOUND, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.METHOD_NOT_FOUND] + ), + }; + } + try { + const result = await handler(message.params ?? null, message); + return { ok: true, result }; + } catch (error) { + if (isRpcError(error)) { + const rpcError: any = { code: error.code, message: error.message }; + if (error.data) { + rpcError.data = error.data; + } + return { ok: false, error: rpcError }; + } + const detail = error instanceof Error ? error.message : 'Unhandled dispatcher error'; + return { + ok: false, + error: buildError( + RPC_ERROR_CODES.INTERNAL_ERROR, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INTERNAL_ERROR], + detail + ), + }; + } + }; + + return { + dispatchRequest, + methods, + notifications, + }; +}; + +module.exports = { + createDispatcher, +}; diff --git a/src/tui-backend/protocol/index.ts b/src/tui-backend/protocol/index.ts new file mode 100644 index 00000000..6b2f2023 --- /dev/null +++ b/src/tui-backend/protocol/index.ts @@ -0,0 +1,13 @@ +const constants = require('./constants'); +const validator = require('./validator'); +const dispatcher = require('./dispatcher'); +const framing = require('./stdio-framing'); + +export const PROTOCOL_VERSION = constants.PROTOCOL_VERSION; +export const MAX_FRAME_BYTES = constants.MAX_FRAME_BYTES; +export const RPC_ERROR_CODES = constants.RPC_ERROR_CODES; +export const RPC_ERROR_MESSAGES = constants.RPC_ERROR_MESSAGES; +export const createValidator = validator.createValidator; +export const createDispatcher = dispatcher.createDispatcher; +export const createFrameParser = framing.createFrameParser; +export const encodeFrame = framing.encodeFrame; diff --git a/src/tui-backend/protocol/schemas.ts b/src/tui-backend/protocol/schemas.ts new file mode 100644 index 00000000..e64b335a --- /dev/null +++ b/src/tui-backend/protocol/schemas.ts @@ -0,0 +1,599 @@ +const idSchema = { + anyOf: [{ type: 'string' }, { type: 'number' }], +}; + +const nullableString = { + anyOf: [{ type: 'string' }, { type: 'null' }], +}; + +const nullableNumber = { + anyOf: [{ type: 'number' }, { type: 'null' }], +}; + +const errorDataSchema = { + type: 'object', + additionalProperties: false, + properties: { + detail: { type: 'string' }, + fields: { + type: 'object', + additionalProperties: { type: 'string' }, + }, + supportedVersions: { type: 'array', items: { type: 'number' } }, + }, +}; + +const errorSchema = { + type: 'object', + additionalProperties: false, + required: ['code', 'message'], + properties: { + code: { type: 'number' }, + message: { type: 'string' }, + data: errorDataSchema, + }, +}; + +const jsonRpcRequestBase = { + type: 'object', + additionalProperties: false, + required: ['jsonrpc', 'id', 'method'], + properties: { + jsonrpc: { const: '2.0' }, + id: idSchema, + method: { type: 'string' }, + params: { type: ['object', 'array', 'null'] }, + }, +}; + +const jsonRpcNotificationBase = { + type: 'object', + additionalProperties: false, + required: ['jsonrpc', 'method'], + properties: { + jsonrpc: { const: '2.0' }, + method: { type: 'string' }, + params: { type: ['object', 'array', 'null'] }, + }, + not: { required: ['id'] }, +}; + +const jsonRpcResponseBase = { + type: 'object', + additionalProperties: false, + required: ['jsonrpc', 'id'], + properties: { + jsonrpc: { const: '2.0' }, + id: idSchema, + // Allow result/error at base layer; method-specific schema handles shape. + result: {}, + error: {}, + }, +}; + +const emptyParamsSchema = { + anyOf: [{ type: 'null' }, { type: 'object', additionalProperties: false, maxProperties: 0 }], +}; + +const clusterSummarySchema = { + type: 'object', + additionalProperties: false, + required: ['id', 'state', 'provider', 'createdAt', 'agentCount', 'messageCount', 'cwd'], + properties: { + id: { type: 'string' }, + state: { type: 'string' }, + provider: nullableString, + createdAt: { type: 'number' }, + agentCount: { type: 'number' }, + messageCount: { type: 'number' }, + cwd: nullableString, + }, +}; + +const clusterMetricsSchema = { + type: 'object', + additionalProperties: false, + required: ['id', 'supported', 'cpuPercent', 'memoryMB'], + properties: { + id: { type: 'string' }, + supported: { type: 'boolean' }, + cpuPercent: nullableNumber, + memoryMB: nullableNumber, + }, +}; + +const clusterLogLineSchema = { + type: 'object', + additionalProperties: false, + required: ['id', 'timestamp', 'text', 'agent', 'role', 'sender'], + properties: { + id: { type: 'string' }, + timestamp: { type: 'number' }, + text: { type: 'string' }, + agent: nullableString, + role: nullableString, + sender: nullableString, + }, +}; + +const timelineEventSchema = { + type: 'object', + additionalProperties: false, + required: ['id', 'timestamp', 'topic', 'label', 'approved', 'sender'], + properties: { + id: { type: 'string' }, + timestamp: { type: 'number' }, + topic: { type: 'string' }, + label: { type: 'string' }, + approved: { anyOf: [{ type: 'boolean' }, { type: 'null' }] }, + sender: nullableString, + }, +}; + +const topologyAgentSchema = { + type: 'object', + additionalProperties: false, + required: ['id', 'role'], + properties: { + id: { type: 'string' }, + role: nullableString, + }, +}; + +const topologyEdgeSchema = { + type: 'object', + additionalProperties: false, + required: ['from', 'to', 'topic', 'kind'], + properties: { + from: { type: 'string' }, + to: { type: 'string' }, + topic: { type: 'string' }, + kind: { enum: ['trigger', 'publish', 'source'] }, + dynamic: { type: 'boolean' }, + }, +}; + +const clusterTopologySchema = { + type: 'object', + additionalProperties: false, + required: ['agents', 'edges', 'topics'], + properties: { + agents: { type: 'array', items: topologyAgentSchema }, + edges: { type: 'array', items: topologyEdgeSchema }, + topics: { type: 'array', items: { type: 'string' } }, + }, +}; + +const guidanceDeliveryResultSchema = { + type: 'object', + additionalProperties: false, + required: ['status', 'reason', 'method'], + properties: { + status: { type: 'string' }, + reason: nullableString, + method: nullableString, + taskId: nullableString, + }, +}; + +const clusterGuidanceSummarySchema = { + type: 'object', + additionalProperties: false, + required: ['injected', 'queued', 'total'], + properties: { + injected: { type: 'number' }, + queued: { type: 'number' }, + total: { type: 'number' }, + }, +}; + +const clusterGuidanceDeliverySchema = { + type: 'object', + additionalProperties: false, + required: ['summary', 'agents', 'timestamp'], + properties: { + summary: clusterGuidanceSummarySchema, + agents: { + type: 'object', + additionalProperties: guidanceDeliveryResultSchema, + }, + timestamp: { type: 'number' }, + }, +}; + +const initializeParamsSchema = { + type: 'object', + additionalProperties: false, + required: ['protocolVersion', 'client'], + properties: { + protocolVersion: { type: 'number' }, + client: { + type: 'object', + additionalProperties: false, + required: ['name', 'version'], + properties: { + name: { type: 'string' }, + version: { type: 'string' }, + pid: { type: 'number' }, + }, + }, + capabilities: { + type: 'object', + additionalProperties: false, + properties: { + wantsMetrics: { type: 'boolean' }, + wantsTopology: { type: 'boolean' }, + }, + }, + }, +}; + +const initializeResultSchema = { + type: 'object', + additionalProperties: false, + required: ['protocolVersion', 'server', 'capabilities'], + properties: { + protocolVersion: { type: 'number' }, + server: { + type: 'object', + additionalProperties: false, + required: ['name', 'version'], + properties: { + name: { type: 'string' }, + version: { type: 'string' }, + }, + }, + capabilities: { + type: 'object', + additionalProperties: false, + required: ['methods', 'notifications'], + properties: { + methods: { type: 'array', items: { type: 'string' } }, + notifications: { type: 'array', items: { type: 'string' } }, + }, + }, + }, +}; + +const pingParamsSchema = emptyParamsSchema; + +const pingResultSchema = { + type: 'object', + additionalProperties: false, + required: ['ok'], + properties: { + ok: { const: true }, + }, +}; + +const listClustersResultSchema = { + type: 'object', + additionalProperties: false, + required: ['clusters'], + properties: { + clusters: { type: 'array', items: clusterSummarySchema }, + }, +}; + +const getClusterSummaryParamsSchema = { + type: 'object', + additionalProperties: false, + required: ['clusterId'], + properties: { + clusterId: { type: 'string' }, + }, +}; + +const getClusterSummaryResultSchema = { + type: 'object', + additionalProperties: false, + required: ['summary'], + properties: { + summary: clusterSummarySchema, + }, +}; + +const listClusterMetricsParamsSchema = { + type: 'object', + additionalProperties: false, + properties: { + clusterIds: { type: 'array', items: { type: 'string' } }, + }, +}; + +const listClusterMetricsResultSchema = { + type: 'object', + additionalProperties: false, + required: ['metrics'], + properties: { + metrics: { type: 'array', items: clusterMetricsSchema }, + }, +}; + +const startClusterFromTextParamsSchema = { + type: 'object', + additionalProperties: false, + required: ['text'], + properties: { + text: { type: 'string' }, + providerOverride: nullableString, + clusterId: { type: 'string' }, + }, +}; + +const startClusterFromIssueParamsSchema = { + type: 'object', + additionalProperties: false, + required: ['ref'], + properties: { + ref: { type: 'string' }, + providerOverride: nullableString, + clusterId: { type: 'string' }, + }, +}; + +const startClusterResultSchema = { + type: 'object', + additionalProperties: false, + required: ['clusterId'], + properties: { + clusterId: { type: 'string' }, + }, +}; + +const sendGuidanceToAgentParamsSchema = { + type: 'object', + additionalProperties: false, + required: ['clusterId', 'agentId', 'text'], + properties: { + clusterId: { type: 'string' }, + agentId: { type: 'string' }, + text: { type: 'string' }, + timeoutMs: { type: 'number' }, + }, +}; + +const sendGuidanceToClusterParamsSchema = { + type: 'object', + additionalProperties: false, + required: ['clusterId', 'text'], + properties: { + clusterId: { type: 'string' }, + text: { type: 'string' }, + timeoutMs: { type: 'number' }, + }, +}; + +const sendGuidanceToAgentResultSchema = { + type: 'object', + additionalProperties: false, + required: ['result'], + properties: { + result: guidanceDeliveryResultSchema, + }, +}; + +const sendGuidanceToClusterResultSchema = { + type: 'object', + additionalProperties: false, + required: ['result'], + properties: { + result: clusterGuidanceDeliverySchema, + }, +}; + +const subscribeClusterLogsParamsSchema = { + type: 'object', + additionalProperties: false, + required: ['clusterId'], + properties: { + clusterId: { type: 'string' }, + agentId: nullableString, + }, +}; + +const subscribeClusterTimelineParamsSchema = { + type: 'object', + additionalProperties: false, + required: ['clusterId'], + properties: { + clusterId: { type: 'string' }, + }, +}; + +const subscribeResultSchema = { + type: 'object', + additionalProperties: false, + required: ['subscriptionId'], + properties: { + subscriptionId: { type: 'string' }, + }, +}; + +const unsubscribeParamsSchema = { + type: 'object', + additionalProperties: false, + required: ['subscriptionId'], + properties: { + subscriptionId: { type: 'string' }, + }, +}; + +const unsubscribeResultSchema = { + type: 'object', + additionalProperties: false, + required: ['removed'], + properties: { + removed: { type: 'boolean' }, + }, +}; + +const getClusterTopologyParamsSchema = { + type: 'object', + additionalProperties: false, + required: ['clusterId'], + properties: { + clusterId: { type: 'string' }, + }, +}; + +const getClusterTopologyResultSchema = { + type: 'object', + additionalProperties: false, + required: ['topology'], + properties: { + topology: clusterTopologySchema, + }, +}; + +const clusterLogLinesNotificationSchema = { + type: 'object', + additionalProperties: false, + required: ['subscriptionId', 'clusterId', 'lines'], + properties: { + subscriptionId: { type: 'string' }, + clusterId: { type: 'string' }, + lines: { type: 'array', items: clusterLogLineSchema }, + droppedCount: { type: 'number' }, + }, +}; + +const clusterTimelineEventsNotificationSchema = { + type: 'object', + additionalProperties: false, + required: ['subscriptionId', 'clusterId', 'events'], + properties: { + subscriptionId: { type: 'string' }, + clusterId: { type: 'string' }, + events: { type: 'array', items: timelineEventSchema }, + droppedCount: { type: 'number' }, + }, +}; + +const buildRequestSchema = (method, paramsSchema, paramsRequired) => { + const schema = { + ...jsonRpcRequestBase, + properties: { + ...jsonRpcRequestBase.properties, + method: { const: method }, + }, + }; + if (paramsSchema) { + schema.properties.params = paramsSchema; + } + if (paramsRequired) { + schema.required = [...jsonRpcRequestBase.required, 'params']; + } + return schema; +}; + +const buildResponseSchema = (resultSchema) => ({ + ...jsonRpcResponseBase, + required: [...jsonRpcResponseBase.required, 'result'], + properties: { + ...jsonRpcResponseBase.properties, + result: resultSchema, + }, +}); + +const buildErrorResponseSchema = () => ({ + ...jsonRpcResponseBase, + required: [...jsonRpcResponseBase.required, 'error'], + properties: { + ...jsonRpcResponseBase.properties, + error: errorSchema, + }, +}); + +const buildNotificationSchema = (method, paramsSchema) => ({ + ...jsonRpcNotificationBase, + properties: { + ...jsonRpcNotificationBase.properties, + method: { const: method }, + params: paramsSchema, + }, + required: [...jsonRpcNotificationBase.required, 'params'], +}); + +const REQUEST_SCHEMAS = { + initialize: buildRequestSchema('initialize', initializeParamsSchema, true), + ping: buildRequestSchema('ping', pingParamsSchema, true), + listClusters: buildRequestSchema('listClusters', emptyParamsSchema, false), + getClusterSummary: buildRequestSchema('getClusterSummary', getClusterSummaryParamsSchema, true), + listClusterMetrics: buildRequestSchema( + 'listClusterMetrics', + listClusterMetricsParamsSchema, + false + ), + startClusterFromText: buildRequestSchema( + 'startClusterFromText', + startClusterFromTextParamsSchema, + true + ), + startClusterFromIssue: buildRequestSchema( + 'startClusterFromIssue', + startClusterFromIssueParamsSchema, + true + ), + sendGuidanceToAgent: buildRequestSchema( + 'sendGuidanceToAgent', + sendGuidanceToAgentParamsSchema, + true + ), + sendGuidanceToCluster: buildRequestSchema( + 'sendGuidanceToCluster', + sendGuidanceToClusterParamsSchema, + true + ), + subscribeClusterLogs: buildRequestSchema( + 'subscribeClusterLogs', + subscribeClusterLogsParamsSchema, + true + ), + subscribeClusterTimeline: buildRequestSchema( + 'subscribeClusterTimeline', + subscribeClusterTimelineParamsSchema, + true + ), + unsubscribe: buildRequestSchema('unsubscribe', unsubscribeParamsSchema, true), + getClusterTopology: buildRequestSchema( + 'getClusterTopology', + getClusterTopologyParamsSchema, + true + ), +}; + +const RESPONSE_SCHEMAS = { + initialize: buildResponseSchema(initializeResultSchema), + ping: buildResponseSchema(pingResultSchema), + listClusters: buildResponseSchema(listClustersResultSchema), + getClusterSummary: buildResponseSchema(getClusterSummaryResultSchema), + listClusterMetrics: buildResponseSchema(listClusterMetricsResultSchema), + startClusterFromText: buildResponseSchema(startClusterResultSchema), + startClusterFromIssue: buildResponseSchema(startClusterResultSchema), + sendGuidanceToAgent: buildResponseSchema(sendGuidanceToAgentResultSchema), + sendGuidanceToCluster: buildResponseSchema(sendGuidanceToClusterResultSchema), + subscribeClusterLogs: buildResponseSchema(subscribeResultSchema), + subscribeClusterTimeline: buildResponseSchema(subscribeResultSchema), + unsubscribe: buildResponseSchema(unsubscribeResultSchema), + getClusterTopology: buildResponseSchema(getClusterTopologyResultSchema), +}; + +const NOTIFICATION_SCHEMAS = { + clusterLogLines: buildNotificationSchema('clusterLogLines', clusterLogLinesNotificationSchema), + clusterTimelineEvents: buildNotificationSchema( + 'clusterTimelineEvents', + clusterTimelineEventsNotificationSchema + ), +}; + +module.exports = { + errorSchema, + jsonRpcRequestBase, + jsonRpcNotificationBase, + jsonRpcResponseBase, + buildErrorResponseSchema, + REQUEST_SCHEMAS, + RESPONSE_SCHEMAS, + NOTIFICATION_SCHEMAS, +}; diff --git a/src/tui-backend/protocol/stdio-framing.ts b/src/tui-backend/protocol/stdio-framing.ts new file mode 100644 index 00000000..275e01a0 --- /dev/null +++ b/src/tui-backend/protocol/stdio-framing.ts @@ -0,0 +1,83 @@ +const { MAX_FRAME_BYTES } = require('./constants'); + +const HEADER_DELIMITER = '\r\n\r\n'; + +const parseContentLength = (headerText) => { + const lines = headerText.split('\r\n'); + for (const line of lines) { + const sepIndex = line.indexOf(':'); + if (sepIndex === -1) { + continue; + } + const name = line.slice(0, sepIndex).trim().toLowerCase(); + if (name !== 'content-length') { + continue; + } + const value = line.slice(sepIndex + 1).trim(); + const length = Number.parseInt(value, 10); + if (!Number.isFinite(length) || length < 0) { + throw new Error('Invalid Content-Length header'); + } + return length; + } + throw new Error('Missing Content-Length header'); +}; + +const createFrameParser = (options: any = {}) => { + const maxFrameBytes = + typeof options.maxFrameBytes === 'number' ? options.maxFrameBytes : MAX_FRAME_BYTES; + let buffer = Buffer.alloc(0); + + const reset = () => { + buffer = Buffer.alloc(0); + }; + + const push = (chunk) => { + if (!chunk || chunk.length === 0) { + return []; + } + buffer = Buffer.concat([buffer, chunk]); + const frames = []; + + while (true) { + const headerIndex = buffer.indexOf(HEADER_DELIMITER); + if (headerIndex === -1) { + break; + } + const headerText = buffer.slice(0, headerIndex).toString('utf8'); + const contentLength = parseContentLength(headerText); + if (contentLength > maxFrameBytes) { + throw new Error('Frame exceeds maximum size'); + } + + const totalLength = headerIndex + HEADER_DELIMITER.length + contentLength; + if (buffer.length < totalLength) { + break; + } + + const payload = buffer.slice(headerIndex + HEADER_DELIMITER.length, totalLength); + frames.push(payload.toString('utf8')); + buffer = buffer.slice(totalLength); + } + + return frames; + }; + + return { push, reset }; +}; + +const encodeFrame = (payload) => { + const payloadBuffer = Buffer.isBuffer(payload) + ? payload + : Buffer.from(typeof payload === 'string' ? payload : JSON.stringify(payload), 'utf8'); + if (payloadBuffer.length > MAX_FRAME_BYTES) { + throw new Error('Frame exceeds maximum size'); + } + const header = `Content-Length: ${payloadBuffer.length}${HEADER_DELIMITER}`; + return Buffer.concat([Buffer.from(header, 'utf8'), payloadBuffer]); +}; + +module.exports = { + createFrameParser, + encodeFrame, +}; diff --git a/src/tui-backend/protocol/types.ts b/src/tui-backend/protocol/types.ts new file mode 100644 index 00000000..32792c2f --- /dev/null +++ b/src/tui-backend/protocol/types.ts @@ -0,0 +1,177 @@ +export type JsonRpcId = string | number; + +export type RpcErrorData = { + detail?: string; + fields?: Record; + supportedVersions?: number[]; +}; + +export type RpcError = { + code: number; + message: string; + data?: RpcErrorData; +}; + +export type JsonRpcRequest = { + jsonrpc: '2.0'; + id: JsonRpcId; + method: string; + params?: TParams | null; +}; + +export type JsonRpcNotification = { + jsonrpc: '2.0'; + method: string; + params?: TParams | null; +}; + +export type JsonRpcSuccessResponse = { + jsonrpc: '2.0'; + id: JsonRpcId; + result: TResult; +}; + +export type JsonRpcErrorResponse = { + jsonrpc: '2.0'; + id: JsonRpcId; + error: RpcError; +}; + +export type ClusterSummary = { + id: string; + state: string; + provider: string | null; + createdAt: number; + agentCount: number; + messageCount: number; + cwd: string | null; +}; + +export type ClusterMetrics = { + id: string; + supported: boolean; + cpuPercent: number | null; + memoryMB: number | null; +}; + +export type ClusterLogLine = { + id: string; + timestamp: number; + text: string; + agent: string | null; + role: string | null; + sender: string | null; +}; + +export type TimelineEvent = { + id: string; + timestamp: number; + topic: string; + label: string; + approved: boolean | null; + sender: string | null; +}; + +export type TopologyAgent = { + id: string; + role: string | null; +}; + +export type TopologyEdge = { + from: string; + to: string; + topic: string; + kind: 'trigger' | 'publish' | 'source'; + dynamic?: boolean; +}; + +export type ClusterTopology = { + agents: TopologyAgent[]; + edges: TopologyEdge[]; + topics: string[]; +}; + +export type GuidanceDeliveryResult = { + status: string; + reason: string | null; + method: string | null; + taskId?: string | null; +}; + +export type ClusterGuidanceSummary = { + injected: number; + queued: number; + total: number; +}; + +export type ClusterGuidanceDelivery = { + summary: ClusterGuidanceSummary; + agents: Record; + timestamp: number; +}; + +export type InitializeParams = { + protocolVersion: number; + client: { name: string; version: string; pid?: number }; + capabilities?: { wantsMetrics?: boolean; wantsTopology?: boolean }; +}; + +export type InitializeResult = { + protocolVersion: number; + server: { name: string; version: string }; + capabilities: { methods: string[]; notifications: string[] }; +}; + +export type PingParams = Record | null; +export type PingResult = { ok: true }; + +export type ListClustersResult = { clusters: ClusterSummary[] }; +export type GetClusterSummaryParams = { clusterId: string }; +export type GetClusterSummaryResult = { summary: ClusterSummary }; +export type ListClusterMetricsParams = { clusterIds?: string[] }; +export type ListClusterMetricsResult = { metrics: ClusterMetrics[] }; +export type StartClusterFromTextParams = { + text: string; + providerOverride?: string | null; + clusterId?: string; +}; +export type StartClusterFromIssueParams = { + ref: string; + providerOverride?: string | null; + clusterId?: string; +}; +export type StartClusterResult = { clusterId: string }; +export type SendGuidanceToAgentParams = { + clusterId: string; + agentId: string; + text: string; + timeoutMs?: number; +}; +export type SendGuidanceToClusterParams = { + clusterId: string; + text: string; + timeoutMs?: number; +}; +export type SendGuidanceToAgentResult = { result: GuidanceDeliveryResult }; +export type SendGuidanceToClusterResult = { result: ClusterGuidanceDelivery }; +export type SubscribeClusterLogsParams = { clusterId: string; agentId?: string | null }; +export type SubscribeClusterTimelineParams = { clusterId: string }; +export type SubscribeResult = { subscriptionId: string }; +export type UnsubscribeParams = { subscriptionId: string }; +export type UnsubscribeResult = { removed: boolean }; +export type GetClusterTopologyParams = { clusterId: string }; +export type GetClusterTopologyResult = { topology: ClusterTopology }; + +export type ClusterLogLinesNotification = { + subscriptionId: string; + clusterId: string; + lines: ClusterLogLine[]; + droppedCount?: number; +}; + +export type ClusterTimelineEventsNotification = { + subscriptionId: string; + clusterId: string; + events: TimelineEvent[]; + droppedCount?: number; +}; diff --git a/src/tui-backend/protocol/validator.ts b/src/tui-backend/protocol/validator.ts new file mode 100644 index 00000000..9cb823bc --- /dev/null +++ b/src/tui-backend/protocol/validator.ts @@ -0,0 +1,250 @@ +const Ajv = require('ajv'); +const { PROTOCOL_VERSION, RPC_ERROR_CODES, RPC_ERROR_MESSAGES } = require('./constants'); +const { + errorSchema, + jsonRpcRequestBase, + jsonRpcNotificationBase, + buildErrorResponseSchema, + REQUEST_SCHEMAS, + RESPONSE_SCHEMAS, + NOTIFICATION_SCHEMAS, +} = require('./schemas'); + +const buildError = (code, message, errors = []) => { + const data = Object.create(null); + if (errors && errors.length) { + const detail = errors + .map((err) => { + const path = err.instancePath || err.schemaPath || ''; + return path ? `${path} ${err.message}` : err.message; + }) + .join('; '); + if (detail) { + data.detail = detail; + } + const fields = Object.create(null); + for (const err of errors) { + const key = err.instancePath || err.schemaPath || ''; + if (key && !fields[key]) { + fields[key] = err.message || 'invalid'; + } + } + if (Object.keys(fields).length) { + data.fields = fields; + } + } + const error = Object.create(null); + error.code = code; + error.message = message; + if (Object.keys(data).length) { + error.data = data; + } + return error; +}; + +const compileSchemaMap = (ajv, schemas) => { + const validators = new Map(); + for (const [key, schema] of Object.entries(schemas)) { + validators.set(key, /** @type {any} */ ajv.compile(schema)); + } + return validators; +}; + +const createValidator = () => { + const ajv = new Ajv({ + allErrors: true, + strict: false, + coerceTypes: false, + removeAdditional: false, + }); + + const validateRequestBase = /** @type {any} */ ajv.compile(jsonRpcRequestBase); + const validateNotificationBase = /** @type {any} */ ajv.compile(jsonRpcNotificationBase); + const validateErrorObject = /** @type {any} */ ajv.compile(errorSchema); + const validateErrorResponse = /** @type {any} */ ajv.compile(buildErrorResponseSchema()); + + const requestValidators = compileSchemaMap(ajv, REQUEST_SCHEMAS); + const responseValidators = compileSchemaMap(ajv, RESPONSE_SCHEMAS); + const notificationValidators = compileSchemaMap(ajv, NOTIFICATION_SCHEMAS); + + const validateRequest = (message) => { + if (!validateRequestBase(message)) { + return { + ok: false, + error: buildError( + RPC_ERROR_CODES.INVALID_REQUEST, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INVALID_REQUEST], + validateRequestBase.errors || [] + ), + }; + } + + const validator = requestValidators.get(message.method); + if (!validator) { + return { + ok: false, + error: buildError( + RPC_ERROR_CODES.METHOD_NOT_FOUND, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.METHOD_NOT_FOUND] + ), + }; + } + + if (!validator(message)) { + return { + ok: false, + error: buildError( + RPC_ERROR_CODES.INVALID_PARAMS, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INVALID_PARAMS], + validator.errors || [] + ), + }; + } + + if ( + message.method === 'initialize' && + message.params && + message.params.protocolVersion !== PROTOCOL_VERSION + ) { + const error = buildError( + RPC_ERROR_CODES.PROTOCOL_VERSION_MISMATCH, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.PROTOCOL_VERSION_MISMATCH] + ); + error.data = { + ...(error.data || {}), + supportedVersions: [PROTOCOL_VERSION], + }; + return { + ok: false, + error, + }; + } + + return { ok: true, value: message }; + }; + + const validateNotification = (message) => { + if (!validateNotificationBase(message)) { + return { + ok: false, + error: buildError( + RPC_ERROR_CODES.INVALID_REQUEST, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INVALID_REQUEST], + validateNotificationBase.errors || [] + ), + }; + } + + const validator = notificationValidators.get(message.method); + if (!validator) { + return { + ok: false, + error: buildError( + RPC_ERROR_CODES.METHOD_NOT_FOUND, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.METHOD_NOT_FOUND] + ), + }; + } + + if (!validator(message)) { + return { + ok: false, + error: buildError( + RPC_ERROR_CODES.INVALID_PARAMS, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INVALID_PARAMS], + validator.errors || [] + ), + }; + } + + return { ok: true, value: message }; + }; + + const isValidId = (id) => typeof id === 'string' || typeof id === 'number'; + const isObject = (value) => value !== null && typeof value === 'object' && !Array.isArray(value); + + const validateResponse = (message, method) => { + if (!isObject(message) || message.jsonrpc !== '2.0' || !isValidId(message.id)) { + return { + ok: false, + error: buildError( + RPC_ERROR_CODES.INVALID_REQUEST, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INVALID_REQUEST], + [] + ), + }; + } + + const hasError = Object.prototype.hasOwnProperty.call(message, 'error'); + const hasResult = Object.prototype.hasOwnProperty.call(message, 'result'); + if ((hasError && hasResult) || (!hasError && !hasResult)) { + return { + ok: false, + error: buildError( + RPC_ERROR_CODES.INVALID_REQUEST, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INVALID_REQUEST] + ), + }; + } + + if (hasError) { + if (!validateErrorResponse(message)) { + return { + ok: false, + error: buildError( + RPC_ERROR_CODES.INVALID_REQUEST, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INVALID_REQUEST], + validateErrorResponse.errors || [] + ), + }; + } + if (!validateErrorObject(message.error)) { + return { + ok: false, + error: buildError( + RPC_ERROR_CODES.INVALID_REQUEST, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INVALID_REQUEST], + validateErrorObject.errors || [] + ), + }; + } + return { ok: true, value: message }; + } + + const validator = responseValidators.get(method); + if (!validator) { + return { + ok: false, + error: buildError( + RPC_ERROR_CODES.METHOD_NOT_FOUND, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.METHOD_NOT_FOUND] + ), + }; + } + + if (!validator(message)) { + return { + ok: false, + error: buildError( + RPC_ERROR_CODES.INVALID_PARAMS, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INVALID_PARAMS], + validator.errors || [] + ), + }; + } + + return { ok: true, value: message }; + }; + + return { + validateRequest, + validateNotification, + validateResponse, + }; +}; + +module.exports = { + createValidator, + RPC_ERROR_CODES, + PROTOCOL_VERSION, +}; diff --git a/src/tui-backend/server.ts b/src/tui-backend/server.ts new file mode 100644 index 00000000..b4424ad8 --- /dev/null +++ b/src/tui-backend/server.ts @@ -0,0 +1,504 @@ +const MOCK_LAUNCH_ENV = 'ZEROSHOT_TUI_BACKEND_MOCK_LAUNCH'; +const MOCK_GUIDANCE_ENV = 'ZEROSHOT_TUI_BACKEND_MOCK_GUIDANCE'; +const METRICS_PLATFORM_ENV = 'ZEROSHOT_TUI_BACKEND_METRICS_PLATFORM'; + +const path = require('path'); +const { + createValidator, + createDispatcher, + createFrameParser, + encodeFrame, + RPC_ERROR_CODES, + RPC_ERROR_MESSAGES, + PROTOCOL_VERSION, +} = require('./protocol'); +const { + listClusters, + getClusterSummary, + listClusterMetrics, + ClusterNotFoundError, +} = require('./services/cluster-registry'); +const { getClusterTopology } = require('./services/cluster-topology'); +const { + launchClusterFromText, + launchClusterFromIssue, + InvalidIssueReferenceError, +} = require('./services/cluster-launcher'); +const { createClusterLogStream, MAX_LOG_LINES } = require('./services/cluster-logs'); +const { createClusterTimelineStream, MAX_TIMELINE_EVENTS } = require('./services/cluster-timeline'); +const { sendAgentGuidance, sendClusterGuidance } = require('./services/guidance-delivery'); +const { createSubscriptionRegistry } = require('./subscriptions'); + +const isValidId = (value) => typeof value === 'string' || typeof value === 'number'; + +const isMockLaunchEnabled = () => process.env[MOCK_LAUNCH_ENV] === '1'; +const isMockGuidanceEnabled = () => process.env[MOCK_GUIDANCE_ENV] === '1'; + +const createMockLauncherDeps = () => ({ + getOrchestrator: async () => ({}), + loadSettings: () => ({ defaultConfig: 'conductor-bootstrap', providerSettings: {} }), + resolveConfigPath: () => 'mock-config', + loadClusterConfig: () => ({}), + startClusterFromText: async () => {}, + startClusterFromIssue: async () => {}, +}); + +const createMockGuidanceDeps = () => ({ + getOrchestrator: async () => ({ + sendGuidanceToAgent: async (clusterId, agentId) => ({ + status: 'injected', + reason: null, + method: 'pty', + taskId: `task-${agentId}`, + }), + sendGuidanceToCluster: async () => ({ + summary: { injected: 1, queued: 1, total: 2 }, + agents: { + 'mock-agent-1': { + status: 'injected', + reason: null, + method: 'pty', + taskId: 'task-mock-agent-1', + }, + 'mock-agent-2': { + status: 'queued', + reason: 'queued', + method: null, + taskId: 'task-mock-agent-2', + }, + }, + timestamp: 1700000000000, + }), + }), +}); + +const loadPackageInfo = () => { + try { + const packagePath = path.resolve(__dirname, '..', '..', 'package.json'); + const pkg = require(packagePath); + return { + name: typeof pkg.name === 'string' ? pkg.name : 'zeroshot', + version: typeof pkg.version === 'string' ? pkg.version : '0.0.0', + }; + } catch (error) { + return { name: 'zeroshot', version: '0.0.0' }; + } +}; + +const writeFrame = (payload) => { + const framed = encodeFrame(payload); + process.stdout.write(framed); +}; + +const writeError = (id, error) => { + writeFrame({ + jsonrpc: '2.0', + id, + error, + }); +}; + +const buildRpcError = (code, message, detail) => + detail ? { code, message, data: { detail } } : { code, message }; + +const logDiagnostic = (message, error) => { + const details = error instanceof Error ? `${message}: ${error.stack || error.message}` : message; + process.stderr.write(`${details}\n`); +}; + +const isNonEmptyString = (value) => typeof value === 'string' && value.trim().length > 0; + +const resolveMetricsPlatformOverride = () => { + const value = process.env[METRICS_PLATFORM_ENV]; + return isNonEmptyString(value) ? value : null; +}; + +const isRpcError = (error) => + error && + typeof error === 'object' && + typeof error.code === 'number' && + typeof error.message === 'string'; + +const isGuidanceInvalidParamsError = (message) => + message.includes('is required') || + message.includes('non-empty string') || + message.includes('agent not found'); + +const isGuidanceClusterNotFoundError = (message) => message.includes('cluster not found'); + +const isTopologyClusterNotFoundError = (error) => + error instanceof Error && /cluster/i.test(error.message) && /not found/i.test(error.message); + +const validateGuidanceText = (text) => { + if (!isNonEmptyString(text)) { + throw buildRpcError( + RPC_ERROR_CODES.INVALID_PARAMS, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INVALID_PARAMS], + 'text must be a non-empty string' + ); + } +}; + +const validateGuidanceId = (value, label) => { + if (!isNonEmptyString(value)) { + throw buildRpcError( + RPC_ERROR_CODES.INVALID_PARAMS, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INVALID_PARAMS], + `${label} must be a non-empty string` + ); + } +}; + +const resolveGuidanceError = (error) => { + if (isRpcError(error)) { + return error; + } + const message = error instanceof Error ? error.message : 'Guidance delivery error'; + if (isGuidanceClusterNotFoundError(message)) { + return buildRpcError( + RPC_ERROR_CODES.CLUSTER_NOT_FOUND, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.CLUSTER_NOT_FOUND], + message + ); + } + if (isGuidanceInvalidParamsError(message)) { + return buildRpcError( + RPC_ERROR_CODES.INVALID_PARAMS, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INVALID_PARAMS], + message + ); + } + return buildRpcError( + RPC_ERROR_CODES.INTERNAL_ERROR, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INTERNAL_ERROR], + message + ); +}; + +const capPayload = (items, maxItems) => { + if (!Array.isArray(items)) { + return { items: [], droppedCount: 0 }; + } + if (items.length <= maxItems) { + return { items, droppedCount: 0 }; + } + const trimmed = items.slice(items.length - maxItems); + return { items: trimmed, droppedCount: items.length - trimmed.length }; +}; + +const startServer = () => { + const registry = createSubscriptionRegistry(); + const notifications = ['clusterLogLines', 'clusterTimelineEvents']; + let shuttingDown = false; + const validator = createValidator(); + const dispatcher = createDispatcher({ + serverInfo: loadPackageInfo(), + protocolVersion: PROTOCOL_VERSION, + notifications, + handlers: { + listClusters: async () => ({ + clusters: await listClusters(), + }), + getClusterSummary: async (params) => { + try { + const summary = await getClusterSummary({ + clusterId: params.clusterId, + }); + return { summary }; + } catch (error) { + if (error instanceof ClusterNotFoundError) { + throw buildRpcError( + RPC_ERROR_CODES.CLUSTER_NOT_FOUND, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.CLUSTER_NOT_FOUND], + error.message + ); + } + throw error; + } + }, + getClusterTopology: async (params) => { + try { + const topology = await getClusterTopology(params.clusterId); + return { topology }; + } catch (error) { + if (isTopologyClusterNotFoundError(error)) { + throw buildRpcError( + RPC_ERROR_CODES.CLUSTER_NOT_FOUND, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.CLUSTER_NOT_FOUND], + error instanceof Error ? error.message : 'Cluster not found' + ); + } + throw error; + } + }, + listClusterMetrics: async (params) => { + const clusterIds = Array.isArray(params?.clusterIds) ? params.clusterIds : undefined; + const platformOverride = resolveMetricsPlatformOverride(); + const metricsById = await listClusterMetrics({ + clusterIds, + deps: platformOverride ? { platform: platformOverride } : undefined, + }); + const metrics = Array.isArray(clusterIds) + ? clusterIds.map((id) => metricsById[id]).filter(Boolean) + : Object.values(metricsById); + return { metrics }; + }, + startClusterFromText: async (params) => { + try { + const result = await launchClusterFromText({ + text: params.text, + providerOverride: params.providerOverride ?? null, + clusterId: params.clusterId, + deps: isMockLaunchEnabled() ? createMockLauncherDeps() : undefined, + }); + return result; + } catch (error) { + throw buildRpcError( + RPC_ERROR_CODES.INTERNAL_ERROR, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INTERNAL_ERROR], + error instanceof Error ? error.message : 'Launcher error' + ); + } + }, + startClusterFromIssue: async (params) => { + try { + const result = await launchClusterFromIssue({ + ref: params.ref, + providerOverride: params.providerOverride ?? null, + clusterId: params.clusterId, + deps: isMockLaunchEnabled() ? createMockLauncherDeps() : undefined, + }); + return result; + } catch (error) { + if (error instanceof InvalidIssueReferenceError) { + throw buildRpcError( + RPC_ERROR_CODES.INVALID_PARAMS, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INVALID_PARAMS], + error.message + ); + } + throw buildRpcError( + RPC_ERROR_CODES.INTERNAL_ERROR, + RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INTERNAL_ERROR], + error instanceof Error ? error.message : 'Launcher error' + ); + } + }, + sendGuidanceToAgent: async (params) => { + try { + validateGuidanceId(params?.clusterId, 'clusterId'); + validateGuidanceId(params?.agentId, 'agentId'); + validateGuidanceText(params?.text); + const result = await sendAgentGuidance({ + clusterId: params.clusterId, + agentId: params.agentId, + text: params.text, + timeoutMs: params.timeoutMs, + deps: isMockGuidanceEnabled() ? createMockGuidanceDeps() : undefined, + }); + return { result }; + } catch (error) { + throw resolveGuidanceError(error); + } + }, + sendGuidanceToCluster: async (params) => { + try { + validateGuidanceId(params?.clusterId, 'clusterId'); + validateGuidanceText(params?.text); + const result = await sendClusterGuidance({ + clusterId: params.clusterId, + text: params.text, + timeoutMs: params.timeoutMs, + deps: isMockGuidanceEnabled() ? createMockGuidanceDeps() : undefined, + }); + return { result }; + } catch (error) { + throw resolveGuidanceError(error); + } + }, + subscribeClusterLogs: async (params) => { + const clusterId = params.clusterId; + const agentId = params.agentId ?? null; + let subscriptionId = ''; + const stream = createClusterLogStream({ + clusterId, + agentId, + maxInitialLines: MAX_LOG_LINES * 5, + onLines: (lines) => { + if (!subscriptionId) return; + const { items, droppedCount } = capPayload(lines, MAX_LOG_LINES); + if (!items.length) return; + const payload = + droppedCount > 0 + ? { + subscriptionId, + clusterId, + lines: items, + droppedCount, + } + : { + subscriptionId, + clusterId, + lines: items, + }; + writeFrame({ + jsonrpc: '2.0', + method: 'clusterLogLines', + params: payload, + }); + }, + }); + subscriptionId = registry.add('clusterLogs', () => stream.close()); + stream.start(); + return { subscriptionId }; + }, + subscribeClusterTimeline: async (params) => { + const clusterId = params.clusterId; + let subscriptionId = ''; + const stream = createClusterTimelineStream({ + clusterId, + maxInitialEvents: MAX_TIMELINE_EVENTS * 5, + onEvents: (events) => { + if (!subscriptionId) return; + const { items, droppedCount } = capPayload(events, MAX_TIMELINE_EVENTS); + if (!items.length) return; + const payload = + droppedCount > 0 + ? { + subscriptionId, + clusterId, + events: items, + droppedCount, + } + : { + subscriptionId, + clusterId, + events: items, + }; + writeFrame({ + jsonrpc: '2.0', + method: 'clusterTimelineEvents', + params: payload, + }); + }, + }); + subscriptionId = registry.add('clusterTimeline', () => stream.close()); + stream.start(); + return { subscriptionId }; + }, + unsubscribe: async (params) => registry.unsubscribe(params.subscriptionId), + }, + }); + const parser = createFrameParser(); + + const shutdown = (code) => { + if (shuttingDown) return; + shuttingDown = true; + registry.closeAll(); + process.exit(code); + }; + + const handleFrame = async (payload) => { + let message; + try { + message = JSON.parse(payload); + } catch (error) { + writeError(null, { + code: RPC_ERROR_CODES.PARSE_ERROR, + message: RPC_ERROR_MESSAGES[RPC_ERROR_CODES.PARSE_ERROR], + }); + logDiagnostic('Invalid JSON payload', error); + return; + } + + if (!message || typeof message !== 'object') { + writeError(null, { + code: RPC_ERROR_CODES.INVALID_REQUEST, + message: RPC_ERROR_MESSAGES[RPC_ERROR_CODES.INVALID_REQUEST], + }); + return; + } + + const hasId = Object.prototype.hasOwnProperty.call(message, 'id'); + if (!hasId) { + const notification = validator.validateNotification(message); + if (!notification.ok) { + logDiagnostic('Invalid notification received', notification.error); + } + return; + } + + const requestValidation = validator.validateRequest(message); + if (!requestValidation.ok) { + const responseId = isValidId(message.id) ? message.id : null; + writeError(responseId, requestValidation.error); + return; + } + + const dispatchResult = await dispatcher.dispatchRequest(requestValidation.value); + if (!dispatchResult.ok) { + writeError(message.id, dispatchResult.error); + return; + } + + writeFrame({ + jsonrpc: '2.0', + id: message.id, + result: dispatchResult.result, + }); + }; + + const handleChunk = (chunk) => { + let frames = []; + try { + frames = parser.push(chunk); + } catch (error) { + parser.reset(); + writeError(null, { + code: RPC_ERROR_CODES.PARSE_ERROR, + message: RPC_ERROR_MESSAGES[RPC_ERROR_CODES.PARSE_ERROR], + data: { detail: error instanceof Error ? error.message : 'Parse error' }, + }); + logDiagnostic('Frame parsing failed', error); + return; + } + + for (const frame of frames) { + void handleFrame(frame); + } + }; + + process.stdin.on('data', handleChunk); + process.stdin.on('end', () => { + shutdown(0); + }); + process.stdin.on('error', (error) => { + logDiagnostic('Stdin error', error); + shutdown(1); + }); + + process.on('uncaughtException', (error) => { + logDiagnostic('Uncaught exception', error); + shutdown(1); + }); + process.on('unhandledRejection', (error) => { + logDiagnostic('Unhandled rejection', error); + shutdown(1); + }); + process.on('exit', () => { + if (!shuttingDown) { + shuttingDown = true; + registry.closeAll(); + } + }); + + process.stdin.resume(); +}; + +if (require.main === module) { + startServer(); +} + +module.exports = { + startServer, +}; diff --git a/src/tui-backend/services/cluster-launcher.ts b/src/tui-backend/services/cluster-launcher.ts new file mode 100644 index 00000000..217e123a --- /dev/null +++ b/src/tui-backend/services/cluster-launcher.ts @@ -0,0 +1,127 @@ +import { loadSettings } from '../../../lib/settings'; +import { + detectRunInput, + loadClusterConfig, + resolveConfigPath, + startClusterFromIssue, + startClusterFromText, +} from '../../../lib/start-cluster'; + +const { generateName } = require('../../../src/name-generator'); + +type ClusterLauncherDeps = { + getOrchestrator?: () => Promise; + loadSettings?: typeof loadSettings; + resolveConfigPath?: typeof resolveConfigPath; + loadClusterConfig?: typeof loadClusterConfig; + startClusterFromText?: typeof startClusterFromText; + startClusterFromIssue?: typeof startClusterFromIssue; + detectRunInput?: typeof detectRunInput; + generateClusterId?: () => string; +}; + +let orchestratorPromise: Promise | null = null; + +export class InvalidIssueReferenceError extends Error { + constructor(ref: string) { + super(`Invalid issue reference: ${ref}`); + this.name = 'InvalidIssueReferenceError'; + Object.setPrototypeOf(this, new.target.prototype); + } +} + +async function getOrchestrator() { + if (!orchestratorPromise) { + const Orchestrator = require('../../../src/orchestrator'); + orchestratorPromise = Orchestrator.create({ quiet: true }); + } + return orchestratorPromise; +} + +export function generateClusterId(): string { + return generateName('cluster'); +} + +type LaunchClusterFromTextArgs = { + text: string; + providerOverride?: string | null; + clusterId?: string; + deps?: ClusterLauncherDeps; +}; + +export async function launchClusterFromText({ + text, + providerOverride = null, + clusterId, + deps = {}, +}: LaunchClusterFromTextArgs): Promise<{ clusterId: string }> { + const getOrchestratorImpl = deps.getOrchestrator ?? getOrchestrator; + const loadSettingsImpl = deps.loadSettings ?? loadSettings; + const resolveConfigPathImpl = deps.resolveConfigPath ?? resolveConfigPath; + const loadClusterConfigImpl = deps.loadClusterConfig ?? loadClusterConfig; + const startClusterFromTextImpl = deps.startClusterFromText ?? startClusterFromText; + const generateClusterIdImpl = deps.generateClusterId ?? generateClusterId; + + const orchestrator = await getOrchestratorImpl(); + const settings = loadSettingsImpl(); + const configName = settings.defaultConfig || 'conductor-bootstrap'; + const configPath = resolveConfigPathImpl(configName); + const config = loadClusterConfigImpl(orchestrator, configPath, settings, providerOverride); + const resolvedClusterId = clusterId || generateClusterIdImpl(); + + await startClusterFromTextImpl({ + orchestrator, + text, + config, + settings, + providerOverride, + clusterId: resolvedClusterId, + }); + + return { clusterId: resolvedClusterId }; +} + +type LaunchClusterFromIssueArgs = { + ref: string; + providerOverride?: string | null; + clusterId?: string; + deps?: ClusterLauncherDeps; +}; + +export async function launchClusterFromIssue({ + ref, + providerOverride = null, + clusterId, + deps = {}, +}: LaunchClusterFromIssueArgs): Promise<{ clusterId: string }> { + const getOrchestratorImpl = deps.getOrchestrator ?? getOrchestrator; + const loadSettingsImpl = deps.loadSettings ?? loadSettings; + const resolveConfigPathImpl = deps.resolveConfigPath ?? resolveConfigPath; + const loadClusterConfigImpl = deps.loadClusterConfig ?? loadClusterConfig; + const startClusterFromIssueImpl = deps.startClusterFromIssue ?? startClusterFromIssue; + const detectRunInputImpl = deps.detectRunInput ?? detectRunInput; + const generateClusterIdImpl = deps.generateClusterId ?? generateClusterId; + + const parsed = detectRunInputImpl(ref); + if (!parsed || typeof parsed !== 'object' || !('issue' in parsed)) { + throw new InvalidIssueReferenceError(ref); + } + + const orchestrator = await getOrchestratorImpl(); + const settings = loadSettingsImpl(); + const configName = settings.defaultConfig || 'conductor-bootstrap'; + const configPath = resolveConfigPathImpl(configName); + const config = loadClusterConfigImpl(orchestrator, configPath, settings, providerOverride); + const resolvedClusterId = clusterId || generateClusterIdImpl(); + + await startClusterFromIssueImpl({ + orchestrator, + issue: parsed.issue, + config, + settings, + providerOverride, + clusterId: resolvedClusterId, + }); + + return { clusterId: resolvedClusterId }; +} diff --git a/src/tui-backend/services/cluster-logs.ts b/src/tui-backend/services/cluster-logs.ts new file mode 100644 index 00000000..7143ee70 --- /dev/null +++ b/src/tui-backend/services/cluster-logs.ts @@ -0,0 +1,299 @@ +import fs from 'fs'; +import os from 'os'; +import path from 'path'; + +const Ledger = require('../../../src/ledger'); + +export const MAX_LOG_LINES = 400; +export const LOG_POLL_INTERVAL_MS = 250; + +type ClusterLogState = 'idle' | 'waiting' | 'ready' | 'error'; + +export type ClusterLogStatus = { + state: ClusterLogState; + message?: string; +}; + +export type ClusterLogLine = { + id: string; + timestamp: number; + text: string; + agent: string | null; + role: string | null; + sender: string | null; +}; + +type ClusterLogStreamOptions = { + clusterId?: string | null; + agentId?: string | null; + onLines: (lines: ClusterLogLine[]) => void; + onStatus?: (status: ClusterLogStatus) => void; + pollIntervalMs?: number; + maxInitialLines?: number; +}; + +export function resolveClusterDbPath(clusterId: string): string { + const envHome = + (typeof process.env.HOME === 'string' && process.env.HOME.trim()) || + (typeof process.env.USERPROFILE === 'string' && process.env.USERPROFILE.trim()) || + (typeof process.env.HOMEDRIVE === 'string' && + typeof process.env.HOMEPATH === 'string' && + `${process.env.HOMEDRIVE}${process.env.HOMEPATH}`.trim()) || + ''; + const homeDir = envHome || os.homedir(); + const storageDir = path.join(homeDir, '.zeroshot'); + const clustersFile = path.join(storageDir, 'clusters.json'); + + try { + if (fs.existsSync(clustersFile)) { + const raw = fs.readFileSync(clustersFile, 'utf8'); + try { + const data = JSON.parse(raw); + const entry = data && typeof data === 'object' ? data[clusterId] : null; + const dbPath = entry?.config?.dbPath; + + if (typeof dbPath === 'string' && dbPath.trim()) { + return dbPath; + } + } catch { + // clusters.json can be mid-write; fall back to default path + } + } + } catch { + // Ignore fs errors; fall back to default path + } + + return path.join(storageDir, `${clusterId}.db`); +} + +export function normalizeAgentOutput(message: any): ClusterLogLine | null { + if (!message || typeof message !== 'object') { + return null; + } + + const content = message.content || {}; + const data = content.data || {}; + const contentText = typeof content.text === 'string' ? content.text : ''; + const dataLine = typeof data.line === 'string' ? data.line : ''; + const text = contentText.trim() ? contentText : dataLine; + + if (!text || !text.trim()) { + return null; + } + + const timestamp = typeof message.timestamp === 'number' ? message.timestamp : Date.now(); + const sender = typeof message.sender === 'string' ? message.sender : null; + const agent = typeof data.agent === 'string' ? data.agent : sender; + const role = typeof data.role === 'string' ? data.role : null; + const id = + typeof message.id === 'string' + ? message.id + : `${timestamp}-${Math.random().toString(16).slice(2)}`; + + return { + id, + timestamp, + text, + agent, + role, + sender, + }; +} + +export function createClusterLogStream({ + clusterId, + agentId, + onLines, + onStatus, + pollIntervalMs = LOG_POLL_INTERVAL_MS, + maxInitialLines = MAX_LOG_LINES, +}: ClusterLogStreamOptions) { + let intervalId: NodeJS.Timeout | null = null; + let ledger: any | null = null; + let lastTimestamp = 0; + let initialized = false; + let closed = false; + let lastStatus: ClusterLogStatus | null = null; + const normalizedAgentId = typeof agentId === 'string' && agentId.trim() ? agentId.trim() : null; + + const filterLines = (lines: ClusterLogLine[]) => { + if (!normalizedAgentId) { + return lines; + } + return lines.filter( + (line) => line.agent === normalizedAgentId || line.sender === normalizedAgentId + ); + }; + + const emitStatus = (status: ClusterLogStatus) => { + if (!onStatus) return; + if (lastStatus && lastStatus.state === status.state && lastStatus.message === status.message) { + return; + } + lastStatus = status; + onStatus(status); + }; + + const emitError = (err: unknown, context: string) => { + const message = + err instanceof Error ? err.message : typeof err === 'string' ? err : 'Unknown error'; + emitStatus({ + state: 'error', + message: context ? `${context}: ${message}` : message, + }); + }; + + const resetLedger = (err: unknown, context: string) => { + emitError(err, context); + if (ledger) { + try { + ledger.close(); + } catch { + // ignore close errors + } + ledger = null; + } + }; + + const ensureLedger = () => { + if (!clusterId) { + emitStatus({ state: 'idle' }); + return false; + } + + if (ledger) { + return true; + } + + const dbPath = resolveClusterDbPath(clusterId); + if (!fs.existsSync(dbPath)) { + emitStatus({ state: 'waiting' }); + return false; + } + + try { + ledger = new Ledger(dbPath); + if (!initialized) { + lastTimestamp = 0; + } + emitStatus({ state: 'ready' }); + return true; + } catch (err) { + resetLedger(err, 'Failed to open log database'); + return false; + } + }; + + const loadInitial = () => { + if (!ledger || initialized || !clusterId) { + return; + } + + let rows: any[] = []; + try { + rows = ledger.query({ + cluster_id: clusterId, + topic: 'AGENT_OUTPUT', + order: 'desc', + limit: maxInitialLines, + }); + } catch (err) { + resetLedger(err, 'Failed to read initial logs'); + return; + } + + const messages = rows.slice().reverse(); + const lines = filterLines( + messages + .map((message: any) => normalizeAgentOutput(message)) + .filter(Boolean) as ClusterLogLine[] + ); + + if (lines.length > 0) { + onLines(lines); + } + + if (messages.length > 0) { + const last = messages[messages.length - 1]; + if (last && typeof last.timestamp === 'number') { + lastTimestamp = Math.max(lastTimestamp, last.timestamp); + } + } + + initialized = true; + }; + + const poll = () => { + if (closed) { + return; + } + + if (!clusterId) { + emitStatus({ state: 'idle' }); + return; + } + + if (!ensureLedger()) { + return; + } + + loadInitial(); + + let rows: any[] = []; + try { + rows = ledger.query({ + cluster_id: clusterId, + topic: 'AGENT_OUTPUT', + since: lastTimestamp + 1, + order: 'asc', + }); + } catch (err) { + resetLedger(err, 'Failed to read logs'); + return; + } + + if (!rows.length) { + return; + } + + const lines = filterLines( + rows.map((message: any) => normalizeAgentOutput(message)).filter(Boolean) as ClusterLogLine[] + ); + + if (lines.length > 0) { + onLines(lines); + } + + const last = rows[rows.length - 1]; + if (last && typeof last.timestamp === 'number') { + lastTimestamp = Math.max(lastTimestamp, last.timestamp); + } + }; + + const start = () => { + if (intervalId) { + return; + } + poll(); + intervalId = setInterval(poll, pollIntervalMs); + }; + + const stop = () => { + if (!intervalId) { + return; + } + clearInterval(intervalId); + intervalId = null; + }; + + const close = () => { + closed = true; + stop(); + if (ledger) { + ledger.close(); + ledger = null; + } + }; + + return { start, stop, close }; +} diff --git a/src/tui-backend/services/cluster-registry.ts b/src/tui-backend/services/cluster-registry.ts new file mode 100644 index 00000000..b8df1a6c --- /dev/null +++ b/src/tui-backend/services/cluster-registry.ts @@ -0,0 +1,295 @@ +import { loadSettings } from '../../../lib/settings'; +import { normalizeProviderName } from '../../../lib/provider-names'; + +const pidusage = require('pidusage'); + +type PidusageStats = Record; +type PidusageFn = (pids: number[]) => Promise; + +type ClusterRegistryDeps = { + getOrchestrator?: () => Promise; + pidusage?: PidusageFn; + platform?: string; + loadSettings?: typeof loadSettings; +}; + +export type ClusterSummary = { + id: string; + state: string; + provider: string | null; + createdAt: number; + agentCount: number; + messageCount: number; + cwd: string | null; +}; + +export type ClusterMetrics = { + id: string; + supported: boolean; + cpuPercent: number | null; + memoryMB: number | null; +}; + +let orchestratorPromise: Promise | null = null; + +async function getOrchestrator() { + if (!orchestratorPromise) { + const Orchestrator = require('../../../src/orchestrator'); + orchestratorPromise = Orchestrator.create({ quiet: true }); + } + return orchestratorPromise; +} + +function resolveClusterCwd(cluster: any): string | null { + if (!cluster || typeof cluster !== 'object') { + return null; + } + if (cluster.worktree?.path) { + return cluster.worktree.path; + } + if (cluster.isolation?.workDir) { + return cluster.isolation.workDir; + } + return null; +} + +function resolveClusterProvider(cluster: any, settings: any): string | null { + if (!cluster || typeof cluster !== 'object') { + const fallback = settings?.defaultProvider ?? null; + const normalizedFallback = normalizeProviderName(fallback); + return typeof normalizedFallback === 'string' ? normalizedFallback : null; + } + const forced = cluster.config?.forceProvider ?? null; + const defaultProvider = cluster.config?.defaultProvider ?? null; + const settingsProvider = settings?.defaultProvider ?? null; + const provider = + forced && typeof forced === 'string' + ? forced + : defaultProvider && typeof defaultProvider === 'string' + ? defaultProvider + : settingsProvider && typeof settingsProvider === 'string' + ? settingsProvider + : null; + const normalized = normalizeProviderName(provider); + return typeof normalized === 'string' ? normalized : null; +} + +function resolveAgentPid(agent: any): number | null { + if (!agent || typeof agent !== 'object') { + return null; + } + const pid = agent.processPid ?? agent.pid ?? null; + if (Number.isFinite(pid) && pid > 0) { + return pid; + } + if (typeof agent.getState === 'function') { + const state = agent.getState(); + const statePid = state?.pid ?? null; + if (Number.isFinite(statePid) && statePid > 0) { + return statePid; + } + } + return null; +} + +function collectAgentPids(cluster: any): number[] { + if (!cluster || typeof cluster !== 'object') { + return []; + } + const agents = Array.isArray(cluster.agents) ? cluster.agents : []; + const pids = new Set(); + for (const agent of agents) { + const pid = resolveAgentPid(agent); + if (pid) { + pids.add(pid); + } + } + return Array.from(pids); +} + +function normalizeSummary(summary: any, orchestrator: any, settings: any): ClusterSummary { + if (!summary || typeof summary !== 'object') { + throw new Error('Invalid cluster summary.'); + } + if (typeof summary.id !== 'string' || summary.id.length === 0) { + throw new Error('Invalid cluster id.'); + } + if (!Number.isFinite(summary.createdAt)) { + throw new Error(`Invalid createdAt for cluster ${summary.id}.`); + } + if (!Number.isFinite(summary.agentCount)) { + throw new Error(`Invalid agentCount for cluster ${summary.id}.`); + } + if (!Number.isFinite(summary.messageCount)) { + throw new Error(`Invalid messageCount for cluster ${summary.id}.`); + } + const cluster = orchestrator.getCluster(summary.id); + const cwd = resolveClusterCwd(cluster); + const provider = resolveClusterProvider(cluster, settings); + return { + id: summary.id, + state: String(summary.state ?? 'unknown'), + provider, + createdAt: summary.createdAt, + agentCount: summary.agentCount, + messageCount: summary.messageCount, + cwd, + }; +} + +export class ClusterNotFoundError extends Error { + clusterId: string; + + constructor(clusterId: string) { + super(`Cluster not found: ${clusterId}`); + this.name = 'ClusterNotFoundError'; + this.clusterId = clusterId; + } +} + +type ListClustersArgs = { + deps?: ClusterRegistryDeps; +}; + +type ListClusterMetricsArgs = { + clusterIds?: string[]; + deps?: ClusterRegistryDeps; +}; + +type GetClusterSummaryArgs = { + clusterId: string; + deps?: ClusterRegistryDeps; +}; + +const SUPPORTED_PLATFORMS = new Set(['darwin', 'linux']); +const BYTES_PER_MB = 1024 * 1024; + +export async function listClusters({ deps = {} }: ListClustersArgs = {}): Promise< + ClusterSummary[] +> { + const getOrchestratorImpl = deps.getOrchestrator ?? getOrchestrator; + const loadSettingsImpl = deps.loadSettings ?? loadSettings; + const orchestrator = await getOrchestratorImpl(); + const settings = loadSettingsImpl(); + const summaries = orchestrator.listClusters(); + const results = summaries.map((summary: any) => + normalizeSummary(summary, orchestrator, settings) + ); + results.sort((left, right) => { + if (left.createdAt !== right.createdAt) { + return left.createdAt - right.createdAt; + } + return left.id.localeCompare(right.id); + }); + return results; +} + +export async function getClusterSummary({ + clusterId, + deps = {}, +}: GetClusterSummaryArgs): Promise { + const getOrchestratorImpl = deps.getOrchestrator ?? getOrchestrator; + const loadSettingsImpl = deps.loadSettings ?? loadSettings; + const orchestrator = await getOrchestratorImpl(); + const settings = loadSettingsImpl(); + const summaries = orchestrator.listClusters(); + const summary = summaries.find((entry: any) => entry.id === clusterId); + if (!summary) { + throw new ClusterNotFoundError(clusterId); + } + return normalizeSummary(summary, orchestrator, settings); +} + +export async function listClusterMetrics({ + clusterIds, + deps = {}, +}: ListClusterMetricsArgs = {}): Promise> { + const getOrchestratorImpl = deps.getOrchestrator ?? getOrchestrator; + const pidusageImpl = deps.pidusage ?? pidusage; + const platform = deps.platform ?? process.platform; + const orchestrator = await getOrchestratorImpl(); + const summaries = orchestrator.listClusters(); + const availableIds = summaries.map((summary: any) => summary.id); + const requestedIds = Array.isArray(clusterIds) + ? clusterIds.filter((id) => typeof id === 'string') + : null; + let resolvedIds = availableIds; + if (requestedIds) { + if (requestedIds.length === 0) { + return {}; + } + const availableSet = new Set(availableIds); + resolvedIds = requestedIds.filter((id) => availableSet.has(id)); + } + + if (resolvedIds.length === 0) { + return {}; + } + + if (!SUPPORTED_PLATFORMS.has(platform)) { + return Object.fromEntries( + resolvedIds.map((id) => [ + id, + { + id, + supported: false, + cpuPercent: null, + memoryMB: null, + }, + ]) + ); + } + + const pidsByCluster = new Map(); + const allPids = new Set(); + for (const clusterId of resolvedIds) { + const cluster = orchestrator.getCluster(clusterId); + const pids = collectAgentPids(cluster); + pidsByCluster.set(clusterId, pids); + for (const pid of pids) { + allPids.add(pid); + } + } + + let statsByPid: PidusageStats = {}; + if (allPids.size > 0) { + try { + statsByPid = await pidusageImpl(Array.from(allPids)); + } catch { + statsByPid = {}; + } + } + + const results: Record = {}; + for (const clusterId of resolvedIds) { + const pids = pidsByCluster.get(clusterId) ?? []; + let cpuTotal = 0; + let memoryTotalBytes = 0; + let hasCpu = false; + let hasMemory = false; + for (const pid of pids) { + const stats = statsByPid[String(pid)] ?? statsByPid[pid as any]; + if (!stats) { + continue; + } + const cpu = Number(stats.cpu); + const memory = Number(stats.memory); + if (Number.isFinite(cpu)) { + cpuTotal += cpu; + hasCpu = true; + } + if (Number.isFinite(memory)) { + memoryTotalBytes += memory; + hasMemory = true; + } + } + results[clusterId] = { + id: clusterId, + supported: true, + cpuPercent: hasCpu ? cpuTotal : null, + memoryMB: hasMemory ? memoryTotalBytes / BYTES_PER_MB : null, + }; + } + + return results; +} diff --git a/src/tui-backend/services/cluster-timeline.ts b/src/tui-backend/services/cluster-timeline.ts new file mode 100644 index 00000000..6b3a8659 --- /dev/null +++ b/src/tui-backend/services/cluster-timeline.ts @@ -0,0 +1,298 @@ +import fs from 'fs'; + +const Ledger = require('../../../src/ledger'); + +import { resolveClusterDbPath } from './cluster-logs'; + +export const MAX_TIMELINE_EVENTS = 40; +export const TIMELINE_POLL_INTERVAL_MS = 750; + +export const WORKFLOW_TRIGGERS = Object.freeze([ + 'ISSUE_OPENED', + 'PLAN_READY', + 'IMPLEMENTATION_READY', + 'VALIDATION_RESULT', + 'CONDUCTOR_ESCALATE', +]); + +type ClusterTimelineState = 'idle' | 'waiting' | 'ready' | 'error'; + +export type ClusterTimelineStatus = { + state: ClusterTimelineState; + message?: string; +}; + +export type TimelineEvent = { + id: string; + timestamp: number; + topic: string; + label: string; + approved: boolean | null; + sender: string | null; +}; + +type ClusterTimelineStreamOptions = { + clusterId?: string | null; + onEvents: (events: TimelineEvent[]) => void; + onStatus?: (status: ClusterTimelineStatus) => void; + pollIntervalMs?: number; + maxInitialEvents?: number; +}; + +function isWorkflowTopic(topic: string): boolean { + return WORKFLOW_TRIGGERS.includes(topic); +} + +function normalizeApproved(value: unknown): boolean | null { + if (value === true || value === 'true') { + return true; + } + if (value === false || value === 'false') { + return false; + } + return null; +} + +function labelForMessage(message: any, approved: boolean | null): string { + switch (message.topic) { + case 'ISSUE_OPENED': + return 'Issue opened'; + case 'PLAN_READY': + return 'Plan ready'; + case 'IMPLEMENTATION_READY': + return 'Implementation ready'; + case 'VALIDATION_RESULT': + if (approved === true) { + return 'Validation approved'; + } + if (approved === false) { + return 'Validation rejected'; + } + return 'Validation result'; + case 'CONDUCTOR_ESCALATE': + return 'Conductor escalated'; + default: + return message.topic || 'Workflow event'; + } +} + +export function normalizeTimelineMessage(message: any): TimelineEvent | null { + if (!message || typeof message !== 'object') { + return null; + } + + const topic = typeof message.topic === 'string' ? message.topic : ''; + if (!topic || !isWorkflowTopic(topic)) { + return null; + } + + const data = message.content?.data || {}; + const approved = normalizeApproved(data.approved); + const timestamp = typeof message.timestamp === 'number' ? message.timestamp : Date.now(); + const id = + typeof message.id === 'string' + ? message.id + : `${timestamp}-${Math.random().toString(16).slice(2)}`; + + return { + id, + timestamp, + topic, + label: labelForMessage(message, approved), + approved, + sender: typeof message.sender === 'string' ? message.sender : null, + }; +} + +export function createClusterTimelineStream({ + clusterId, + onEvents, + onStatus, + pollIntervalMs = TIMELINE_POLL_INTERVAL_MS, + maxInitialEvents = MAX_TIMELINE_EVENTS, +}: ClusterTimelineStreamOptions) { + let intervalId: NodeJS.Timeout | null = null; + let ledger: any | null = null; + let lastTimestamp = 0; + let initialized = false; + let closed = false; + let lastStatus: ClusterTimelineStatus | null = null; + + const emitStatus = (status: ClusterTimelineStatus) => { + if (!onStatus) return; + if (lastStatus && lastStatus.state === status.state && lastStatus.message === status.message) { + return; + } + lastStatus = status; + onStatus(status); + }; + + const emitError = (err: unknown, context: string) => { + const message = + err instanceof Error ? err.message : typeof err === 'string' ? err : 'Unknown error'; + emitStatus({ + state: 'error', + message: context ? `${context}: ${message}` : message, + }); + }; + + const resetLedger = (err: unknown, context: string) => { + emitError(err, context); + if (ledger) { + try { + ledger.close(); + } catch { + // ignore close errors + } + ledger = null; + } + }; + + const ensureLedger = () => { + if (!clusterId) { + emitStatus({ state: 'idle' }); + return false; + } + + if (ledger) { + return true; + } + + const dbPath = resolveClusterDbPath(clusterId); + if (!fs.existsSync(dbPath)) { + emitStatus({ state: 'waiting' }); + return false; + } + + try { + ledger = new Ledger(dbPath); + if (!initialized) { + lastTimestamp = 0; + } + emitStatus({ state: 'ready' }); + return true; + } catch (err) { + resetLedger(err, 'Failed to open timeline database'); + return false; + } + }; + + const queryWorkflowMessages = (since?: number): any[] => { + if (!ledger || !clusterId) { + return []; + } + const messages: any[] = []; + const hasSince = typeof since === 'number' && Number.isFinite(since); + for (const topic of WORKFLOW_TRIGGERS) { + const criteria: any = { + cluster_id: clusterId, + topic, + order: 'asc', + }; + if (hasSince && since! > 0) { + criteria.since = since; + } + try { + const rows = ledger.query(criteria); + if (rows.length) { + messages.push(...rows); + } + } catch (err) { + resetLedger(err, 'Failed to read timeline entries'); + return []; + } + } + messages.sort((a, b) => (a.timestamp || 0) - (b.timestamp || 0)); + return messages; + }; + + const loadInitial = () => { + if (!ledger || initialized || !clusterId) { + return; + } + + const messages = queryWorkflowMessages(); + if (messages.length) { + const events = messages + .map((message) => normalizeTimelineMessage(message)) + .filter(Boolean) as TimelineEvent[]; + if (events.length) { + const trimmed = + events.length > maxInitialEvents + ? events.slice(events.length - maxInitialEvents) + : events; + onEvents(trimmed); + } + + const last = messages[messages.length - 1]; + if (last && typeof last.timestamp === 'number') { + lastTimestamp = Math.max(lastTimestamp, last.timestamp); + } + } + + initialized = true; + }; + + const poll = () => { + if (closed) { + return; + } + + if (!clusterId) { + emitStatus({ state: 'idle' }); + return; + } + + if (!ensureLedger()) { + return; + } + + loadInitial(); + + const since = lastTimestamp > 0 ? lastTimestamp + 1 : 1; + const messages = queryWorkflowMessages(since); + if (!messages.length) { + return; + } + + const events = messages + .map((message) => normalizeTimelineMessage(message)) + .filter(Boolean) as TimelineEvent[]; + + if (events.length) { + onEvents(events); + } + + const last = messages[messages.length - 1]; + if (last && typeof last.timestamp === 'number') { + lastTimestamp = Math.max(lastTimestamp, last.timestamp); + } + }; + + const start = () => { + if (intervalId) { + return; + } + poll(); + intervalId = setInterval(poll, pollIntervalMs); + }; + + const stop = () => { + if (!intervalId) { + return; + } + clearInterval(intervalId); + intervalId = null; + }; + + const close = () => { + closed = true; + stop(); + if (ledger) { + ledger.close(); + ledger = null; + } + }; + + return { start, stop, close }; +} diff --git a/src/tui-backend/services/cluster-topology.ts b/src/tui-backend/services/cluster-topology.ts new file mode 100644 index 00000000..46035f0d --- /dev/null +++ b/src/tui-backend/services/cluster-topology.ts @@ -0,0 +1,145 @@ +type ClusterTopologyDeps = { + getOrchestrator?: () => Promise; +}; + +export type TopologyAgent = { + id: string; + role: string | null; +}; + +export type TopologyEdge = { + from: string; + to: string; + topic: string; + kind: 'trigger' | 'publish' | 'source'; + dynamic?: boolean; +}; + +export type ClusterTopology = { + agents: TopologyAgent[]; + edges: TopologyEdge[]; + topics: string[]; +}; + +let orchestratorPromise: Promise | null = null; + +async function getOrchestrator() { + if (!orchestratorPromise) { + const Orchestrator = require('../../../src/orchestrator'); + orchestratorPromise = Orchestrator.create({ quiet: true }); + } + return orchestratorPromise; +} + +function normalizeTopic(value: any): string | null { + if (typeof value !== 'string') { + return null; + } + const topic = value.trim(); + return topic ? topic : null; +} + +function extractTopicsFromScript(script: any): string[] { + if (typeof script !== 'string') { + return []; + } + const topics = new Set(); + const regex = /topic\s*:\s*['"`]([A-Za-z0-9_:-]+)['"`]/g; + let match: RegExpExecArray | null = null; + while ((match = regex.exec(script)) !== null) { + if (match[1]) { + topics.add(match[1]); + } + } + return Array.from(topics); +} + +export function buildTopologyModel(config: any): ClusterTopology { + const agents: TopologyAgent[] = []; + const edges: TopologyEdge[] = []; + const topics = new Set(); + const edgeKeys = new Set(); + + const addEdge = ( + from: string, + to: string, + topic: string, + kind: TopologyEdge['kind'], + dynamic?: boolean + ) => { + if (!from || !to || !topic) { + return; + } + const key = `${from}::${to}`; + if (edgeKeys.has(key)) { + return; + } + edgeKeys.add(key); + edges.push({ from, to, topic, kind, dynamic }); + }; + + topics.add('ISSUE_OPENED'); + addEdge('system', 'ISSUE_OPENED', 'ISSUE_OPENED', 'source'); + + const agentConfigs = Array.isArray(config?.agents) ? config.agents : []; + for (const agent of agentConfigs) { + const id = typeof agent?.id === 'string' ? agent.id : null; + if (!id) { + continue; + } + agents.push({ + id, + role: typeof agent.role === 'string' ? agent.role : null, + }); + + const triggers = Array.isArray(agent.triggers) ? agent.triggers : []; + for (const trigger of triggers) { + const topic = normalizeTopic(trigger?.topic); + if (!topic) { + continue; + } + topics.add(topic); + addEdge(topic, id, topic, 'trigger'); + } + + const outputTopic = normalizeTopic(agent?.hooks?.onComplete?.config?.topic); + if (outputTopic) { + topics.add(outputTopic); + addEdge(id, outputTopic, outputTopic, 'publish'); + } + + const hookLogicScript = agent?.hooks?.onComplete?.logic?.script; + for (const topic of extractTopicsFromScript(hookLogicScript)) { + topics.add(topic); + addEdge(id, topic, topic, 'publish', true); + } + + const hookTransformScript = agent?.hooks?.onComplete?.transform?.script; + for (const topic of extractTopicsFromScript(hookTransformScript)) { + topics.add(topic); + addEdge(id, topic, topic, 'publish', true); + } + } + + return { + agents, + edges, + topics: Array.from(topics), + }; +} + +export async function getClusterTopology( + clusterId: string | null | undefined, + { deps = {} }: { deps?: ClusterTopologyDeps } = {} +): Promise { + if (!clusterId) { + return { agents: [], edges: [], topics: [] }; + } + const getOrchestratorImpl = deps.getOrchestrator ?? getOrchestrator; + const orchestrator = await getOrchestratorImpl(); + const cluster = orchestrator.getCluster(clusterId); + if (!cluster?.config) { + throw new Error(`Cluster ${clusterId} not found.`); + } + return buildTopologyModel(cluster.config); +} diff --git a/src/tui-backend/services/guidance-delivery.ts b/src/tui-backend/services/guidance-delivery.ts new file mode 100644 index 00000000..17602594 --- /dev/null +++ b/src/tui-backend/services/guidance-delivery.ts @@ -0,0 +1,74 @@ +type GuidanceDeliveryResult = { + status: string; + reason: string | null; + method: string | null; + taskId?: string | null; +}; + +type ClusterGuidanceSummary = { + injected: number; + queued: number; + total: number; +}; + +type ClusterGuidanceDelivery = { + summary: ClusterGuidanceSummary; + agents: Record; + timestamp: number; +}; + +type GuidanceDeliveryDeps = { + getOrchestrator?: () => Promise; +}; + +let orchestratorPromise: Promise | null = null; + +async function getOrchestrator() { + if (!orchestratorPromise) { + const Orchestrator = require('../../../src/orchestrator'); + orchestratorPromise = Orchestrator.create({ quiet: true }); + } + return orchestratorPromise; +} + +type SendAgentGuidanceArgs = { + clusterId: string; + agentId: string; + text: string; + timeoutMs?: number; + deps?: GuidanceDeliveryDeps; +}; + +export async function sendAgentGuidance({ + clusterId, + agentId, + text, + timeoutMs, + deps = {}, +}: SendAgentGuidanceArgs): Promise { + const getOrchestratorImpl = deps.getOrchestrator ?? getOrchestrator; + const orchestrator = await getOrchestratorImpl(); + return await orchestrator.sendGuidanceToAgent(clusterId, agentId, text, { + timeoutMs, + }); +} + +type SendClusterGuidanceArgs = { + clusterId: string; + text: string; + timeoutMs?: number; + deps?: GuidanceDeliveryDeps; +}; + +export async function sendClusterGuidance({ + clusterId, + text, + timeoutMs, + deps = {}, +}: SendClusterGuidanceArgs): Promise { + const getOrchestratorImpl = deps.getOrchestrator ?? getOrchestrator; + const orchestrator = await getOrchestratorImpl(); + return await orchestrator.sendGuidanceToCluster(clusterId, text, { + timeoutMs, + }); +} diff --git a/src/tui-backend/services/index.ts b/src/tui-backend/services/index.ts new file mode 100644 index 00000000..e537d853 --- /dev/null +++ b/src/tui-backend/services/index.ts @@ -0,0 +1,7 @@ +export * from './cluster-launcher'; +export * from './cluster-registry'; +export * from './cluster-logs'; +export * from './cluster-timeline'; +export * from './cluster-topology'; +export * from './guidance-delivery'; +export * from './start-cluster'; diff --git a/src/tui-backend/services/start-cluster.ts b/src/tui-backend/services/start-cluster.ts new file mode 100644 index 00000000..c4ec87f6 --- /dev/null +++ b/src/tui-backend/services/start-cluster.ts @@ -0,0 +1 @@ +export { startClusterFromText, startClusterFromIssue } from '../../../lib/start-cluster'; diff --git a/src/tui-backend/subscriptions/index.ts b/src/tui-backend/subscriptions/index.ts new file mode 100644 index 00000000..79a59ad4 --- /dev/null +++ b/src/tui-backend/subscriptions/index.ts @@ -0,0 +1,69 @@ +import { randomUUID } from 'crypto'; + +export type SubscriptionKind = string; + +export type SubscriptionEntry = { + id: string; + kind: SubscriptionKind; + close: () => void; + closed: boolean; +}; + +export type SubscriptionRegistry = { + add: (kind: SubscriptionKind, close: () => void) => string; + unsubscribe: (id: string) => { removed: boolean }; + closeAll: () => number; + size: () => number; +}; + +export function createSubscriptionRegistry(): SubscriptionRegistry { + const entries = new Map(); + + const add = (kind: SubscriptionKind, close: () => void) => { + if (typeof close !== 'function') { + throw new TypeError('Subscription close must be a function.'); + } + const id = randomUUID(); + entries.set(id, { + id, + kind, + close, + closed: false, + }); + return id; + }; + + const unsubscribe = (id: string) => { + const entry = entries.get(id); + if (!entry) { + return { removed: false }; + } + entries.delete(id); + if (!entry.closed) { + entry.closed = true; + entry.close(); + } + return { removed: true }; + }; + + const closeAll = () => { + const values = Array.from(entries.values()); + entries.clear(); + for (const entry of values) { + if (!entry.closed) { + entry.closed = true; + entry.close(); + } + } + return values.length; + }; + + const size = () => entries.size; + + return { + add, + unsubscribe, + closeAll, + size, + }; +} diff --git a/src/tui/CHANGES.txt b/src/tui/CHANGES.txt deleted file mode 100644 index 1e8e4ac1..00000000 --- a/src/tui/CHANGES.txt +++ /dev/null @@ -1,133 +0,0 @@ -TUI Performance & UX Improvements -================================== - -## New Features - -### 4. Two-Level Navigation šŸŽÆ -**Feature:** Completely separate layouts for overview vs detail -**Implementation:** -- Overview mode: ONLY shows clusters table + stats (agents/logs hidden) -- Detail mode: ONLY shows agents + logs (clusters/stats hidden) -- Enter key to drill into detail, Escape to return -- Help text updates dynamically based on current view -- Widgets physically shown/hidden (not just empty data) - -## Fixed Issues - -### 1. Slow Startup āš” -**Problem:** TUI took 5-10 seconds to start due to synchronous cluster loading -**Solution:** -- Deferred initial polls by 50-100ms to let UI render first -- Shows "Loading..." message immediately -- Lazy-loads cluster ledgers only when needed -- Startup now instant (<100ms) - -### 2. Default Filter šŸŽÆ -**Problem:** Showed all clusters (including stopped) by default -**Solution:** -- Changed default filter from "all" to "running" -- User only sees active clusters -- Can still use `--filter all` to see everything - -### 3. Cluster Selection šŸ“ -**Problem:** Agents and logs weren't properly filtered by selected cluster -**Solution:** -- Renderer now tracks selectedClusterId -- Agents shown are for the selected cluster only -- Logs filtered to show only messages from selected cluster -- Messages cleared when switching between clusters -- Navigate with ā†‘ā†“ or jk keys - -## Performance Improvements - -**Before:** -- Startup: 5-10 seconds -- All clusters loaded synchronously -- All ledgers opened on startup -- Unfiltered logs from all clusters - -**After:** -- Startup: <100ms instant -- Clusters loaded async after UI renders -- Ledgers lazy-loaded when needed -- Logs filtered to selected cluster only - -## Usage - -```bash -# Shows only running clusters (default) -vibe watch - -# Show all clusters (including stopped) -vibe watch --filter all - -# Show only stopped clusters -vibe watch --filter stopped -``` - -## Keyboard Navigation - -### Two-Level Navigation -- **Overview Mode** (default): ONLY clusters + stats visible - - Large clusters table (16 rows) with system stats sidebar - - No agents or logs shown - - `ā†‘` / `k` - Select previous cluster - - `ā†“` / `j` - Select next cluster - - `Enter` - Switch to detail view for selected cluster - -- **Detail Mode**: ONLY agents + logs visible - - Full-width agents table (9 rows) - - Full-width live logs (9 rows) - - Clusters table and stats hidden - - `Escape` - Switch back to overview mode - -Agents and logs auto-update in real-time when in detail view - -## Technical Changes - -### data-poller.js -- Line 45-53: Deferred initial polls with setTimeout -- Line 196-208: Added lazy loading for cluster ledgers -- Line 201-203: Check if ledger DB exists before loading - -### index.js (TUI) -- Line 21: Changed default filter to 'running' -- Line 34-36: Added viewMode state ('overview' or 'detail') and detailClusterId -- Line 48-49: Show "Loading..." message on startup -- Line 107-119: Conditional rendering - agents only shown in detail view -- Line 130-137: Conditional rendering for resource_stats case - -### keybindings.js -- Line 14-37: Enter key handler to switch to detail view -- Line 39-59: Escape key handler to switch to overview view -- Line 22, 45: Clear messages when switching views -- Line 29-34: Detail mode - hide clusters/stats, show agents/logs -- Line 52-57: Overview mode - show clusters/stats, hide agents/logs -- Line 24-27, 47-50: Update help text based on view mode - -### layout.js -- Line 36: Expanded clusters table to 16 rows (from 6) -- Line 66: Expanded stats box to 16 rows (from 6) -- Line 85: Repositioned agent table to row 0, 9 rows, full width -- Line 119: Repositioned logs to row 9, 9 rows, full width -- Line 165-167: Initially hide agent table and logs (overview mode default) - -### cli/index.js -- Line 1161: Changed default filter to 'running' in CLI option - -## Testing - -Run integration tests: -```bash -node tests/tui-integration.test.js # Basic TUI startup and data loading -node tests/tui-navigation-test.js # Two-level navigation functionality -``` - -Expected: All tests pass, TUI starts instantly - -## Notes - -- Messages are cluster-scoped (only show for selected cluster) -- Selection persists across refreshes -- Empty clusters (no agents) still show in list -- Logs clear when switching clusters to avoid confusion diff --git a/src/tui/LAYOUT.md b/src/tui/LAYOUT.md deleted file mode 100644 index 53cb23b3..00000000 --- a/src/tui/LAYOUT.md +++ /dev/null @@ -1,261 +0,0 @@ -# TUI Dashboard Layout Module - -Dashboard layout builder for real-time cluster monitoring with blessed-contrib. - -## Overview - -The layout module creates a responsive terminal UI with a 20x12 grid layout containing: - -- **Clusters Table** (top-left): View all active clusters, status, agent count, and uptime -- **System Stats** (top-right): CPU, memory, and cluster statistics -- **Agents Table** (middle): List all agents with role, status, iteration, and resource usage -- **Live Logs** (lower): Real-time event stream with color-coded severity levels -- **Help Bar** (bottom): Keyboard shortcut reference - -## Grid Layout - -``` -ā”Œā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”¬ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā” -ā”‚ Clusters Table (6 rows x 8 cols) ā”‚ System Stats Box ā”‚ -ā”‚ ā”‚ (6 rows x 4 cols) ā”‚ -ā”œā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”“ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”¤ -ā”‚ Agents Table (6 rows x 12 cols) ā”‚ -ā”œā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”¤ -ā”‚ Live Logs (6 rows x 12 cols) ā”‚ -ā”œā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”¤ -ā”‚ Help Bar (2 rows x 12 cols) ā”‚ -ā””ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”˜ -``` - -## Usage - -### Basic Setup - -```javascript -const blessed = require('blessed'); -const { createLayout } = require('./layout'); - -// Create screen -const screen = blessed.screen({ mouse: true, title: 'Cluster Dashboard' }); - -// Create layout -const layout = createLayout(screen); - -// Render -screen.render(); - -// Exit handler -screen.key(['q', 'C-c'], () => process.exit(0)); -``` - -### Updating Tables - -```javascript -const { updateClustersTable, updateAgentsTable, updateStatsBox, addLogEntry } = require('./layout'); - -// Update clusters table -updateClustersTable(layout.clustersTable, [ - { - id: 'cluster-swift-falcon', - status: 'running', - agentCount: 5, - config: 'default', - uptime: '2h 30m', - }, -]); - -// Update agents table -updateAgentsTable(layout.agentTable, [ - { - clusterId: 'cluster-swift-falcon', - id: 'worker-1', - role: 'worker', - status: 'running', - iteration: 3, - cpu: '12.5%', - memory: '245 MB', - }, -]); - -// Update system stats -updateStatsBox(layout.statsBox, { - activeClusters: 2, - totalAgents: 5, - usedMemory: '512 MB', - totalMemory: '8 GB', - totalCPU: '26.2%', -}); - -// Add log entry -addLogEntry(layout.logsBox, 'Cluster started successfully', 'info'); -addLogEntry(layout.logsBox, 'Warning: High CPU usage', 'warn'); -addLogEntry(layout.logsBox, 'Error: Agent crashed', 'error'); -``` - -## API Reference - -### createLayout(screen) - -Creates the dashboard layout with all widgets. - -**Parameters:** - -- `screen` (blessed.screen): Blessed screen instance - -**Returns:** - -```javascript -{ - (screen, // Blessed screen - grid, // blessed-contrib grid - clustersTable, // Clusters table widget - agentTable, // Agents table widget - statsBox, // System stats box widget - logsBox, // Live logs widget - helpBar, // Help bar widget - widgets, // Array of interactive widgets [clustersTable, agentTable, logsBox] - focus(index), // Function to focus widget by index - getCurrentFocus()); // Function to get current focus index -} -``` - -### updateClustersTable(clustersTable, clusters) - -Updates the clusters table with current data. - -**Parameters:** - -- `clustersTable`: Clusters table widget -- `clusters` (array): Array of cluster objects with properties: - - `id` (string): Cluster identifier - - `status` (string): running | stopped | initializing | stopping | failed | killed - - `agentCount` (number): Number of agents - - `config` (string): Configuration name - - `uptime` (string): Formatted uptime (e.g., "2h 30m") - -### updateAgentsTable(agentTable, agents) - -Updates the agents table with current data. - -**Parameters:** - -- `agentTable`: Agents table widget -- `agents` (array): Array of agent objects with properties: - - `clusterId` (string): Parent cluster ID - - `id` (string): Agent identifier - - `role` (string): worker | validator | orchestrator - - `status` (string): running | idle | failed - - `iteration` (number): Current iteration count - - `cpu` (string): CPU percentage (e.g., "12.5%") - - `memory` (string): Memory usage (e.g., "245 MB") - -### updateStatsBox(statsBox, stats) - -Updates the system stats box. - -**Parameters:** - -- `statsBox`: Stats box widget -- `stats` (object): - - `activeClusters` (number): Count of active clusters - - `totalAgents` (number): Total agent count - - `usedMemory` (string): Formatted memory usage - - `totalMemory` (string): Formatted total memory - - `totalCPU` (string): Total CPU percentage - -### addLogEntry(logsBox, message, level) - -Adds a timestamped log entry. - -**Parameters:** - -- `logsBox`: Logs box widget -- `message` (string): Log message -- `level` (string): info | warn | error | debug (default: info) - -### clearLogs(logsBox) - -Clears all log entries. - -**Parameters:** - -- `logsBox`: Logs box widget - -## Keyboard Navigation - -| Key | Action | -| --------- | -------------------------- | -| Tab | Next widget | -| Shift+Tab | Previous widget | -| ā†‘/ā†“ | Navigate in focused widget | -| Enter | Select/activate | -| q | Quit | - -## Color Scheme - -- **Borders**: Cyan -- **Headers**: Cyan (bold) -- **Text**: White -- **Selection**: Black text on cyan background -- **Log levels**: - - info: White - - warn: Yellow - - error: Red - - debug: Gray - -## Demo - -Run the included demo: - -```bash -node src/tui/demo.js -``` - -Keyboard shortcuts in demo: - -- [r] - Refresh data -- [c] - Simulate cluster start -- [k] - Simulate cluster kill -- [s] - Simulate warning -- [q] - Quit - -## Testing - -Run tests: - -```bash -npm test -- tests/tui-layout.test.js -``` - -Tests verify: - -- Layout creation and widget initialization -- Data update functions -- Focus navigation -- Log entry handling -- Edge cases (empty data, missing properties) - -## Styling Customization - -Widgets can be customized by modifying the configuration objects in `createLayout()`: - -```javascript -const clustersTable = grid.set(0, 0, 6, 8, contrib.table, { - fg: 'white', // Foreground color - selectedFg: 'black', // Selected foreground - selectedBg: 'cyan', // Selected background - border: { type: 'line', fg: 'cyan' }, - style: { - header: { fg: 'cyan', bold: true }, - cell: { selected: { fg: 'black', bg: 'cyan' } }, - }, -}); -``` - -## Related Modules - -- `formatters.js` - Value formatting utilities (timestamps, bytes, CPU) -- `renderer.js` - Additional rendering helpers -- `keybindings.js` - Keyboard event handlers -- `data-poller.js` - Real-time data collection -- `index.js` - Main dashboard integration diff --git a/src/tui/README.txt b/src/tui/README.txt deleted file mode 100644 index d246f5a1..00000000 --- a/src/tui/README.txt +++ /dev/null @@ -1,192 +0,0 @@ -VIBE WATCH - Interactive TUI Dashboard -======================================= - -Launch with: vibe watch - -OVERVIEW --------- -The vibe watch command provides a real-time, htop/k9s-style dashboard for monitoring all active vibe clusters. - -FEATURES --------- -āœ“ Real-time cluster state monitoring (1s refresh) -āœ“ CPU and memory tracking per agent (via pidusage) -āœ“ Live message streaming from cluster ledgers -āœ“ Interactive keyboard controls (kill, stop, export) -āœ“ Automatic detection of new clusters -āœ“ System-wide statistics (active clusters, agents, avg resources) - -LAYOUT ------- -ā”Œā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā” -ā”‚ VIBE CLUSTER WATCH [q] Quit ā”‚ -ā”œā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”¤ -ā”‚ ā”Œā”€ Clusters ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā” ā”Œā”€ System Stats ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā” ā”‚ -ā”‚ ā”‚ ID State Time ā”‚ ā”‚ Active: 2 CPU: 12% ā”‚ ā”‚ -ā”‚ ā”‚ ā— a-38 RUN 5m ā”‚ ā”‚ Agents: 7 Mem: 245 MB ā”‚ ā”‚ -ā”‚ ā”‚ ā— s-62 RUN 2m ā”‚ ā””ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”˜ ā”‚ -ā”‚ ā””ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”˜ ā”‚ -ā”‚ ā”Œā”€ Agents ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā” ā”‚ -ā”‚ ā”‚ Agent Role State Iter CPU% Mem(MB) ā”‚ ā”‚ -ā”‚ ā”‚ worker impl exec 3 8.5 67 ā”‚ ā”‚ -ā”‚ ā”‚ validator val idle 1 0.1 42 ā”‚ ā”‚ -ā”‚ ā””ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”˜ ā”‚ -ā”‚ ā”Œā”€ Live Logs ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā” ā”‚ -ā”‚ ā”‚ [09:45:23] worker: TASK_STARTED (iteration 3) ā”‚ ā”‚ -ā”‚ ā”‚ [09:45:24] worker: Implementing feature X... ā”‚ ā”‚ -ā”‚ ā””ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”˜ ā”‚ -ā”‚ ā”Œā”€ Help ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā” ā”‚ -ā”‚ ā”‚ [ā†‘/ā†“] Nav [K] Kill [s] Stop [e] Export [q] Quitā”‚ ā”‚ -ā”‚ ā””ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”˜ ā”‚ -ā””ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”˜ - -KEYBOARD SHORTCUTS ------------------- -Navigation: - ā†‘ / k Move selection up - ā†“ / j Move selection down - -Actions (on selected cluster): - K Kill cluster (force, with confirmation) - s Stop cluster (graceful, with confirmation) - e Export cluster conversation (markdown) - l Open full logs in new terminal window - -View Controls: - r Force refresh all data - f Toggle filter (running/stopped/all) - ? / h Show help dialog - -Exit: - q / Ctrl-C Quit (with confirmation if clusters active) - -USAGE ------ -# Launch with defaults -vibe watch - -# Filter to only running clusters -vibe watch --filter running - -# Faster refresh (500ms instead of 1s) -vibe watch --refresh-rate 500 - -# Show only stopped clusters -vibe watch --filter stopped - -ARCHITECTURE ------------- -The TUI is composed of 6 modular components: - -1. index.js - Main coordinator, initializes screen/layout/poller -2. layout.js - Creates blessed-contrib grid with 5 widgets -3. renderer.js - Transforms data into widget updates -4. data-poller.js - Polls orchestrator at 4 different intervals -5. keybindings.js - Keyboard event handlers and confirmations -6. formatters.js - Utility functions (time, bytes, CPU, icons) - -Data Flow: - Orchestrator ā†’ DataPoller ā†’ TUI.onUpdate ā†’ Renderer ā†’ Widgets ā†’ Screen - -Polling Strategy: - - Cluster states: 1s (main refresh rate) - - Resource stats: 2s (expensive pidusage calls) - - New clusters: 2s (rare event) - - Log messages: 500ms per cluster (real-time feel) - -DEPENDENCIES ------------- -- blessed@0.1.81 Terminal UI framework -- blessed-contrib@4.11.0 Dashboard widgets (grid, table, log) -- pidusage@4.0.1 Cross-platform CPU/memory monitoring - -TESTING -------- -# Run integration test -node tests/tui-integration.test.js - -# Run layout demo (interactive) -node src/tui/demo.js - -# Run unit tests -npm test - -TROUBLESHOOTING ---------------- -Q: TUI shows "No clusters found" -A: Start a cluster first: vibe run "test task" or vibe task run "test" - -Q: CPU/Memory shows 0% -A: Process may have died. Check cluster state. - -Q: Logs not streaming -A: Ledger database may be locked. Wait a few seconds. - -Q: Terminal garbled on exit -A: Try running: reset - -Q: Keyboard shortcuts not working -A: Make sure terminal supports key events (most modern terminals do) - -DEMO MODE ---------- -To see the TUI with mock data: - node src/tui/demo.js - -This starts a live dashboard with simulated clusters that auto-updates. -Press [r] to refresh, [c] to add cluster, [k] to kill, [q] to quit. - -FILES ------ -src/tui/ -ā”œā”€ā”€ index.js Main TUI class (6.0K) -ā”œā”€ā”€ layout.js Widget creation (8.1K) -ā”œā”€ā”€ renderer.js Data ā†’ widgets (5.6K) -ā”œā”€ā”€ data-poller.js Data collection (8.1K) -ā”œā”€ā”€ keybindings.js User input (8.9K) -ā”œā”€ā”€ formatters.js Utilities (3.4K) -ā”œā”€ā”€ demo.js Interactive demo (5.0K) -ā””ā”€ā”€ LAYOUT.md API documentation (7.4K) - -tests/ -ā””ā”€ā”€ tui-integration.test.js Integration test - -MODIFICATIONS TO EXISTING FILES -------------------------------- -src/agent-wrapper.js - - Added: this.processPid tracking (line 42) - - Added: PID capture on spawn (line 605-607) - - Added: PROCESS_SPAWNED lifecycle event - - Added: TASK_ID_ASSIGNED lifecycle event - - Added: pid field in getState() (line 1350) - -cli/index.js - - Added: vibe watch command with options - -lib/completion.js - - Added: Shell completion for watch command - -package.json - - Added: blessed, blessed-contrib, pidusage dependencies - -FUTURE ENHANCEMENTS -------------------- -Potential improvements: -- [ ] Sorting clusters by various fields -- [ ] Filtering by cluster config name -- [ ] Graph view for CPU/memory over time -- [ ] Search/filter logs by keyword -- [ ] Export selected logs to file -- [ ] Cluster health indicators -- [ ] Alert notifications for failures -- [ ] Docker container stats (for isolation mode) -- [ ] Network I/O stats -- [ ] Agent communication graph visualization - -CREDITS -------- -Implementation: 6 parallel agents (Dec 2024) -Architecture: blessed-contrib grid system -Inspiration: htop, k9s, lazydocker - -For issues or feature requests, see vibe/cluster GitHub repo. diff --git a/src/tui/TWO-LEVEL-NAVIGATION.md b/src/tui/TWO-LEVEL-NAVIGATION.md deleted file mode 100644 index eeaaf562..00000000 --- a/src/tui/TWO-LEVEL-NAVIGATION.md +++ /dev/null @@ -1,186 +0,0 @@ -# Two-Level Navigation - Implementation Summary - -## Overview - -Completely redesigned TUI layout with separate views for overview and detail modes: - -1. **Overview mode** (default): ONLY clusters + stats - clean, focused view -2. Press Enter ā†’ **Detail mode**: ONLY agents + logs for selected cluster -3. Press Escape ā†’ Return to overview - -## User Experience - -### Overview Mode (Default) - -- **ONLY visible:** Large clusters table (16 rows) + system stats sidebar -- **Hidden:** Agent table and logs (completely invisible) -- Clean, spacious layout focusing on cluster selection -- Help text: `[Enter] View [ā†‘/ā†“] Navigate [k] Kill [s] Stop [l] Logs [r] Refresh [q] Quit` - -### Detail Mode (After pressing Enter) - -- **ONLY visible:** Full-width agents table (9 rows) + full-width logs (9 rows) -- **Hidden:** Clusters table and stats box (completely invisible) -- Dedicated space for monitoring single cluster in depth -- Help text: `[Esc] Back [k] Kill [s] Stop [e] Export [l] Logs [r] Refresh [q] Quit` - -### Navigation Flow - -``` -Overview (ONLY clusters + stats) - ā†“ Enter -Detail (ONLY agents + logs) - ā†“ Escape -Overview (ONLY clusters + stats) -``` - -## Implementation Details - -### Layout Design - -**Overview mode layout** (`src/tui/layout.js`): - -- Clusters table: rows 0-16 (16 rows), cols 0-8 -- Stats box: rows 0-16 (16 rows), cols 8-12 -- Help bar: rows 18-20 -- Agents/logs: **hidden** (`.hide()` called on initialization) - -**Detail mode layout** (`src/tui/layout.js`): - -- Agents table: rows 0-9 (9 rows), cols 0-12 (full width) -- Logs box: rows 9-18 (9 rows), cols 0-12 (full width) -- Help bar: rows 18-20 -- Clusters/stats: **hidden** (`.hide()` called on mode switch) - -### State Management - -**New state in TUI class (`src/tui/index.js`):** - -```javascript -this.viewMode = 'overview'; // or 'detail' -this.detailClusterId = null; // cluster ID when in detail mode -``` - -### Keybindings - -**Enter key** (`src/tui/keybindings.js` lines 14-37): - -- Checks if in overview mode with clusters available -- Sets `viewMode = 'detail'` and `detailClusterId` -- **Hides** clusters table and stats box (`.hide()`) -- **Shows** agents table and logs box (`.show()`) -- Updates help text -- Clears old messages - -**Escape key** (`src/tui/keybindings.js` lines 39-59): - -- Checks if in detail mode -- Sets `viewMode = 'overview'` and `detailClusterId = null` -- **Shows** clusters table and stats box (`.show()`) -- **Hides** agents table and logs box (`.hide()`) -- Updates help text -- Clears messages - -### Conditional Rendering - -**Cluster state updates** (`src/tui/index.js` lines 107-119): - -```javascript -if (this.viewMode === 'detail' && this.detailClusterId) { - // Show agents for detail cluster - const status = this.orchestrator.getStatus(this.detailClusterId); - this.renderer.renderAgentTable(status.agents, this.resourceStats); -} else if (this.viewMode === 'overview') { - // Don't show agents in overview - this.renderer.renderAgentTable([], this.resourceStats); -} -``` - -**Resource stats updates** (`src/tui/index.js` lines 130-137): - -- Same conditional logic as above -- Only renders agents in detail mode - -## Testing - -### Automated Tests - -**Test 1: `tests/tui-integration.test.js`** - -- Basic TUI startup -- Data loading -- Module integration -- āœ… PASSING - -**Test 2: `tests/tui-navigation-test.js`** - -- Initial state verification -- Enter detail view -- Verify agents shown -- Return to overview -- Conditional rendering logic -- āœ… PASSING - -### Manual Testing - -**Run the manual test:** - -```bash -chmod +x tests/tui-keybindings-manual-test.js -node tests/tui-keybindings-manual-test.js -``` - -**Instructions:** - -1. Press ā†‘/ā†“ or j/k to navigate clusters -2. Press Enter to drill into detail view ā†’ clusters/stats hide, agents/logs appear -3. Press Escape to return to overview ā†’ clusters/stats reappear, agents/logs hide -4. Verify help text updates correctly - -## Files Modified - -| File | Changes | -| ------------------------ | ------------------------------------------------------- | -| `src/tui/index.js` | Added viewMode state, conditional rendering | -| `src/tui/keybindings.js` | Added Enter/Escape handlers, widget visibility toggling | -| `src/tui/layout.js` | Updated help text to show Enter key | -| `src/tui/CHANGES.txt` | Documented feature and technical changes | - -## Files Created - -| File | Purpose | -| -------------------------------------- | --------------------------------------- | -| `tests/tui-navigation-test.js` | Automated test for two-level navigation | -| `tests/tui-keybindings-manual-test.js` | Interactive manual test | -| `src/tui/TWO-LEVEL-NAVIGATION.md` | This document | - -## Performance Impact - -- **Startup:** No impact (viewMode check is O(1)) -- **Rendering:** Slight improvement in overview mode (no agent data fetching) -- **Memory:** Minimal increase (2 new state variables) - -## Known Limitations - -None. Feature is complete and tested. - -## Usage - -```bash -# Start TUI (shows overview by default) -zeroshot watch - -# In overview: -# - Use ā†‘/ā†“ or j/k to select cluster -# - Press Enter to drill into detail view - -# In detail: -# - View agents and logs for selected cluster -# - Press Escape to return to overview -``` - -## Future Enhancements (Optional) - -- Add breadcrumb showing current cluster in detail mode -- Add keybinding to jump directly to a cluster by ID -- Add "pinning" to keep detail view on specific cluster even when new clusters spawn diff --git a/src/tui/data-poller.js b/src/tui/data-poller.js deleted file mode 100644 index b777467d..00000000 --- a/src/tui/data-poller.js +++ /dev/null @@ -1,349 +0,0 @@ -/** - * DataPoller - Aggregates cluster data for TUI display - * - * Polls all data sources at appropriate intervals: - * - Cluster states (1s) - * - Resource stats via pidusage (2s) - * - New cluster detection (2s) - * - Ledger message streaming (500ms per cluster) - */ - -const pidusage = require('pidusage'); -const Ledger = require('../ledger'); -const fs = require('fs'); -const path = require('path'); -const os = require('os'); - -class DataPoller { - constructor(orchestrator, options = {}) { - this.orchestrator = orchestrator; - this.intervals = []; - this.ledgers = new Map(); // clusterId -> Ledger instance - this.ledgerStopFns = new Map(); // clusterId -> stop function for pollForMessages - this.onUpdate = options.onUpdate || (() => {}); // Callback for updates - this.watchForNewClustersStopFn = null; - } - - /** - * Start all polling intervals - */ - start() { - // Poll cluster states (1s) - const clusterStateInterval = setInterval(() => { - this._pollClusterStates(); - }, 1000); - this.intervals.push(clusterStateInterval); - - // Poll resource stats (2s) - const resourceStatsInterval = setInterval(() => { - this._pollResourceStats(); - }, 2000); - this.intervals.push(resourceStatsInterval); - - // Watch for new clusters (2s) - this._watchForNewClusters(); - - // Defer initial polls to avoid blocking UI startup - // Run in background after 50ms to let UI render first - setTimeout(() => { - this._pollClusterStates(); - }, 50); - - setTimeout(() => { - this._pollResourceStats(); - }, 100); - } - - /** - * Stop all polling intervals and clean up resources - */ - stop() { - // Clear all intervals - for (const intervalId of this.intervals) { - clearInterval(intervalId); - } - this.intervals = []; - - // Stop watching for new clusters - if (this.watchForNewClustersStopFn) { - this.watchForNewClustersStopFn(); - this.watchForNewClustersStopFn = null; - } - - // Stop all ledger polling - for (const stopFn of this.ledgerStopFns.values()) { - stopFn(); - } - this.ledgerStopFns.clear(); - - // Close all ledger connections - for (const ledger of this.ledgers.values()) { - try { - ledger.close(); - } catch { - // Ignore errors during cleanup - } - } - this.ledgers.clear(); - } - - /** - * Poll cluster states (1s interval) - * Gets all clusters and their agent states from orchestrator - * @private - */ - _pollClusterStates() { - try { - const clusters = this.orchestrator.listClusters(); - - // Get detailed status for each cluster - const clustersWithStatus = clusters.map((cluster) => { - try { - const status = this.orchestrator.getStatus(cluster.id); - // Add agentCount for stats calculation - return { - ...status, - agentCount: status.agents ? status.agents.length : 0, - }; - } catch (error) { - console.error( - `[DataPoller] Failed to get status for cluster ${cluster.id}:`, - error.message - ); - return { - id: cluster.id, - state: 'unknown', - createdAt: cluster.createdAt, - agents: [], - agentCount: 0, - messageCount: 0, - }; - } - }); - - this.onUpdate({ - type: 'cluster_state', - clusters: clustersWithStatus, - }); - } catch (error) { - console.error('[DataPoller] _pollClusterStates error:', error.message); - } - } - - /** - * Poll resource stats (2s interval) - * Uses pidusage to get CPU and memory for all agent processes - * @private - */ - async _pollResourceStats() { - try { - const clusters = this.orchestrator.listClusters(); - const stats = {}; - const pids = this._collectClusterPids(clusters); - - if (pids.length > 0) { - const pidStats = await this._safePidUsage(pids); - this._populatePidStats(stats, pids, pidStats); - } - - this.onUpdate({ - type: 'resource_stats', - stats, - }); - } catch (error) { - console.error('[DataPoller] _pollResourceStats error:', error.message); - } - } - - /** - * Watch for new clusters (2s interval) - * Uses orchestrator.watchForNewClusters to detect new clusters - * and start streaming their ledger messages - * @private - */ - _watchForNewClusters() { - this.watchForNewClustersStopFn = this.orchestrator.watchForNewClusters((cluster) => { - // Lazy load ledger only when we need to stream messages - // This avoids loading all ledgers on startup - const ledger = this._loadLedgerForCluster(cluster, { - label: 'cluster', - requireExisting: true, - }); - - if (!ledger) { - return; - } - - // Start streaming messages - this._streamLedgerMessages(cluster.id); - - // Emit update about new cluster - this.onUpdate({ - type: 'new_cluster', - cluster, - }); - }, 2000); - - // Also load ledgers for all existing clusters - const existingClusters = this.orchestrator.listClusters(); - for (const cluster of existingClusters) { - const ledger = this._loadLedgerForCluster(cluster, { label: 'existing cluster' }); - if (!ledger) { - continue; - } - - this._streamLedgerMessages(cluster.id); - } - } - - /** - * Stream ledger messages for a cluster (500ms interval) - * Uses ledger.pollForMessages to get new messages - * @param {string} clusterId - Cluster ID to stream messages from - * @private - */ - _streamLedgerMessages(clusterId) { - const ledger = this.ledgers.get(clusterId); - if (!ledger) { - console.error(`[DataPoller] No ledger found for cluster ${clusterId}`); - return; - } - - // Stop existing polling if any - const existingStopFn = this.ledgerStopFns.get(clusterId); - if (existingStopFn) { - existingStopFn(); - } - - // Start polling for messages - const stopFn = ledger.pollForMessages( - clusterId, - (message) => { - this.onUpdate({ - type: 'new_message', - clusterId, - message, - }); - }, - 500, // Poll every 500ms - 50 // Show last 50 messages initially - ); - - this.ledgerStopFns.set(clusterId, stopFn); - } - - /** - * Collect resource stats for all agent PIDs - * @returns {Object} Map of pid -> { cpu, memory } - * @private - */ - async _collectResourceStats() { - const stats = {}; - const clusters = this.orchestrator.listClusters(); - - for (const cluster of clusters) { - const pids = this._getClusterAgentPids(cluster); - for (const pid of pids) { - stats[pid] = await this._getSinglePidStat(pid); - } - } - - return stats; - } - - _collectClusterPids(clusters) { - const pids = []; - - for (const cluster of clusters) { - pids.push(...this._getClusterAgentPids(cluster)); - } - - return pids; - } - - _getClusterAgentPids(cluster) { - try { - const status = this.orchestrator.getStatus(cluster.id); - const pids = []; - - for (const agent of status.agents || []) { - if (agent.pid) { - pids.push(agent.pid); - } - } - - return pids; - } catch { - // Skip clusters that error - return []; - } - } - - async _getSinglePidStat(pid) { - try { - const pidStat = await pidusage(pid); - return { - cpu: pidStat.cpu || 0, - memory: pidStat.memory || 0, - }; - } catch { - // Process died - set to zero - return { cpu: 0, memory: 0 }; - } - } - - async _safePidUsage(pids) { - try { - return await pidusage(pids); - } catch { - return null; - } - } - - _populatePidStats(stats, pids, pidStats) { - for (const pid of pids) { - const entry = pidStats?.[pid]; - stats[pid] = { - cpu: entry?.cpu || 0, - memory: entry?.memory || 0, - }; - } - } - - _loadLedgerForCluster(cluster, options) { - const { label, requireExisting = false } = options; - - try { - return this._ensureLedger(cluster.id, { requireExisting }); - } catch (error) { - console.error( - `[DataPoller] Failed to load ledger for ${label} ${cluster.id}:`, - error.message - ); - return null; - } - } - - _ensureLedger(clusterId, { requireExisting = false } = {}) { - const existing = this.ledgers.get(clusterId); - if (existing) { - return existing; - } - - const dbPath = this._getLedgerPath(clusterId); - if (requireExisting && !fs.existsSync(dbPath)) { - return null; - } - - const ledger = new Ledger(dbPath); - this.ledgers.set(clusterId, ledger); - return ledger; - } - - _getLedgerPath(clusterId) { - const storageDir = this.orchestrator.storageDir || path.join(os.homedir(), '.zeroshot'); - return path.join(storageDir, `${clusterId}.db`); - } -} - -module.exports = DataPoller; diff --git a/src/tui/demo.js b/src/tui/demo.js deleted file mode 100644 index 6d5bbf2b..00000000 --- a/src/tui/demo.js +++ /dev/null @@ -1,208 +0,0 @@ -/** - * TUI Dashboard Demo - * Simple demonstration of the dashboard layout with mock data - * - * Run: node src/tui/demo.js - * Press: [q] to quit - */ - -const blessed = require('blessed'); -const { - createLayout, - updateClustersTable, - updateAgentsTable, - updateStatsBox, - addLogEntry, -} = require('./layout'); -const { formatTimestamp } = require('./formatters'); - -// Create main screen -const screen = blessed.screen({ - mouse: true, - title: 'Cluster Dashboard - Demo', - smartCSR: true, -}); - -// Create layout -const layout = createLayout(screen); - -// Mock data generators -const mockClusters = [ - { - id: 'cluster-swift-falcon', - status: 'running', - agentCount: 5, - config: 'default', - uptime: formatTimestamp(2 * 60 * 60 * 1000 + 30 * 60 * 1000), // 2h 30m - }, - { - id: 'cluster-bold-panther', - status: 'running', - agentCount: 3, - config: 'simple', - uptime: formatTimestamp(45 * 60 * 1000), // 45m - }, - { - id: 'cluster-quick-eagle', - status: 'stopped', - agentCount: 0, - config: 'default', - uptime: '0s', - }, -]; - -const mockAgents = [ - { - clusterId: 'cluster-swift-falcon', - id: 'worker-1', - role: 'worker', - status: 'running', - iteration: 3, - cpu: '12.5%', - memory: '245 MB', - }, - { - clusterId: 'cluster-swift-falcon', - id: 'validator-req', - role: 'validator', - status: 'idle', - iteration: 0, - cpu: '0.1%', - memory: '128 MB', - }, - { - clusterId: 'cluster-swift-falcon', - id: 'validator-sec', - role: 'validator', - status: 'idle', - iteration: 0, - cpu: '0.2%', - memory: '135 MB', - }, - { - clusterId: 'cluster-bold-panther', - id: 'worker-2', - role: 'worker', - status: 'running', - iteration: 1, - cpu: '8.3%', - memory: '189 MB', - }, - { - clusterId: 'cluster-bold-panther', - id: 'validator-qa', - role: 'validator', - status: 'running', - iteration: 1, - cpu: '5.1%', - memory: '156 MB', - }, -]; - -const mockStats = { - activeClusters: 2, - totalAgents: 5, - usedMemory: '853 MB', - totalMemory: '8 GB', - totalCPU: '26.2%', -}; - -// Keyboard shortcuts -screen.key(['q', 'C-c'], () => { - return process.exit(0); -}); - -screen.key(['r'], () => { - updateClustersTable(layout.clustersTable, mockClusters); - updateAgentsTable(layout.agentTable, mockAgents); - updateStatsBox(layout.statsBox, mockStats); - addLogEntry(layout.logsBox, 'Dashboard refreshed', 'info'); - screen.render(); -}); - -screen.key(['c'], () => { - addLogEntry(layout.logsBox, 'Cluster started: cluster-wandering-wolf', 'info'); - screen.render(); -}); - -screen.key(['k'], () => { - addLogEntry(layout.logsBox, 'Cluster killed: cluster-quick-eagle', 'warn'); - screen.render(); -}); - -screen.key(['s'], () => { - addLogEntry(layout.logsBox, 'Warning: High memory usage on cluster-swift-falcon', 'warn'); - screen.render(); -}); - -// Initialize with mock data -updateClustersTable(layout.clustersTable, mockClusters); -updateAgentsTable(layout.agentTable, mockAgents); -updateStatsBox(layout.statsBox, mockStats); - -// Add initial log entries -addLogEntry(layout.logsBox, 'Dashboard initialized', 'info'); -addLogEntry(layout.logsBox, 'Monitoring 2 active clusters', 'info'); -addLogEntry(layout.logsBox, 'System CPU: 26.2% | Memory: 853 MB / 8 GB', 'info'); - -// Simulate live updates -const updateInterval = setInterval(() => { - // Update uptime for running clusters - mockClusters.forEach((cluster) => { - if (cluster.status === 'running') { - const uptimeMs = Math.random() * 3 * 60 * 60 * 1000; // Random uptime - cluster.uptime = formatTimestamp(uptimeMs); - } - }); - - // Simulate CPU/Memory changes - mockAgents.forEach((agent) => { - if (agent.status === 'running') { - agent.cpu = (Math.random() * 20).toFixed(1) + '%'; - agent.memory = Math.floor(Math.random() * 200 + 100) + ' MB'; - } - }); - - mockStats.totalCPU = (Math.random() * 50).toFixed(1) + '%'; - - updateClustersTable(layout.clustersTable, mockClusters); - updateAgentsTable(layout.agentTable, mockAgents); - updateStatsBox(layout.statsBox, mockStats); - - screen.render(); -}, 3000); - -// Display help on startup -setTimeout(() => { - addLogEntry( - layout.logsBox, - 'Press [r] to refresh | [c] to add cluster | [k] to kill | [s] for warning | [q] to quit', - 'info' - ); - screen.render(); -}, 500); - -// Cleanup on exit -process.on('exit', () => { - clearInterval(updateInterval); -}); - -// Render initial screen -screen.render(); - -console.log( - '\n' + - '===============================================\n' + - ' Cluster Dashboard - Demo Mode\n' + - '===============================================\n' + - 'Keyboard shortcuts:\n' + - ' [ā†‘/ā†“] Navigate between widgets\n' + - ' [Tab] Next widget\n' + - ' [Shift+Tab] Previous widget\n' + - ' [r] Refresh data\n' + - ' [c] Simulate cluster start\n' + - ' [k] Simulate cluster kill\n' + - ' [s] Simulate warning\n' + - ' [q] Quit\n' + - '===============================================\n\n' -); diff --git a/src/tui/formatters.js b/src/tui/formatters.js deleted file mode 100644 index 6ad9c585..00000000 --- a/src/tui/formatters.js +++ /dev/null @@ -1,129 +0,0 @@ -/** - * TUI Display Formatters - * Converts raw values to human-readable formats for terminal display - */ - -/** - * Convert milliseconds to human-readable uptime string - * @param {number} ms - Milliseconds - * @returns {string} Formatted uptime (e.g., "5m 23s", "2h 15m", "3d 4h") - */ -const formatTimestamp = (ms) => { - if (!ms || ms < 0) return '0s'; - - const seconds = Math.floor(ms / 1000); - - if (seconds < 60) { - return `${seconds}s`; - } - - const minutes = Math.floor(seconds / 60); - if (minutes < 60) { - const remainingSeconds = seconds % 60; - return remainingSeconds > 0 ? `${minutes}m ${remainingSeconds}s` : `${minutes}m`; - } - - const hours = Math.floor(minutes / 60); - if (hours < 24) { - const remainingMinutes = minutes % 60; - return remainingMinutes > 0 ? `${hours}h ${remainingMinutes}m` : `${hours}h`; - } - - const days = Math.floor(hours / 24); - const remainingHours = hours % 24; - return remainingHours > 0 ? `${days}d ${remainingHours}h` : `${days}d`; -}; - -/** - * Convert bytes to human-readable size string - * @param {number} bytes - Number of bytes - * @returns {string} Formatted size (e.g., "245 MB", "1.2 GB", "512 KB") - */ -const formatBytes = (bytes) => { - if (!bytes || bytes < 0) return '0 B'; - - const units = ['B', 'KB', 'MB', 'GB', 'TB']; - let size = bytes; - let unitIndex = 0; - - while (size >= 1000 && unitIndex < units.length - 1) { - size /= 1000; - unitIndex++; - } - - const formatted = size < 10 ? size.toFixed(1) : Math.round(size); - return `${formatted} ${units[unitIndex]}`; -}; - -/** - * Format CPU percentage with consistent precision - * @param {number} percent - CPU percentage (0-100) - * @returns {string} Formatted percentage (e.g., "12.3%", "0.1%") - */ -const formatCPU = (percent) => { - if (typeof percent !== 'number' || percent < 0) return '0.0%'; - - // Assert normalized range (should never exceed 100% after per-core normalization) - let normalizedPercent = percent; - if (normalizedPercent > 100) { - console.warn(`[formatCPU] CPU percent ${percent}% exceeds 100% - normalization bug?`); - normalizedPercent = 100; - } - - return `${normalizedPercent.toFixed(1)}%`; -}; - -/** - * Map cluster state to unicode icon - * @param {string} state - Cluster state (running, stopped, initializing, stopping, failed, killed) - * @returns {string} Unicode icon representing state - */ -const stateIcon = (state) => { - const icons = { - running: 'ā—', // filled circle (green) - stopped: 'ā—‹', // hollow circle - initializing: 'ā—', // half circle - stopping: 'ā—‘', // half circle other way - failed: 'āš ', // warning - killed: 'āš ', // warning - }; - - return icons[state] || '?'; -}; - -/** - * Truncate string with ellipsis if exceeds max length - * @param {string} str - String to truncate - * @param {number} maxLen - Maximum length - * @returns {string} Truncated string with "..." if needed - */ -const truncate = (str, maxLen) => { - if (!str || typeof str !== 'string') return ''; - if (str.length <= maxLen) return str; - - return str.substring(0, maxLen - 3) + '...'; -}; - -/** - * Format duration between two timestamps - * @param {number} startMs - Start timestamp in milliseconds - * @param {number} endMs - End timestamp in milliseconds (null = now) - * @returns {string} Formatted duration (e.g., "5m 23s", "2h 15m") - */ -const formatDuration = (startMs, endMs) => { - if (!startMs || startMs < 0) return '0s'; - - const end = endMs && endMs > 0 ? endMs : Date.now(); - const duration = Math.max(0, end - startMs); - - return formatTimestamp(duration); -}; - -module.exports = { - formatTimestamp, - formatBytes, - formatCPU, - stateIcon, - truncate, - formatDuration, -}; diff --git a/src/tui/index.js b/src/tui/index.js deleted file mode 100644 index b11e9667..00000000 --- a/src/tui/index.js +++ /dev/null @@ -1,193 +0,0 @@ -/** - * TUI - Main interactive dashboard - * - * Coordinates: - * - Screen and layout - * - Data polling - * - Rendering - * - Keybindings - * - State management - */ - -const blessed = require('blessed'); -const { createLayout } = require('./layout'); -const DataPoller = require('./data-poller'); -const Renderer = require('./renderer'); -const { setupKeybindings } = require('./keybindings'); - -class TUI { - constructor(options) { - this.orchestrator = options.orchestrator; - this.filter = options.filter || 'running'; - this.refreshRate = options.refreshRate || 1000; - - // State - this.clusters = []; - this.resourceStats = new Map(); - this.messages = []; - this.selectedIndex = 0; - this.poller = null; - this.renderer = null; - this.widgets = null; - this.screen = null; - - // View mode: 'overview' or 'detail' - this.viewMode = 'overview'; - this.detailClusterId = null; - } - - start() { - // Create screen - this.screen = blessed.screen({ - smartCSR: true, - title: 'Vibe Cluster Watch', - dockBorders: true, - fullUnicode: true, - }); - - // Create layout - this.widgets = createLayout(this.screen); - - // Show immediate loading message - this.widgets.statsBox.setContent('{center}{bold}Loading...{/bold}{/center}'); - this.screen.render(); - - // Create renderer - this.renderer = new Renderer(this.widgets, this.screen); - - // Setup keybindings (pass TUI instance for state management) - setupKeybindings(this.screen, this.widgets, this, this.orchestrator); - - // Create data poller - this.poller = new DataPoller(this.orchestrator, { - refreshRate: this.refreshRate, - onUpdate: (update) => this._handleUpdate(update), - }); - - // Initial message - this.messages.push({ - timestamp: new Date().toISOString(), - text: 'TUI started. Press ? for help.', - level: 'info', - }); - this.renderer.renderLogs(this.messages.slice(-20)); - - // Start polling - this.poller.start(); - - // Initial render - this.screen.render(); - } - - _handleUpdate(update) { - // Update state based on update.type - switch (update.type) { - case 'cluster_state': - // Update cluster list - this.clusters = update.clusters; - - // Apply filter - let filteredClusters = this.clusters; - if (this.filter === 'running') { - // For "running" filter, only show truly active (running) clusters - // Exclude initializing, stopped, failed, etc. - filteredClusters = this.clusters.filter((c) => c.state === 'running'); - } else if (this.filter !== 'all') { - // For other specific filters, match exact state - filteredClusters = this.clusters.filter((c) => c.state === this.filter); - } - - // Ensure selectedIndex is valid - if (this.selectedIndex >= filteredClusters.length) { - this.selectedIndex = Math.max(0, filteredClusters.length - 1); - } - - // Render clusters table - this.renderer.renderClustersTable(filteredClusters, this.selectedIndex); - - // Render system stats - this.renderer.renderSystemStats(this.clusters, this.resourceStats); - - // Update agent table for selected cluster (ONLY in detail view) - if (this.viewMode === 'detail' && this.detailClusterId) { - // In detail view, show agents for the detail cluster - try { - const status = this.orchestrator.getStatus(this.detailClusterId); - this.renderer.renderAgentTable(status.agents, this.resourceStats); - } catch { - // Cluster might have been stopped/killed - this.renderer.renderAgentTable([], this.resourceStats); - } - } else if (this.viewMode === 'overview') { - // In overview view, don't show agents (or show empty) - this.renderer.renderAgentTable([], this.resourceStats); - } - break; - - case 'resource_stats': - // Update resource stats - this.resourceStats = update.stats; - - // Re-render system stats - this.renderer.renderSystemStats(this.clusters, this.resourceStats); - - // Update agent table with new resource stats (ONLY in detail view) - if (this.viewMode === 'detail' && this.detailClusterId) { - try { - const status = this.orchestrator.getStatus(this.detailClusterId); - this.renderer.renderAgentTable(status.agents, this.resourceStats); - } catch { - this.renderer.renderAgentTable([], this.resourceStats); - } - } - break; - - case 'new_message': - // Only add messages from the selected cluster - const selectedClusterId = this.renderer.selectedClusterId; - if (selectedClusterId && update.clusterId === selectedClusterId) { - // Add new message to log - this.messages.push(update.message); - - // Keep only last 100 messages in memory - if (this.messages.length > 100) { - this.messages = this.messages.slice(-100); - } - - // Render last 20 messages - this.renderer.renderLogs(this.messages.slice(-20)); - } - break; - - case 'error': - // Add error to log - this.messages.push({ - timestamp: new Date().toISOString(), - text: `āœ— ${update.error}`, - level: 'error', - }); - - if (this.messages.length > 100) { - this.messages = this.messages.slice(-100); - } - - this.renderer.renderLogs(this.messages.slice(-20)); - break; - } - - // Render screen - this.screen.render(); - } - - exit() { - if (this.poller) { - this.poller.stop(); - } - if (this.screen) { - this.screen.destroy(); - } - process.exit(0); - } -} - -module.exports = TUI; diff --git a/src/tui/keybindings.js b/src/tui/keybindings.js deleted file mode 100644 index 43084594..00000000 --- a/src/tui/keybindings.js +++ /dev/null @@ -1,324 +0,0 @@ -/** - * Keybindings for TUI - * - * Handles: - * - Navigation (up/down, j/k) - * - Actions (kill, stop, export, logs) - * - Confirmations for destructive actions - */ - -const blessed = require('blessed'); -const fs = require('fs'); -const { spawn } = require('child_process'); -const { execSync } = require('../lib/safe-exec'); // Enforces timeouts - -const HELP_TEXT_DETAIL = - '{cyan-fg}[Esc]{/} Back {cyan-fg}[k]{/} Kill {cyan-fg}[s]{/} Stop {cyan-fg}[e]{/} Export {cyan-fg}[l]{/} Logs {cyan-fg}[r]{/} Refresh {cyan-fg}[q]{/} Quit'; -const HELP_TEXT_OVERVIEW = - '{cyan-fg}[Enter]{/} View {cyan-fg}[ā†‘/ā†“]{/} Navigate {cyan-fg}[k]{/} Kill {cyan-fg}[s]{/} Stop {cyan-fg}[l]{/} Logs {cyan-fg}[r]{/} Refresh {cyan-fg}[q]{/} Quit'; - -function getSelectedCluster(tui) { - if (tui.clusters.length === 0) { - return null; - } - - return tui.clusters[tui.selectedIndex] || null; -} - -function pushLogMessage(tui, text, level) { - tui.messages.push({ - timestamp: new Date().toISOString(), - text, - level, - }); - tui.renderer.renderLogs(tui.messages.slice(-20)); -} - -function enterDetailView(screen, widgets, tui) { - if (tui.viewMode !== 'overview') { - return; - } - - const selectedCluster = getSelectedCluster(tui); - if (!selectedCluster) { - return; - } - - tui.viewMode = 'detail'; - tui.detailClusterId = selectedCluster.id; - tui.renderer.setSelectedCluster(selectedCluster.id); - tui.messages = []; - - widgets.helpBar.setContent(HELP_TEXT_DETAIL); - widgets.clustersTable.hide(); - widgets.statsBox.hide(); - widgets.agentTable.show(); - widgets.logsBox.show(); - screen.render(); -} - -function exitDetailView(screen, widgets, tui) { - if (tui.viewMode !== 'detail') { - return; - } - - tui.viewMode = 'overview'; - tui.detailClusterId = null; - tui.renderer.setSelectedCluster(null); - tui.messages = []; - - widgets.helpBar.setContent(HELP_TEXT_OVERVIEW); - widgets.clustersTable.show(); - widgets.statsBox.show(); - widgets.agentTable.hide(); - widgets.logsBox.hide(); - screen.render(); -} - -function moveSelection(screen, tui, orchestrator, delta) { - if (tui.clusters.length === 0) { - return; - } - - tui.selectedIndex = Math.min(tui.clusters.length - 1, Math.max(0, tui.selectedIndex + delta)); - tui.renderer.renderClustersTable(tui.clusters, tui.selectedIndex); - - const selectedCluster = tui.clusters[tui.selectedIndex]; - if (selectedCluster) { - tui.renderer.setSelectedCluster(selectedCluster.id); - tui.messages = []; - - const status = orchestrator.getStatus(selectedCluster.id); - tui.renderer.renderAgentTable(status.agents, tui.resourceStats); - } - - screen.render(); -} - -function createConfirmationDialog(screen, label, color) { - const labelText = color - ? ` {bold}{${color}-fg}${label}{/${color}-fg}{/bold} ` - : ` {bold}${label}{/bold} `; - - return blessed.question({ - parent: screen, - border: 'line', - height: 'shrink', - width: 'half', - top: 'center', - left: 'center', - label: labelText, - tags: true, - keys: true, - vi: true, - }); -} - -function confirmClusterAction(options) { - const { screen, tui, selectedCluster, label, color, prompt, action, successText, failureText } = - options; - const question = createConfirmationDialog(screen, label, color); - - question.ask(prompt, async (err, value) => { - if (err || !value) { - return; - } - - try { - await action(selectedCluster); - pushLogMessage(tui, successText(selectedCluster), 'success'); - } catch (error) { - pushLogMessage(tui, failureText(error), 'error'); - } - - screen.render(); - }); -} - -function handleKillCluster(screen, tui, orchestrator) { - const selectedCluster = getSelectedCluster(tui); - if (!selectedCluster) { - return; - } - - confirmClusterAction({ - screen, - tui, - selectedCluster, - label: 'Confirm Kill', - color: 'red', - prompt: `Kill cluster ${selectedCluster.id}?\n\n(This will force-stop all agents)`, - action: (cluster) => orchestrator.kill(cluster.id), - successText: (cluster) => `āœ“ Killed cluster ${cluster.id}`, - failureText: (error) => `āœ— Failed to kill cluster: ${error.message}`, - }); -} - -function handleStopCluster(screen, tui, orchestrator) { - const selectedCluster = getSelectedCluster(tui); - if (!selectedCluster) { - return; - } - - confirmClusterAction({ - screen, - tui, - selectedCluster, - label: 'Confirm Stop', - color: 'yellow', - prompt: `Stop cluster ${selectedCluster.id}?\n\n(This will gracefully stop all agents)`, - action: (cluster) => orchestrator.stop(cluster.id), - successText: (cluster) => `āœ“ Stopped cluster ${cluster.id}`, - failureText: (error) => `āœ— Failed to stop cluster: ${error.message}`, - }); -} - -function handleExportCluster(screen, tui, orchestrator) { - const selectedCluster = getSelectedCluster(tui); - if (!selectedCluster) { - return; - } - - try { - const markdown = orchestrator.export(selectedCluster.id, 'markdown'); - const filename = `${selectedCluster.id}-export.md`; - fs.writeFileSync(filename, markdown); - pushLogMessage(tui, `āœ“ Exported cluster to ${filename}`, 'success'); - } catch (error) { - pushLogMessage(tui, `āœ— Failed to export cluster: ${error.message}`, 'error'); - } - - screen.render(); -} - -function findTerminalCommand() { - const terminals = ['gnome-terminal', 'konsole', 'xterm', 'urxvt', 'alacritty', 'kitty']; - - for (const terminal of terminals) { - try { - execSync(`which ${terminal}`, { stdio: 'ignore' }); - return terminal; - } catch { - // Ignore missing terminal - } - } - - return 'xterm'; -} - -function buildLogCommand(clusterId) { - const term = - process.env.TERM_PROGRAM || (process.env.COLORTERM ? 'gnome-terminal' : null) || 'xterm'; - - if (term === 'iTerm.app' || term === 'Apple_Terminal') { - return { - cmd: 'osascript', - args: ['-e', `tell application "Terminal" to do script "zeroshot logs ${clusterId} -f"`], - }; - } - - const cmd = findTerminalCommand(); - const logCommand = `zeroshot logs ${clusterId} -f; read -p "Press enter to close..."`; - - if (cmd === 'gnome-terminal' || cmd === 'konsole') { - return { cmd, args: ['--', 'bash', '-c', logCommand] }; - } - - return { cmd, args: ['-e', 'bash', '-c', logCommand] }; -} - -function handleOpenLogs(screen, tui) { - const selectedCluster = getSelectedCluster(tui); - if (!selectedCluster) { - return; - } - - try { - const { cmd, args } = buildLogCommand(selectedCluster.id); - spawn(cmd, args, { detached: true, stdio: 'ignore' }); - pushLogMessage(tui, `āœ“ Opened logs for ${selectedCluster.id} in new terminal`, 'success'); - } catch (error) { - pushLogMessage(tui, `āœ— Failed to open logs: ${error.message}`, 'error'); - } - - screen.render(); -} - -function handleRefresh(screen, tui) { - pushLogMessage(tui, 'ā†» Refreshing...', 'info'); - screen.render(); - - if (tui.poller) { - tui.poller.poll(); - } -} - -function handleExit(screen, tui) { - const question = createConfirmationDialog(screen, 'Confirm Exit'); - - question.ask('Exit TUI?\n\n(Clusters will continue running)', (err, value) => { - if (err || !value) { - return; - } - - tui.exit(); - }); -} - -function handleHelp(screen) { - const helpBox = blessed.box({ - parent: screen, - border: 'line', - height: '80%', - width: '60%', - top: 'center', - left: 'center', - label: ' {bold}Keybindings{/bold} ', - tags: true, - keys: true, - vi: true, - scrollable: true, - alwaysScroll: true, - content: ` -{bold}Navigation:{/bold} - ā†‘/k Move selection up - ā†“/j Move selection down - -{bold}Actions:{/bold} - K Kill selected cluster (force stop) - s Stop selected cluster (graceful) - e Export selected cluster to markdown - l Show full logs in new terminal - r Force refresh - -{bold}Other:{/bold} - ?/h Show this help - q/Ctrl-C Exit TUI - -Press any key to close... - `.trim(), - }); - - helpBox.key(['escape', 'q', 'enter', 'space'], () => { - helpBox.destroy(); - screen.render(); - }); - - screen.render(); -} - -function setupKeybindings(screen, widgets, tui, orchestrator) { - screen.key(['enter'], () => enterDetailView(screen, widgets, tui)); - screen.key(['escape'], () => exitDetailView(screen, widgets, tui)); - screen.key(['up', 'k'], () => moveSelection(screen, tui, orchestrator, -1)); - screen.key(['down', 'j'], () => moveSelection(screen, tui, orchestrator, 1)); - screen.key(['K'], () => handleKillCluster(screen, tui, orchestrator)); - screen.key(['s'], () => handleStopCluster(screen, tui, orchestrator)); - screen.key(['e'], () => handleExportCluster(screen, tui, orchestrator)); - screen.key(['l'], () => handleOpenLogs(screen, tui)); - screen.key(['r'], () => handleRefresh(screen, tui)); - screen.key(['q', 'C-c'], () => handleExit(screen, tui)); - screen.key(['?', 'h'], () => handleHelp(screen)); -} - -module.exports = { setupKeybindings }; diff --git a/src/tui/layout.js b/src/tui/layout.js index 7fde06d4..9e4e700d 100644 --- a/src/tui/layout.js +++ b/src/tui/layout.js @@ -10,7 +10,12 @@ */ const blessed = require('blessed'); -const contrib = require('blessed-contrib'); +// Pull in only the widgets we use to avoid loading optional picture widget. +const contrib = { + grid: require('blessed-contrib/lib/layout/grid'), + table: require('blessed-contrib/lib/widget/table'), + log: require('blessed-contrib/lib/widget/log'), +}; /** * Create main TUI layout with grid-based widget organization diff --git a/src/tui/renderer.js b/src/tui/renderer.js deleted file mode 100644 index f714cf04..00000000 --- a/src/tui/renderer.js +++ /dev/null @@ -1,184 +0,0 @@ -/** - * TUI Screen Renderer - * Transforms polled data into widget updates using formatters and layout widgets - */ - -const { formatTimestamp, formatBytes, formatCPU, stateIcon, truncate } = require('./formatters'); - -class Renderer { - /** - * Create renderer instance - * @param {object} widgets - Widget objects from layout.js - * @param {object} screen - Blessed screen instance - */ - constructor(widgets, screen) { - if (!widgets) { - throw new Error('Renderer requires widgets object from layout'); - } - if (!screen) { - throw new Error('Renderer requires screen instance'); - } - - this.widgets = widgets; - this.screen = screen; - this.selectedClusterId = null; - } - - /** - * Set the currently selected cluster ID - * @param {string|null} id - Cluster ID to select - */ - setSelectedCluster(id) { - this.selectedClusterId = id; - } - - /** - * Render clusters table with state icons and uptime - * @param {Array} clusters - Array of cluster objects - */ - renderClustersTable(clusters) { - const clusterList = !clusters || !Array.isArray(clusters) ? [] : clusters; - - const data = clusterList.map((c) => { - if (!c) return ['', '', '', '']; - - const icon = stateIcon(c.state || 'unknown'); - const uptime = - c.state === 'running' && c.createdAt ? formatTimestamp(Date.now() - c.createdAt) : '-'; - const clusterId = truncate(c.id || '', 18); - const state = (c.state || 'unknown').toUpperCase(); - const agentCount = `${c.agentCount || 0} agents`; - - return [`${icon} ${clusterId}`, state, agentCount, uptime]; - }); - - if (this.widgets.clustersTable && this.widgets.clustersTable.setData) { - this.widgets.clustersTable.setData({ - headers: ['ID', 'State', 'Agents', 'Uptime'], - data, - }); - } - } - - /** - * Render system statistics box with aggregate metrics - * @param {Array} clusters - Array of cluster objects - * @param {Map} resourceStats - Map of PID -> {cpu, memory} - */ - renderSystemStats(clusters, resourceStats) { - const clusterList = !clusters || !Array.isArray(clusters) ? [] : clusters; - const statsMap = !resourceStats || !(resourceStats instanceof Map) ? new Map() : resourceStats; - - // Calculate aggregate stats - const activeClusters = clusterList.filter((c) => c && c.state === 'running').length; - const totalAgents = clusterList.reduce((sum, c) => sum + (c?.agentCount || 0), 0); - - // Calculate average CPU and memory from resource stats - let totalCpu = 0; - let totalMemory = 0; - let statCount = 0; - - statsMap.forEach((stat) => { - if (stat && typeof stat.cpu === 'number' && typeof stat.memory === 'number') { - totalCpu += stat.cpu; - totalMemory += stat.memory; - statCount++; - } - }); - - const avgCpu = statCount > 0 ? totalCpu / statCount : 0; - const avgMemory = statCount > 0 ? totalMemory / statCount : 0; - - // Format output with blessed color tags - const statsText = [ - '{cyan-fg}Active Clusters:{/} ' + activeClusters, - '{cyan-fg}Total Agents:{/} ' + totalAgents, - '{cyan-fg}Avg CPU:{/} ' + formatCPU(avgCpu), - '{cyan-fg}Avg Memory:{/} ' + formatBytes(avgMemory), - ].join('\n'); - - if (this.widgets.statsBox && this.widgets.statsBox.setContent) { - this.widgets.statsBox.setContent(statsText); - } - } - - /** - * Render agent table for selected cluster - * @param {Array} agents - Array of agent objects - * @param {Map} resourceStats - Map of PID -> {cpu, memory} - */ - renderAgentTable(agents, resourceStats) { - if (!this.selectedClusterId) { - // No cluster selected, show empty table - if (this.widgets.agentTable && this.widgets.agentTable.setData) { - this.widgets.agentTable.setData({ - headers: ['Agent', 'Role', 'State', 'Iter', 'CPU%', 'Mem'], - data: [], - }); - } - return; - } - - const agentList = !agents || !Array.isArray(agents) ? [] : agents; - const statsMap = !resourceStats || !(resourceStats instanceof Map) ? new Map() : resourceStats; - - const data = agentList.map((a) => { - if (!a) return ['', '', '', '', '', '']; - - const pid = a.pid; - const stats = statsMap.get(pid) || { cpu: 0, memory: 0 }; - - const agentId = truncate(a.id || '', 12); - const role = truncate(a.role || '', 12); - const state = a.state || 'unknown'; - const iteration = `${a.iteration || 0}/${a.maxIterations || 0}`; - const cpu = formatCPU(stats.cpu); - const memory = formatBytes(stats.memory); - - return [agentId, role, state, iteration, cpu, memory]; - }); - - if (this.widgets.agentTable && this.widgets.agentTable.setData) { - this.widgets.agentTable.setData({ - headers: ['Agent', 'Role', 'State', 'Iter', 'CPU%', 'Mem'], - data, - }); - } - } - - /** - * Render log messages to log widget - * @param {Array} messages - Array of message objects - */ - renderLogs(messages) { - if (!messages || !Array.isArray(messages)) { - return; - } - - if (!this.widgets.logsBox || !this.widgets.logsBox.log) { - return; - } - - messages.forEach((msg) => { - if (!msg) return; - - const timestamp = msg.timestamp || Date.now(); - const time = new Date(timestamp).toLocaleTimeString(); - const sender = truncate(msg.sender || 'unknown', 15); - const text = truncate(msg.content?.text || '', 60); - - this.widgets.logsBox.log(`[${time}] ${sender}: ${text}`); - }); - } - - /** - * Trigger screen render to update display - */ - render() { - if (this.screen && this.screen.render) { - this.screen.render(); - } - } -} - -module.exports = Renderer; diff --git a/task-lib/attachable-watcher.js b/task-lib/attachable-watcher.js index 42524f00..fedeaf0e 100644 --- a/task-lib/attachable-watcher.js +++ b/task-lib/attachable-watcher.js @@ -9,7 +9,11 @@ import { appendFileSync, existsSync, mkdirSync } from 'fs'; import { join } from 'path'; import { homedir } from 'os'; import { updateTask } from './store.js'; -import { detectStreamingModeError, recoverStructuredOutput } from './claude-recovery.js'; +import { + detectFatalClaudeError, + detectStreamingModeError, + recoverStructuredOutput, +} from './claude-recovery.js'; import { createRequire } from 'module'; // ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā• @@ -70,6 +74,7 @@ const cwd = cwdArg; const logFile = logFileArg; const args = JSON.parse(argsJsonArg); const config = configJsonArg ? JSON.parse(configJsonArg) : {}; +let server = null; const SOCKET_DIR = join(homedir(), '.zeroshot', 'sockets'); const socketPath = join(SOCKET_DIR, `${taskId}.sock`); @@ -95,6 +100,7 @@ const silentJsonMode = let finalResultJson = null; let outputBuffer = ''; let streamingModeError = null; +let fatalError = null; function splitBufferLines(buffer, chunk) { const nextBuffer = buffer + chunk; @@ -103,6 +109,29 @@ function splitBufferLines(buffer, chunk) { return { lines, remaining }; } +function maybeHandleFatalError(line, timestamp) { + if (!enableRecovery || fatalError) { + return false; + } + + const detected = detectFatalClaudeError(line); + if (!detected) { + return false; + } + + fatalError = detected; + + if (silentJsonMode) { + log(`[${timestamp}]${line}\n`); + } + log(`[${timestamp}][FATAL] ${detected}\n`); + + if (server) { + server.stop('SIGTERM').catch(() => {}); + } + return true; +} + function captureStreamingError(line, timestamp) { if (!enableRecovery) { return false; @@ -131,6 +160,7 @@ function maybeCaptureStructuredOutput(line) { function handleSilentJsonLines(lines, timestamp) { for (const line of lines) { if (!line.trim()) continue; + maybeHandleFatalError(line, timestamp); if (captureStreamingError(line, timestamp)) { continue; } @@ -140,6 +170,7 @@ function handleSilentJsonLines(lines, timestamp) { function handleStreamingLines(lines, timestamp) { for (const line of lines) { + maybeHandleFatalError(line, timestamp); if (captureStreamingError(line, timestamp)) { continue; } @@ -159,6 +190,7 @@ function flushOutputBuffer(timestamp) { return; } + maybeHandleFatalError(outputBuffer, timestamp); if (captureStreamingError(outputBuffer, timestamp)) { return; } @@ -205,7 +237,7 @@ function writeCompletionFooter(code, signal) { log(`Exit code: ${code}, Signal: ${signal}\n`); } -const server = new AttachServer({ +server = new AttachServer({ id: taskId, socketPath, command, @@ -244,13 +276,13 @@ server.on('exit', async ({ exitCode, signal }) => { writeCompletionFooter(code, signal); - const resolvedCode = recovered?.payload ? 0 : code; + const resolvedCode = fatalError ? 1 : recovered?.payload ? 0 : code; const status = resolvedCode === 0 ? 'completed' : 'failed'; try { await updateTask(taskId, { status, exitCode: resolvedCode, - error: resolvedCode !== 0 && signal ? `Killed by ${signal}` : null, + error: fatalError || (resolvedCode !== 0 && signal ? `Killed by ${signal}` : null), socketPath: null, }); } catch (updateError) { diff --git a/task-lib/claude-recovery.js b/task-lib/claude-recovery.js index 81462c02..1220704f 100644 --- a/task-lib/claude-recovery.js +++ b/task-lib/claude-recovery.js @@ -3,6 +3,7 @@ import { join } from 'path'; import { homedir } from 'os'; export const STREAMING_MODE_ERROR = 'only prompt commands are supported in streaming mode'; +export const NO_MESSAGES_RETURNED = 'No messages returned'; export function detectStreamingModeError(line) { const trimmed = typeof line === 'string' ? line.trim() : ''; @@ -30,6 +31,27 @@ export function detectStreamingModeError(line) { return null; } +export function detectFatalClaudeError(line) { + if (typeof line !== 'string') return null; + const trimmed = line.trim(); + if (!trimmed) return null; + + if (trimmed.startsWith('{')) { + try { + JSON.parse(trimmed); + return null; + } catch { + // Not valid JSON, continue detection + } + } + + if (trimmed.toLowerCase().includes(NO_MESSAGES_RETURNED.toLowerCase())) { + return `Claude CLI error: ${NO_MESSAGES_RETURNED}`; + } + + return null; +} + function findSessionJsonlPath(sessionId) { const claudeDir = process.env.CLAUDE_CONFIG_DIR || join(homedir(), '.claude'); const projectsDir = join(claudeDir, 'projects'); diff --git a/task-lib/config.js b/task-lib/config.js index f3c1a0d7..47a9f449 100644 --- a/task-lib/config.js +++ b/task-lib/config.js @@ -1,7 +1,10 @@ -import { homedir } from 'os'; import { join } from 'path'; +import { homedir } from 'os'; + +const HOME_DIR = + process.env.ZEROSHOT_HOME || process.env.HOME || process.env.USERPROFILE || homedir(); -export const TASKS_DIR = join(homedir(), '.claude-zeroshot'); +export const TASKS_DIR = join(HOME_DIR, '.claude-zeroshot'); export const LOGS_DIR = join(TASKS_DIR, 'logs'); export const SCHEDULER_PID_FILE = join(TASKS_DIR, 'scheduler.pid'); export const SCHEDULER_LOG = join(TASKS_DIR, 'scheduler.log'); diff --git a/task-lib/tui/layout.js b/task-lib/tui/layout.js index d2127621..ac32b8ba 100644 --- a/task-lib/tui/layout.js +++ b/task-lib/tui/layout.js @@ -9,7 +9,8 @@ */ import blessed from 'blessed'; -import contrib from 'blessed-contrib'; +import Grid from 'blessed-contrib/lib/layout/grid.js'; +import Log from 'blessed-contrib/lib/widget/log.js'; /** * Create task logs TUI layout @@ -19,7 +20,7 @@ import contrib from 'blessed-contrib'; */ function createLayout(screen, taskId) { // Create 20x12 grid for responsive layout - const grid = new contrib.grid({ rows: 20, cols: 12, screen }); + const grid = new Grid({ rows: 20, cols: 12, screen }); // ============================================================ // TASK INFO BOX (3 rows x 12 cols) @@ -46,7 +47,7 @@ function createLayout(screen, taskId) { // Scrollable log output with auto-scroll // ============================================================ - const logsBox = grid.set(3, 0, 15, 12, contrib.log, { + const logsBox = grid.set(3, 0, 15, 12, Log, { fg: 'white', label: ' Live Logs ', border: { type: 'line', fg: 'cyan' }, diff --git a/task-lib/watcher.js b/task-lib/watcher.js index 7ad58348..ea36169b 100644 --- a/task-lib/watcher.js +++ b/task-lib/watcher.js @@ -8,7 +8,11 @@ import { spawn } from 'child_process'; import { appendFileSync } from 'fs'; import { updateTask } from './store.js'; -import { detectStreamingModeError, recoverStructuredOutput } from './claude-recovery.js'; +import { + detectFatalClaudeError, + detectStreamingModeError, + recoverStructuredOutput, +} from './claude-recovery.js'; import { createRequire } from 'module'; const require = createRequire(import.meta.url); @@ -42,6 +46,7 @@ const silentJsonMode = let finalResultJson = null; let streamingModeError = null; +let fatalError = null; let stdoutBuffer = ''; let stderrBuffer = ''; @@ -53,6 +58,42 @@ function splitBufferLines(buffer, chunk) { return { lines, remaining }; } +function maybeHandleFatalError(line, timestamp) { + if (!enableRecovery || fatalError) { + return false; + } + + const detected = detectFatalClaudeError(line); + if (!detected) { + return false; + } + + fatalError = detected; + + if (silentJsonMode) { + log(`[${timestamp}]${line}\n`); + } + log(`[${timestamp}][FATAL] ${detected}\n`); + + try { + child.kill('SIGTERM'); + } catch { + // Ignore - process may already be dead + } + + setTimeout(() => { + if (child.exitCode === null) { + try { + child.kill('SIGKILL'); + } catch { + // Ignore - process may already be dead + } + } + }, 5000); + + return true; +} + function captureStreamingError(line, timestamp) { if (!enableRecovery) { return false; @@ -81,6 +122,7 @@ function maybeCaptureStructuredOutput(line) { function handleSilentJsonLines(lines, timestamp) { for (const line of lines) { if (!line.trim()) continue; + maybeHandleFatalError(line, timestamp); if (captureStreamingError(line, timestamp)) { continue; } @@ -90,6 +132,7 @@ function handleSilentJsonLines(lines, timestamp) { function handleStreamingLines(lines, timestamp) { for (const line of lines) { + maybeHandleFatalError(line, timestamp); if (captureStreamingError(line, timestamp)) { continue; } @@ -109,6 +152,7 @@ function flushStdoutBuffer(timestamp) { return; } + maybeHandleFatalError(stdoutBuffer, timestamp); if (captureStreamingError(stdoutBuffer, timestamp)) { return; } @@ -123,6 +167,7 @@ function flushStdoutBuffer(timestamp) { function flushStderrBuffer(timestamp) { if (stderrBuffer.trim()) { + maybeHandleFatalError(stderrBuffer, timestamp); log(`[${timestamp}]${stderrBuffer}\n`); } } @@ -201,13 +246,13 @@ child.on('close', async (code, signal) => { writeCompletionFooter(code, signal); - const resolvedCode = recovered?.payload ? 0 : code; + const resolvedCode = fatalError ? 1 : recovered?.payload ? 0 : code; const status = resolvedCode === 0 ? 'completed' : 'failed'; try { await updateTask(taskId, { status, exitCode: resolvedCode, - error: resolvedCode !== 0 && signal ? `Killed by ${signal}` : null, + error: fatalError || (resolvedCode !== 0 && signal ? `Killed by ${signal}` : null), }); } catch (updateError) { log(`[${Date.now()}][ERROR] Failed to update task status: ${updateError.message}\n`); diff --git a/tests/add-agents-trigger-merge.test.js b/tests/add-agents-trigger-merge.test.js new file mode 100644 index 00000000..e0bf3d34 --- /dev/null +++ b/tests/add-agents-trigger-merge.test.js @@ -0,0 +1,257 @@ +/** + * Test: add_agents should REPLACE agents with duplicate IDs entirely + * + * HISTORY: + * - Original behavior: Merged triggers but kept old hooks ā†’ BUG + * - Bug manifestation: heavy-validation consensus-coordinator used quick-validation's + * hooks, publishing QUICK_VALIDATION_PASSED instead of VALIDATION_RESULT ā†’ infinite loop + * - Fix: Replace agent entirely when same ID encountered + * + * DESIGN DECISION: Same ID = same agent = full replacement + * If you need different behavior, use different agent IDs. + */ + +const assert = require('assert'); +const fs = require('fs'); +const path = require('path'); +const os = require('os'); +const Orchestrator = require('../src/orchestrator.js'); +const MockTaskRunner = require('./helpers/mock-task-runner.js'); + +// Isolate tests from user settings +const testSettingsDir = fs.mkdtempSync(path.join(os.tmpdir(), 'zeroshot-test-settings-')); +const testSettingsFile = path.join(testSettingsDir, 'settings.json'); +fs.writeFileSync(testSettingsFile, JSON.stringify({ maxModel: 'opus', minModel: null })); +process.env.ZEROSHOT_SETTINGS_FILE = testSettingsFile; + +function createTempDir() { + const tmpBase = path.join(os.tmpdir(), 'zeroshot-test'); + if (!fs.existsSync(tmpBase)) { + fs.mkdirSync(tmpBase, { recursive: true }); + } + return fs.mkdtempSync(path.join(tmpBase, 'agent-replace-')); +} + +function cleanupTempDir(tmpDir) { + if (fs.existsSync(tmpDir)) { + fs.rmSync(tmpDir, { recursive: true, force: true }); + } +} + +describe('add_agents duplicate ID handling', function () { + this.timeout(10000); + + let tmpDir; + let orchestrator; + let mockRunner; + + beforeEach(function () { + tmpDir = createTempDir(); + mockRunner = new MockTaskRunner(); + }); + + afterEach(function () { + if (orchestrator) { + orchestrator.close(); + orchestrator = null; + } + cleanupTempDir(tmpDir); + }); + + it('should REPLACE agent entirely when adding agent with duplicate ID', async function () { + orchestrator = new Orchestrator({ + dataDir: tmpDir, + taskRunner: mockRunner, + quiet: true, + }); + + // Initial agent with QUICK trigger and quick-specific hooks + const initialConfig = { + agents: [ + { + id: 'consensus-coordinator', + role: 'coordinator', + modelLevel: 'level2', + triggers: [{ topic: 'QUICK_VALIDATION_RESULT', action: 'execute_task' }], + prompt: 'Quick validation coordinator.', + hooks: { + onComplete: { + action: 'publish_message', + config: { topic: 'QUICK_VALIDATION_PASSED', content: { text: 'Stage 1 passed' } }, + }, + }, + }, + ], + }; + + const result = await orchestrator.start(initialConfig, { text: 'Test task' }); + const clusterId = result.id; + const cluster = orchestrator.getCluster(clusterId); + + // Verify initial state + const agentBefore = cluster.agents.find((a) => a.id === 'consensus-coordinator'); + assert.ok(agentBefore, 'consensus-coordinator should exist'); + assert.strictEqual(agentBefore.config.triggers.length, 1); + assert.strictEqual(agentBefore.config.hooks.onComplete.config.topic, 'QUICK_VALIDATION_PASSED'); + + // Add agent with SAME ID but DIFFERENT triggers and hooks (simulating heavy-validation) + await orchestrator._opAddAgents( + cluster, + { + agents: [ + { + id: 'consensus-coordinator', // Same ID! + role: 'coordinator', + modelLevel: 'level2', + triggers: [{ topic: 'HEAVY_VALIDATION_RESULT', action: 'execute_task' }], + prompt: 'Heavy validation coordinator.', + hooks: { + onComplete: { + action: 'publish_message', + config: { topic: 'VALIDATION_RESULT', content: { text: 'All validations passed' } }, + }, + }, + }, + ], + }, + {} + ); + + // Verify REPLACEMENT occurred (not merge) + const clusterAfter = orchestrator.getCluster(clusterId); + const agentAfter = clusterAfter.agents.find((a) => a.id === 'consensus-coordinator'); + + assert.ok(agentAfter, 'consensus-coordinator should still exist'); + + // CRITICAL: Should have ONLY the new trigger (not merged) + assert.strictEqual( + agentAfter.config.triggers.length, + 1, + 'Should have 1 trigger (replaced, not merged)' + ); + assert.strictEqual( + agentAfter.config.triggers[0].topic, + 'HEAVY_VALIDATION_RESULT', + 'Should have the NEW trigger' + ); + + // CRITICAL: Should have NEW hooks (the bug was keeping old hooks) + assert.strictEqual( + agentAfter.config.hooks.onComplete.config.topic, + 'VALIDATION_RESULT', + 'Should have NEW hooks (not old QUICK_VALIDATION_PASSED)' + ); + + // Verify prompt was also replaced + assert.strictEqual(agentAfter.config.prompt, 'Heavy validation coordinator.'); + + await orchestrator.stop(clusterId); + }); + + it('should allow adding agents with different IDs', async function () { + orchestrator = new Orchestrator({ + dataDir: tmpDir, + taskRunner: mockRunner, + quiet: true, + }); + + const initialConfig = { + agents: [ + { + id: 'quick-consensus', + role: 'coordinator', + triggers: [{ topic: 'QUICK_RESULT', action: 'execute_task' }], + prompt: 'Quick coordinator.', + }, + ], + }; + + const result = await orchestrator.start(initialConfig, { text: 'Test task' }); + const clusterId = result.id; + const cluster = orchestrator.getCluster(clusterId); + + // Add a DIFFERENT agent (different ID) + await orchestrator._opAddAgents( + cluster, + { + agents: [ + { + id: 'heavy-consensus', // Different ID + role: 'coordinator', + triggers: [{ topic: 'HEAVY_RESULT', action: 'execute_task' }], + prompt: 'Heavy coordinator.', + }, + ], + }, + {} + ); + + const clusterAfter = orchestrator.getCluster(clusterId); + + // Should have BOTH agents + assert.strictEqual(clusterAfter.agents.length, 2, 'Should have 2 agents'); + + const quickAgent = clusterAfter.agents.find((a) => a.id === 'quick-consensus'); + const heavyAgent = clusterAfter.agents.find((a) => a.id === 'heavy-consensus'); + + assert.ok(quickAgent, 'quick-consensus should exist'); + assert.ok(heavyAgent, 'heavy-consensus should exist'); + + await orchestrator.stop(clusterId); + }); + + it('should replace agent instance entirely when same ID added', async function () { + orchestrator = new Orchestrator({ + dataDir: tmpDir, + taskRunner: mockRunner, + quiet: true, + }); + + const initialConfig = { + agents: [ + { + id: 'test-agent', + role: 'validator', + triggers: [{ topic: 'TEST', action: 'execute_task' }], + prompt: 'Test agent.', + }, + ], + }; + + const result = await orchestrator.start(initialConfig, { text: 'Test task' }); + const clusterId = result.id; + const cluster = orchestrator.getCluster(clusterId); + + // cluster.agents contains AgentWrapper instances directly + const agentBefore = cluster.agents.find((a) => a.id === 'test-agent'); + assert.ok(agentBefore, 'Agent should exist before replacement'); + + // Replace the agent + await orchestrator._opAddAgents( + cluster, + { + agents: [ + { + id: 'test-agent', + role: 'validator', + triggers: [{ topic: 'TEST2', action: 'execute_task' }], + prompt: 'Replaced agent.', + }, + ], + }, + {} + ); + + const clusterAfter = orchestrator.getCluster(clusterId); + const agentAfter = clusterAfter.agents.find((a) => a.id === 'test-agent'); + + // Should be a DIFFERENT AgentWrapper instance (not the same object) + assert.notStrictEqual(agentAfter, agentBefore, 'Should be new AgentWrapper instance'); + + // Verify the new config was applied + assert.strictEqual(agentAfter.config.triggers[0].topic, 'TEST2'); + assert.strictEqual(agentAfter.config.prompt, 'Replaced agent.'); + + await orchestrator.stop(clusterId); + }); +}); diff --git a/tests/agent-task-not-found.test.js b/tests/agent-task-not-found.test.js new file mode 100644 index 00000000..d253ea61 --- /dev/null +++ b/tests/agent-task-not-found.test.js @@ -0,0 +1,159 @@ +/** + * TEST: Agent Task Not Found - Fail-Safe Restart Behavior + * + * Verifies that when zeroshot status returns "ID not found", the agent + * immediately fails with restart error instead of polling 30 times. + * + * POSTMORTEM (2026-01-29): Agent worker polling failed 30 times when task + * completed and was cleaned up before worker could detect it. Worker treated + * "not found" as retryable error, wasting 30+ seconds before giving up. + * + * FIX: Detect "ID not found" immediately ā†’ return error ā†’ trigger restart (fail-safe) + * + * This test verifies the fix at the code level by reading the implementation. + * Integration test would require complex mocking of child processes and timers. + */ + +const assert = require('assert'); +const fs = require('fs'); +const path = require('path'); + +describe('Agent Task Not Found - Fail-Safe Restart', () => { + const sourceFile = path.join(__dirname, '..', 'src', 'agent', 'agent-task-executor.js'); + let sourceCode; + + before(() => { + sourceCode = fs.readFileSync(sourceFile, 'utf8'); + }); + + it('should have handleStatusExecError function that detects "ID not found"', () => { + // Verify the function exists + assert.ok( + sourceCode.includes('function handleStatusExecError'), + 'handleStatusExecError function should exist' + ); + + // Verify it checks for "ID not found" pattern + assert.ok( + sourceCode.includes('ID not found') || sourceCode.includes('Not found in tasks'), + 'Should check for "ID not found" or "Not found in tasks" patterns' + ); + }); + + it('should check both error.message and stderr for "not found" patterns', () => { + // Extract the handleStatusExecError function + const functionMatch = sourceCode.match( + /function handleStatusExecError\([^)]*\)\s*{[\s\S]*?^}/m + ); + + assert.ok(functionMatch, 'Should find handleStatusExecError function'); + + const functionBody = functionMatch[0]; + + // Verify it checks error message + assert.ok( + functionBody.includes('error.message') || functionBody.includes('errorMessage'), + 'Should check error.message for patterns' + ); + + // Verify it checks stderr + assert.ok( + functionBody.includes('stderr') || functionBody.includes('stderrMessage'), + 'Should check stderr for patterns' + ); + + // Verify it looks for both "ID not found" and "Not found in tasks" + const hasIdNotFound = functionBody.includes('ID not found'); + const hasNotFoundInTasks = functionBody.includes('Not found in tasks'); + + assert.ok( + hasIdNotFound && hasNotFoundInTasks, + 'Should check for both "ID not found" and "Not found in tasks" patterns' + ); + }); + + it('should return error immediately when task not found (not retry)', () => { + const functionMatch = sourceCode.match( + /function handleStatusExecError\([^)]*\)\s*{[\s\S]*?^}/m + ); + + assert.ok(functionMatch, 'Should find handleStatusExecError function'); + + const functionBody = functionMatch[0]; + + // Verify it has a dedicated "not found" check BEFORE the retry counter + const notFoundCheckIndex = functionBody.indexOf('ID not found'); + const retryCounterIndex = functionBody.indexOf('consecutiveExecFailures++'); + + assert.ok( + notFoundCheckIndex > 0 && notFoundCheckIndex < retryCounterIndex, + 'Should check for "not found" BEFORE incrementing retry counter' + ); + + // Verify it resolves immediately when not found + const notFoundSection = functionBody.substring(notFoundCheckIndex, retryCounterIndex); + + assert.ok( + notFoundSection.includes('resolve('), + 'Should call resolve() immediately when task not found' + ); + + assert.ok( + notFoundSection.includes('success: false'), + 'Should resolve with success: false when task not found' + ); + + assert.ok( + notFoundSection.includes('restarting') || notFoundSection.includes('restart'), + 'Error message should mention restarting' + ); + }); + + it('should publish AGENT_ERROR event when task not found', () => { + const functionMatch = sourceCode.match( + /function handleStatusExecError\([^)]*\)\s*{[\s\S]*?^}/m + ); + + const functionBody = functionMatch[0]; + + // Find the "not found" section + const notFoundStart = functionBody.indexOf('ID not found'); + const retryCounterIndex = functionBody.indexOf('consecutiveExecFailures++'); + const notFoundSection = functionBody.substring(notFoundStart, retryCounterIndex); + + // Verify it publishes an error event + assert.ok( + notFoundSection.includes('_publish') && notFoundSection.includes('AGENT_ERROR'), + 'Should publish AGENT_ERROR event when task not found' + ); + + // Verify error type is appropriate + assert.ok( + notFoundSection.includes('task_not_found') || notFoundSection.includes('not_found'), + 'Error event should have appropriate error type' + ); + }); + + it('should have improved warning message', () => { + const functionMatch = sourceCode.match( + /function handleStatusExecError\([^)]*\)\s*{[\s\S]*?^}/m + ); + + const functionBody = functionMatch[0]; + + // Verify warning message is helpful + const notFoundStart = functionBody.indexOf('ID not found'); + const retryCounterIndex = functionBody.indexOf('consecutiveExecFailures++'); + const notFoundSection = functionBody.substring(notFoundStart, retryCounterIndex); + + assert.ok( + notFoundSection.includes('will restart') || notFoundSection.includes('restarting'), + 'Warning message should explain that task will be restarted' + ); + + assert.ok( + notFoundSection.includes('ensure completion') || notFoundSection.includes('safety'), + 'Warning message should explain fail-safe reasoning' + ); + }); +}); diff --git a/tests/cannot-validate-status.test.js b/tests/cannot-validate-status.test.js index 25f68a37..83baf75d 100644 --- a/tests/cannot-validate-status.test.js +++ b/tests/cannot-validate-status.test.js @@ -48,6 +48,7 @@ const baseContextParams = (overrides = {}) => ({ messageBus: { query: () => [] }, cluster: { id: 'test-cluster', createdAt: Date.now() - 60000 }, triggeringMessage: { topic: 'IMPLEMENTATION_READY', sender: 'worker' }, + isolation: null, ...overrides, }); @@ -258,6 +259,28 @@ describe('CANNOT_VALIDATE Context Builder - Core Behavior', function () { assert.ok(context.includes('Do NOT re-attempt'), 'Missing skip instruction'); }); + it('should ignore platform mismatch reasons when running in docker isolation', function () { + const criteria = [ + { + id: 'AC1', + status: 'CANNOT_VALIDATE', + reason: 'npm install fails on darwin-arm64 (EBADPLATFORM for @esbuild/linux-x64)', + }, + { id: 'AC2', status: 'CANNOT_VALIDATE', reason: 'No SSH access to prod' }, + ]; + + const context = buildContext( + baseContextParams({ + messageBus: mockBusWithCriteria(criteria), + isolation: { enabled: true }, + }) + ); + + assert.ok(!context.includes('**AC1**'), 'Should skip platform mismatch criteria'); + assert.ok(!context.includes('EBADPLATFORM'), 'Should remove platform mismatch reason'); + assert.ok(context.includes('**AC2**'), 'Should keep non-platform criteria'); + }); + it('should NOT inject skip section for non-validator roles', function () { const criteria = [{ id: 'AC1', status: 'CANNOT_VALIDATE', reason: 'test' }]; diff --git a/tests/cluster-operations.test.js b/tests/cluster-operations.test.js index e90a2e92..ccdc5b42 100644 --- a/tests/cluster-operations.test.js +++ b/tests/cluster-operations.test.js @@ -116,6 +116,34 @@ describe('CLUSTER_OPERATIONS', function () { 'Should have error about missing ISSUE_OPENED trigger' ); }); + + it('should include agents from load_config when building proposed config (parameterized)', function () { + const existing = [ + { + id: 'worker', + role: 'implementation', + triggers: [{ topic: 'ISSUE_OPENED', action: 'execute_task' }], + }, + ]; + + const ops = [ + { + action: 'load_config', + config: { base: 'quick-validation', params: {} }, + }, + ]; + + const proposed = orchestrator._buildProposedAgentConfigs(existing, ops); + + assert( + proposed.some((a) => a.id === 'validator-requirements'), + 'Expected load_config to add validator-requirements' + ); + assert( + proposed.some((a) => a.id === 'validator-code'), + 'Expected load_config to add validator-code' + ); + }); }); describe('VALID_OPERATIONS constant', function () { diff --git a/tests/conductor-republish-metadata.test.js b/tests/conductor-republish-metadata.test.js index 88cc0446..586cd773 100644 --- a/tests/conductor-republish-metadata.test.js +++ b/tests/conductor-republish-metadata.test.js @@ -31,7 +31,9 @@ describe('Conductor Republish Metadata', function () { // Parse and validate logic script const script = issueTrigger.logic.script; expect(script, 'Logic script should check sender=system').to.include('message.sender === '); - expect(script, 'Logic script should exclude _republished').to.include('!message.metadata?._republished'); + expect(script, 'Logic script should exclude _republished').to.include( + '!message.metadata?._republished' + ); }); it('junior conductor transform should set _republished metadata', function () { diff --git a/tests/config-validator.test.js b/tests/config-validator.test.js index 80970198..d84b6aba 100644 --- a/tests/config-validator.test.js +++ b/tests/config-validator.test.js @@ -137,6 +137,7 @@ describe('modelRules validation', function () { ], }); assert.ok(result.errors.some((e) => e.includes('Add catch-all rule'))); + assert.ok(result.errors.some((e) => e.includes('"modelLevel": "level2"'))); }); it('should accept modelRules with "all" catch-all', function () { @@ -184,7 +185,7 @@ describe('modelRules validation', function () { }, ], }); - assert.ok(result.warnings.some((w) => w.includes('model "gpt4"') && w.includes('claude'))); + assert.ok(result.warnings.some((w) => w.includes('model "gpt4"') && w.includes('not valid'))); }); }); @@ -2298,6 +2299,37 @@ describe('Semantic Validation - Medium Gaps (8-9)', function () { const contextWarnings = result.warnings.filter((w) => w.includes('[Gap 9]')); assert.ok(contextWarnings.length > 0, 'Should have Gap 9 warning'); }); + + it('should not warn for STATE_SNAPSHOT context source', function () { + const config = { + agents: [ + { + id: 'worker', + role: 'implementation', + triggers: [{ topic: 'ISSUE_OPENED' }], + contextStrategy: { + sources: [{ topic: 'STATE_SNAPSHOT', amount: 1 }], + }, + hooks: { + onComplete: { + action: 'publish_message', + config: { topic: 'DONE', content: {} }, + }, + }, + }, + { + id: 'completion', + role: 'orchestrator', + triggers: [{ topic: 'DONE', action: 'stop_cluster' }], + }, + ], + }; + const result = validateConfig(config); + const snapshotWarnings = result.warnings.filter( + (warning) => warning.includes('[Gap 9]') && warning.includes('STATE_SNAPSHOT') + ); + assert.strictEqual(snapshotWarnings.length, 0, 'STATE_SNAPSHOT should not trigger Gap 9'); + }); }); }); @@ -2590,7 +2622,7 @@ describe('Semantic Validation - Medium Gap 14', function () { role: 'implementation', triggers: [{ topic: 'ISSUE_OPENED' }], contextStrategy: { - sources: [{ topic: 'TEST' }], // Missing amount + sources: [{ topic: 'TEST', strategy: 'latest' }], // Missing amount }, hooks: { onComplete: { @@ -2622,6 +2654,48 @@ describe('Semantic Validation - Medium Gap 14', function () { assert.ok(amountWarnings.length > 0, 'Should have Gap 14 warning'); assert.ok(amountWarnings[0].includes('amount')); }); + + it('should accept limit as amount alias without warning', function () { + const config = { + agents: [ + { + id: 'worker', + role: 'implementation', + triggers: [{ topic: 'ISSUE_OPENED' }], + contextStrategy: { + sources: [{ topic: 'TEST', limit: 1 }], + }, + hooks: { + onComplete: { + action: 'publish_message', + config: { topic: 'DONE', content: {} }, + }, + }, + }, + { + id: 'tester', + role: 'validator', + triggers: [{ topic: 'X' }], + hooks: { + onComplete: { + action: 'publish_message', + config: { topic: 'TEST', content: {} }, + }, + }, + }, + { + id: 'completion', + role: 'orchestrator', + triggers: [{ topic: 'DONE', action: 'stop_cluster' }], + }, + ], + }; + const result = validateConfig(config); + const amountWarnings = result.warnings.filter( + (w) => w.includes('[Gap 14]') && w.includes('amount') + ); + assert.strictEqual(amountWarnings.length, 0); + }); }); }); diff --git a/tests/context-packs.test.js b/tests/context-packs.test.js new file mode 100644 index 00000000..cf5dd38d --- /dev/null +++ b/tests/context-packs.test.js @@ -0,0 +1,144 @@ +const assert = require('assert'); +const AgentWrapper = require('../src/agent-wrapper'); +const MessageBus = require('../src/message-bus'); +const Ledger = require('../src/ledger'); +const { buildContextPacks } = require('../src/agent/context-pack-builder'); +const path = require('path'); +const fs = require('fs'); +const os = require('os'); + +describe('context packs', () => { + let tempDir; + let ledger; + let messageBus; + let clusterId; + let clusterCreatedAt; + + beforeEach(() => { + tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'zeroshot-context-packs-')); + const dbPath = path.join(tempDir, 'test-ledger.db'); + + ledger = new Ledger(dbPath); + messageBus = new MessageBus(ledger); + + clusterId = 'test-cluster-789'; + clusterCreatedAt = Date.now(); + }); + + afterEach(() => { + if (ledger) ledger.close(); + try { + fs.rmSync(tempDir, { recursive: true, force: true }); + } catch { + // Ignore cleanup errors + } + }); + + function createWorker(contextStrategy) { + const workerConfig = { + id: 'worker', + role: 'implementation', + modelLevel: 'level2', + timeout: 0, + contextStrategy, + }; + + const mockCluster = { + id: clusterId, + createdAt: clusterCreatedAt, + agents: [], + }; + + return new AgentWrapper(workerConfig, messageBus, mockCluster, { + testMode: true, + mockSpawnFn: () => {}, + }); + } + + function publishMessage(topic, text, timestamp) { + messageBus.publish({ + cluster_id: clusterId, + topic, + sender: 'system', + content: { text }, + timestamp, + }); + } + + function buildTriggeringMessage(timestamp, text = 'triggered') { + return { + topic: 'WORKER_PROGRESS', + sender: 'system', + timestamp, + content: { text }, + }; + } + + it('keeps triggering message and required anchors under tight budgets', () => { + const baseTime = Date.now(); + publishMessage('ISSUE_OPENED', 'Implement feature X', baseTime); + publishMessage('PLAN_READY', '1. Do the thing', baseTime + 10); + publishMessage('OPTIONAL_TOPIC', 'optional-detail', baseTime + 20); + publishMessage('VALIDATION_RESULT', 'rejected: missing test', baseTime + 30); + + const worker = createWorker({ + sources: [ + { topic: 'ISSUE_OPENED', amount: 1 }, + { topic: 'PLAN_READY', amount: 1 }, + { topic: 'OPTIONAL_TOPIC', amount: 3 }, + { topic: 'VALIDATION_RESULT', amount: 3 }, + ], + maxTokens: 1, + }); + + const context = worker._buildContext(buildTriggeringMessage(baseTime + 40, 'kickoff')); + + assert(context.includes('## Triggering Message'), 'Triggering message section must exist'); + assert(context.includes('kickoff'), 'Triggering message content must be preserved'); + assert(context.includes('Messages from topic: ISSUE_OPENED'), 'Issue anchor must be preserved'); + assert(context.includes('Messages from topic: PLAN_READY'), 'Plan anchor must be preserved'); + assert(!context.includes('optional-detail'), 'Low-priority context should be dropped'); + }); + + it('compacts high-priority packs before skipping low-priority packs', () => { + const packs = [ + { + id: 'header', + section: 'header', + priority: 'required', + render: () => 'REQ\n', + }, + { + id: 'high', + section: 'sources', + priority: 'high', + render: () => 'H'.repeat(40), + compact: () => 'H\n', + }, + { + id: 'low', + section: 'sources', + priority: 'low', + render: () => 'L'.repeat(40), + compact: () => 'L\n', + }, + { + id: 'trigger', + section: 'triggeringMessage', + priority: 'required', + preserve: true, + render: () => 'TRIG\n', + }, + ]; + + const result = buildContextPacks({ packs, maxTokens: 4 }); + const highPack = result.packDecisions.find((pack) => pack.id === 'high'); + const lowPack = result.packDecisions.find((pack) => pack.id === 'low'); + + assert.strictEqual(highPack.status, 'included'); + assert.strictEqual(highPack.variant, 'compact'); + assert.strictEqual(lowPack.status, 'skipped'); + assert(result.context.includes('H\n'), 'High-priority compact content should be included'); + assert(!result.context.includes('L'.repeat(40)), 'Low-priority full content should be dropped'); + }); +}); diff --git a/tests/context-source-selection.test.js b/tests/context-source-selection.test.js new file mode 100644 index 00000000..0b35b7c4 --- /dev/null +++ b/tests/context-source-selection.test.js @@ -0,0 +1,171 @@ +const assert = require('assert'); +const AgentWrapper = require('../src/agent-wrapper'); +const MessageBus = require('../src/message-bus'); +const Ledger = require('../src/ledger'); +const path = require('path'); +const fs = require('fs'); +const os = require('os'); + +describe('context source selection', () => { + let tempDir; + let ledger; + let messageBus; + let clusterId; + let clusterCreatedAt; + + beforeEach(() => { + tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'zeroshot-test-')); + const dbPath = path.join(tempDir, 'test-ledger.db'); + + ledger = new Ledger(dbPath); + messageBus = new MessageBus(ledger); + + clusterId = 'test-cluster-456'; + clusterCreatedAt = Date.now(); + }); + + afterEach(() => { + if (ledger) ledger.close(); + try { + fs.rmSync(tempDir, { recursive: true, force: true }); + } catch { + // Ignore cleanup errors + } + }); + + function createWorker(contextStrategy) { + const workerConfig = { + id: 'worker', + role: 'implementation', + modelLevel: 'level2', + timeout: 0, + contextStrategy, + }; + + const mockCluster = { + id: clusterId, + createdAt: clusterCreatedAt, + agents: [], + }; + + return new AgentWrapper(workerConfig, messageBus, mockCluster, { + testMode: true, + mockSpawnFn: () => {}, + }); + } + + function publishMessage(topic, text, timestamp) { + messageBus.publish({ + cluster_id: clusterId, + topic, + sender: 'system', + content: { text }, + timestamp, + }); + } + + function buildTriggeringMessage(timestamp) { + return { + topic: 'ISSUE_OPENED', + sender: 'system', + timestamp, + }; + } + + it('selects latest messages in chronological order', () => { + const baseTime = Date.now(); + publishMessage('TEST_TOPIC', 'first-message', baseTime); + publishMessage('TEST_TOPIC', 'second-message', baseTime + 10); + publishMessage('TEST_TOPIC', 'third-message', baseTime + 20); + + const worker = createWorker({ + sources: [{ topic: 'TEST_TOPIC', amount: 2, strategy: 'latest' }], + }); + + const context = worker._buildContext(buildTriggeringMessage(baseTime + 30)); + + assert(!context.includes('first-message'), 'Should not include oldest message'); + assert(context.includes('second-message'), 'Should include second message'); + assert(context.includes('third-message'), 'Should include latest message'); + assert( + context.indexOf('second-message') < context.indexOf('third-message'), + 'Latest messages should render in chronological order' + ); + }); + + it('selects oldest messages in chronological order', () => { + const baseTime = Date.now(); + publishMessage('TEST_TOPIC', 'alpha-message', baseTime); + publishMessage('TEST_TOPIC', 'beta-message', baseTime + 10); + publishMessage('TEST_TOPIC', 'gamma-message', baseTime + 20); + + const worker = createWorker({ + sources: [{ topic: 'TEST_TOPIC', amount: 2, strategy: 'oldest' }], + }); + + const context = worker._buildContext(buildTriggeringMessage(baseTime + 30)); + + assert(context.includes('alpha-message'), 'Should include oldest message'); + assert(context.includes('beta-message'), 'Should include second message'); + assert(!context.includes('gamma-message'), 'Should not include latest message'); + assert( + context.indexOf('alpha-message') < context.indexOf('beta-message'), + 'Oldest messages should render in chronological order' + ); + }); + + it('selects all messages in chronological order', () => { + const baseTime = Date.now(); + publishMessage('TEST_TOPIC', 'one-message', baseTime); + publishMessage('TEST_TOPIC', 'two-message', baseTime + 10); + publishMessage('TEST_TOPIC', 'three-message', baseTime + 20); + + const worker = createWorker({ + sources: [{ topic: 'TEST_TOPIC', strategy: 'all' }], + }); + + const context = worker._buildContext(buildTriggeringMessage(baseTime + 30)); + + assert(context.includes('one-message'), 'Should include first message'); + assert(context.includes('two-message'), 'Should include second message'); + assert(context.includes('three-message'), 'Should include third message'); + assert( + context.indexOf('one-message') < context.indexOf('two-message'), + 'All messages should render in chronological order' + ); + assert( + context.indexOf('two-message') < context.indexOf('three-message'), + 'All messages should render in chronological order' + ); + }); + + it('uses limit as amount alias with latest default', () => { + const baseTime = Date.now(); + publishMessage('TEST_TOPIC', 'old-message', baseTime); + publishMessage('TEST_TOPIC', 'newer-message', baseTime + 10); + + const worker = createWorker({ + sources: [{ topic: 'TEST_TOPIC', limit: 1 }], + }); + + const context = worker._buildContext(buildTriggeringMessage(baseTime + 20)); + + assert(!context.includes('old-message'), 'Should not include older message'); + assert(context.includes('newer-message'), 'Should include latest message'); + }); + + it('prefers amount when both amount and limit are set', () => { + const baseTime = Date.now(); + publishMessage('TEST_TOPIC', 'older-message', baseTime); + publishMessage('TEST_TOPIC', 'newest-message', baseTime + 10); + + const worker = createWorker({ + sources: [{ topic: 'TEST_TOPIC', amount: 1, limit: 2, strategy: 'latest' }], + }); + + const context = worker._buildContext(buildTriggeringMessage(baseTime + 20)); + + assert(!context.includes('older-message'), 'Should honor amount over limit'); + assert(context.includes('newest-message'), 'Should include latest message'); + }); +}); diff --git a/tests/fixtures/detached-daemon.js b/tests/fixtures/detached-daemon.js new file mode 100644 index 00000000..524be6e7 --- /dev/null +++ b/tests/fixtures/detached-daemon.js @@ -0,0 +1,55 @@ +const Orchestrator = require('../../src/orchestrator'); +const MockTaskRunner = require('../helpers/mock-task-runner'); + +const storageDir = process.env.ZEROSHOT_TEST_STORAGE; +const clusterId = process.env.ZEROSHOT_TEST_CLUSTER_ID; + +if (!storageDir || !clusterId) { + console.error('Missing ZEROSHOT_TEST_STORAGE or ZEROSHOT_TEST_CLUSTER_ID'); + process.exit(1); +} + +const mockRunner = new MockTaskRunner(); +const orchestrator = new Orchestrator({ + quiet: true, + storageDir, + taskRunner: mockRunner, +}); + +const config = { + agents: [ + { + id: 'worker', + role: 'implementation', + timeout: 0, + triggers: [{ topic: 'NEVER', action: 'execute_task' }], + prompt: 'Idle agent for detached stop test', + }, + ], +}; + +async function startCluster() { + await orchestrator.start(config, { text: 'Detached stop test' }, { clusterId }); + console.log('READY'); +} + +async function shutdown(signal) { + try { + await orchestrator.stop(clusterId); + console.log(`[DAEMON] Stopped cluster ${clusterId} from ${signal}`); + } catch (error) { + console.error(`[DAEMON] Failed to stop cluster ${clusterId}: ${error.message}`); + } + process.exit(0); +} + +process.on('SIGTERM', () => shutdown('SIGTERM')); +process.on('SIGINT', () => shutdown('SIGINT')); + +startCluster().catch((error) => { + console.error(`[DAEMON] Failed to start cluster: ${error.message}`); + console.error(error.stack); + process.exit(1); +}); + +setInterval(() => {}, 1000); diff --git a/tests/fixtures/tui-v2/protocol/invalid.params.getClusterSummary.json b/tests/fixtures/tui-v2/protocol/invalid.params.getClusterSummary.json new file mode 100644 index 00000000..3ce90532 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/invalid.params.getClusterSummary.json @@ -0,0 +1,8 @@ +{ + "jsonrpc": "2.0", + "id": 100, + "method": "getClusterSummary", + "params": { + "clusterId": 123 + } +} diff --git a/tests/fixtures/tui-v2/protocol/invalid.request.missing-method.json b/tests/fixtures/tui-v2/protocol/invalid.request.missing-method.json new file mode 100644 index 00000000..f2816670 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/invalid.request.missing-method.json @@ -0,0 +1,4 @@ +{ + "jsonrpc": "2.0", + "id": 99 +} diff --git a/tests/fixtures/tui-v2/protocol/notification.clusterLogLines.json b/tests/fixtures/tui-v2/protocol/notification.clusterLogLines.json new file mode 100644 index 00000000..c3ab9036 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/notification.clusterLogLines.json @@ -0,0 +1,19 @@ +{ + "jsonrpc": "2.0", + "method": "clusterLogLines", + "params": { + "subscriptionId": "sub-logs-1", + "clusterId": "cluster-123", + "lines": [ + { + "id": "log-1", + "timestamp": 1769811111000, + "text": "Agent output", + "agent": "worker", + "role": "implementation", + "sender": "worker" + } + ], + "droppedCount": 0 + } +} diff --git a/tests/fixtures/tui-v2/protocol/notification.clusterTimelineEvents.json b/tests/fixtures/tui-v2/protocol/notification.clusterTimelineEvents.json new file mode 100644 index 00000000..9cb74ae3 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/notification.clusterTimelineEvents.json @@ -0,0 +1,19 @@ +{ + "jsonrpc": "2.0", + "method": "clusterTimelineEvents", + "params": { + "subscriptionId": "sub-timeline-1", + "clusterId": "cluster-123", + "events": [ + { + "id": "evt-1", + "timestamp": 1769811111000, + "topic": "PLAN_READY", + "label": "Plan ready", + "approved": null, + "sender": "planner" + } + ], + "droppedCount": 0 + } +} diff --git a/tests/fixtures/tui-v2/protocol/request.getClusterSummary.json b/tests/fixtures/tui-v2/protocol/request.getClusterSummary.json new file mode 100644 index 00000000..e5e9bd26 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/request.getClusterSummary.json @@ -0,0 +1,8 @@ +{ + "jsonrpc": "2.0", + "id": 3, + "method": "getClusterSummary", + "params": { + "clusterId": "cluster-123" + } +} diff --git a/tests/fixtures/tui-v2/protocol/request.getClusterTopology.json b/tests/fixtures/tui-v2/protocol/request.getClusterTopology.json new file mode 100644 index 00000000..042c26f5 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/request.getClusterTopology.json @@ -0,0 +1,8 @@ +{ + "jsonrpc": "2.0", + "id": 11, + "method": "getClusterTopology", + "params": { + "clusterId": "cluster-123" + } +} diff --git a/tests/fixtures/tui-v2/protocol/request.initialize.json b/tests/fixtures/tui-v2/protocol/request.initialize.json new file mode 100644 index 00000000..5f004610 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/request.initialize.json @@ -0,0 +1,17 @@ +{ + "jsonrpc": "2.0", + "id": 1, + "method": "initialize", + "params": { + "protocolVersion": 1, + "client": { + "name": "zeroshot-tui", + "version": "0.1.0", + "pid": 4242 + }, + "capabilities": { + "wantsMetrics": true, + "wantsTopology": false + } + } +} diff --git a/tests/fixtures/tui-v2/protocol/request.listClusterMetrics.json b/tests/fixtures/tui-v2/protocol/request.listClusterMetrics.json new file mode 100644 index 00000000..bed45323 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/request.listClusterMetrics.json @@ -0,0 +1,8 @@ +{ + "jsonrpc": "2.0", + "id": 4, + "method": "listClusterMetrics", + "params": { + "clusterIds": ["cluster-123", "cluster-456"] + } +} diff --git a/tests/fixtures/tui-v2/protocol/request.listClusters.json b/tests/fixtures/tui-v2/protocol/request.listClusters.json new file mode 100644 index 00000000..9371b943 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/request.listClusters.json @@ -0,0 +1,5 @@ +{ + "jsonrpc": "2.0", + "id": 2, + "method": "listClusters" +} diff --git a/tests/fixtures/tui-v2/protocol/request.sendGuidanceToAgent.json b/tests/fixtures/tui-v2/protocol/request.sendGuidanceToAgent.json new file mode 100644 index 00000000..f2314e9e --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/request.sendGuidanceToAgent.json @@ -0,0 +1,11 @@ +{ + "jsonrpc": "2.0", + "id": 7, + "method": "sendGuidanceToAgent", + "params": { + "clusterId": "cluster-123", + "agentId": "worker", + "text": "Focus on tests", + "timeoutMs": 5000 + } +} diff --git a/tests/fixtures/tui-v2/protocol/request.sendGuidanceToCluster.json b/tests/fixtures/tui-v2/protocol/request.sendGuidanceToCluster.json new file mode 100644 index 00000000..bc635dfd --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/request.sendGuidanceToCluster.json @@ -0,0 +1,10 @@ +{ + "jsonrpc": "2.0", + "id": 8, + "method": "sendGuidanceToCluster", + "params": { + "clusterId": "cluster-123", + "text": "Ship it", + "timeoutMs": 5000 + } +} diff --git a/tests/fixtures/tui-v2/protocol/request.startClusterFromIssue.json b/tests/fixtures/tui-v2/protocol/request.startClusterFromIssue.json new file mode 100644 index 00000000..6d7201d8 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/request.startClusterFromIssue.json @@ -0,0 +1,9 @@ +{ + "jsonrpc": "2.0", + "id": 6, + "method": "startClusterFromIssue", + "params": { + "ref": "covibes/zeroshot#240", + "providerOverride": null + } +} diff --git a/tests/fixtures/tui-v2/protocol/request.startClusterFromText.json b/tests/fixtures/tui-v2/protocol/request.startClusterFromText.json new file mode 100644 index 00000000..d142b669 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/request.startClusterFromText.json @@ -0,0 +1,10 @@ +{ + "jsonrpc": "2.0", + "id": 5, + "method": "startClusterFromText", + "params": { + "text": "Implement the requested feature", + "providerOverride": "codex", + "clusterId": "cluster-789" + } +} diff --git a/tests/fixtures/tui-v2/protocol/request.subscribeClusterLogs.json b/tests/fixtures/tui-v2/protocol/request.subscribeClusterLogs.json new file mode 100644 index 00000000..2f6fde6d --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/request.subscribeClusterLogs.json @@ -0,0 +1,9 @@ +{ + "jsonrpc": "2.0", + "id": 9, + "method": "subscribeClusterLogs", + "params": { + "clusterId": "cluster-123", + "agentId": "worker" + } +} diff --git a/tests/fixtures/tui-v2/protocol/request.subscribeClusterTimeline.json b/tests/fixtures/tui-v2/protocol/request.subscribeClusterTimeline.json new file mode 100644 index 00000000..60fe45e5 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/request.subscribeClusterTimeline.json @@ -0,0 +1,8 @@ +{ + "jsonrpc": "2.0", + "id": 10, + "method": "subscribeClusterTimeline", + "params": { + "clusterId": "cluster-123" + } +} diff --git a/tests/fixtures/tui-v2/protocol/request.unsubscribe.json b/tests/fixtures/tui-v2/protocol/request.unsubscribe.json new file mode 100644 index 00000000..f1369030 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/request.unsubscribe.json @@ -0,0 +1,8 @@ +{ + "jsonrpc": "2.0", + "id": 21, + "method": "unsubscribe", + "params": { + "subscriptionId": "sub-123" + } +} diff --git a/tests/fixtures/tui-v2/protocol/response.getClusterSummary.json b/tests/fixtures/tui-v2/protocol/response.getClusterSummary.json new file mode 100644 index 00000000..511e43dc --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/response.getClusterSummary.json @@ -0,0 +1,15 @@ +{ + "jsonrpc": "2.0", + "id": 3, + "result": { + "summary": { + "id": "cluster-123", + "state": "running", + "provider": "codex", + "createdAt": 1769810000000, + "agentCount": 3, + "messageCount": 120, + "cwd": "/Users/tom/.zeroshot/worktrees/prime-falcon-67" + } + } +} diff --git a/tests/fixtures/tui-v2/protocol/response.getClusterTopology.json b/tests/fixtures/tui-v2/protocol/response.getClusterTopology.json new file mode 100644 index 00000000..1ffd22db --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/response.getClusterTopology.json @@ -0,0 +1,33 @@ +{ + "jsonrpc": "2.0", + "id": 11, + "result": { + "topology": { + "agents": [ + { "id": "planner", "role": "planning" }, + { "id": "worker", "role": "implementation" } + ], + "edges": [ + { + "from": "system", + "to": "ISSUE_OPENED", + "topic": "ISSUE_OPENED", + "kind": "source" + }, + { + "from": "ISSUE_OPENED", + "to": "planner", + "topic": "ISSUE_OPENED", + "kind": "trigger" + }, + { + "from": "planner", + "to": "PLAN_READY", + "topic": "PLAN_READY", + "kind": "publish" + } + ], + "topics": ["ISSUE_OPENED", "PLAN_READY"] + } + } +} diff --git a/tests/fixtures/tui-v2/protocol/response.initialize.json b/tests/fixtures/tui-v2/protocol/response.initialize.json new file mode 100644 index 00000000..cf371728 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/response.initialize.json @@ -0,0 +1,28 @@ +{ + "jsonrpc": "2.0", + "id": 1, + "result": { + "protocolVersion": 1, + "server": { + "name": "zeroshot-backend", + "version": "5.4.0" + }, + "capabilities": { + "methods": [ + "initialize", + "listClusters", + "getClusterSummary", + "listClusterMetrics", + "startClusterFromText", + "startClusterFromIssue", + "sendGuidanceToAgent", + "sendGuidanceToCluster", + "subscribeClusterLogs", + "subscribeClusterTimeline", + "unsubscribe", + "getClusterTopology" + ], + "notifications": ["clusterLogLines", "clusterTimelineEvents"] + } + } +} diff --git a/tests/fixtures/tui-v2/protocol/response.listClusterMetrics.json b/tests/fixtures/tui-v2/protocol/response.listClusterMetrics.json new file mode 100644 index 00000000..19b90ff8 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/response.listClusterMetrics.json @@ -0,0 +1,20 @@ +{ + "jsonrpc": "2.0", + "id": 4, + "result": { + "metrics": [ + { + "id": "cluster-123", + "supported": true, + "cpuPercent": 12.5, + "memoryMB": 256.4 + }, + { + "id": "cluster-456", + "supported": false, + "cpuPercent": null, + "memoryMB": null + } + ] + } +} diff --git a/tests/fixtures/tui-v2/protocol/response.listClusters.json b/tests/fixtures/tui-v2/protocol/response.listClusters.json new file mode 100644 index 00000000..f047025f --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/response.listClusters.json @@ -0,0 +1,26 @@ +{ + "jsonrpc": "2.0", + "id": 2, + "result": { + "clusters": [ + { + "id": "cluster-123", + "state": "running", + "provider": "codex", + "createdAt": 1769810000000, + "agentCount": 3, + "messageCount": 120, + "cwd": "/Users/tom/.zeroshot/worktrees/prime-falcon-67" + }, + { + "id": "cluster-456", + "state": "idle", + "provider": null, + "createdAt": 1769810500000, + "agentCount": 0, + "messageCount": 0, + "cwd": null + } + ] + } +} diff --git a/tests/fixtures/tui-v2/protocol/response.sendGuidanceToAgent.json b/tests/fixtures/tui-v2/protocol/response.sendGuidanceToAgent.json new file mode 100644 index 00000000..39b538d6 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/response.sendGuidanceToAgent.json @@ -0,0 +1,12 @@ +{ + "jsonrpc": "2.0", + "id": 7, + "result": { + "result": { + "status": "injected", + "reason": null, + "method": "pty", + "taskId": "task-1" + } + } +} diff --git a/tests/fixtures/tui-v2/protocol/response.sendGuidanceToCluster.json b/tests/fixtures/tui-v2/protocol/response.sendGuidanceToCluster.json new file mode 100644 index 00000000..0cdadaca --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/response.sendGuidanceToCluster.json @@ -0,0 +1,28 @@ +{ + "jsonrpc": "2.0", + "id": 8, + "result": { + "result": { + "summary": { + "injected": 1, + "queued": 0, + "total": 1 + }, + "agents": { + "worker": { + "status": "injected", + "reason": null, + "method": "pty", + "taskId": "task-1" + }, + "validator": { + "status": "queued", + "reason": "busy", + "method": null, + "taskId": null + } + }, + "timestamp": 1769811111000 + } + } +} diff --git a/tests/fixtures/tui-v2/protocol/response.startClusterFromIssue.json b/tests/fixtures/tui-v2/protocol/response.startClusterFromIssue.json new file mode 100644 index 00000000..3c3fd9ac --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/response.startClusterFromIssue.json @@ -0,0 +1,7 @@ +{ + "jsonrpc": "2.0", + "id": 6, + "result": { + "clusterId": "cluster-456" + } +} diff --git a/tests/fixtures/tui-v2/protocol/response.startClusterFromText.json b/tests/fixtures/tui-v2/protocol/response.startClusterFromText.json new file mode 100644 index 00000000..17986870 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/response.startClusterFromText.json @@ -0,0 +1,7 @@ +{ + "jsonrpc": "2.0", + "id": 5, + "result": { + "clusterId": "cluster-789" + } +} diff --git a/tests/fixtures/tui-v2/protocol/response.subscribeClusterLogs.json b/tests/fixtures/tui-v2/protocol/response.subscribeClusterLogs.json new file mode 100644 index 00000000..f0a0a54d --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/response.subscribeClusterLogs.json @@ -0,0 +1,7 @@ +{ + "jsonrpc": "2.0", + "id": 9, + "result": { + "subscriptionId": "sub-logs-1" + } +} diff --git a/tests/fixtures/tui-v2/protocol/response.subscribeClusterTimeline.json b/tests/fixtures/tui-v2/protocol/response.subscribeClusterTimeline.json new file mode 100644 index 00000000..5dbdb505 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/response.subscribeClusterTimeline.json @@ -0,0 +1,7 @@ +{ + "jsonrpc": "2.0", + "id": 10, + "result": { + "subscriptionId": "sub-timeline-1" + } +} diff --git a/tests/fixtures/tui-v2/protocol/response.unsubscribe.json b/tests/fixtures/tui-v2/protocol/response.unsubscribe.json new file mode 100644 index 00000000..8d42ff14 --- /dev/null +++ b/tests/fixtures/tui-v2/protocol/response.unsubscribe.json @@ -0,0 +1,7 @@ +{ + "jsonrpc": "2.0", + "id": 21, + "result": { + "removed": true + } +} diff --git a/tests/integration/context-metrics.test.js b/tests/integration/context-metrics.test.js new file mode 100644 index 00000000..febe161e --- /dev/null +++ b/tests/integration/context-metrics.test.js @@ -0,0 +1,133 @@ +const assert = require('assert'); +const fs = require('fs'); +const os = require('os'); +const path = require('path'); + +const Ledger = require('../../src/ledger'); +const MessageBus = require('../../src/message-bus'); +const { buildContext } = require('../../src/agent/agent-context-builder'); + +const MAX_CONTEXT_CHARS = 500000; + +describe('Context Metrics Integration', function () { + let tempDir; + let ledger; + let messageBus; + let originalMetricsEnv; + let originalLedgerEnv; + + beforeEach(() => { + tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'zeroshot-context-metrics-')); + const dbPath = path.join(tempDir, 'test-ledger.db'); + ledger = new Ledger(dbPath); + messageBus = new MessageBus(ledger); + + originalMetricsEnv = process.env.ZEROSHOT_CONTEXT_METRICS; + originalLedgerEnv = process.env.ZEROSHOT_CONTEXT_METRICS_LEDGER; + process.env.ZEROSHOT_CONTEXT_METRICS = '0'; + process.env.ZEROSHOT_CONTEXT_METRICS_LEDGER = '1'; + }); + + afterEach(() => { + if (tempDir && fs.existsSync(tempDir)) { + fs.rmSync(tempDir, { recursive: true, force: true }); + } + + if (originalMetricsEnv === undefined) { + delete process.env.ZEROSHOT_CONTEXT_METRICS; + } else { + process.env.ZEROSHOT_CONTEXT_METRICS = originalMetricsEnv; + } + + if (originalLedgerEnv === undefined) { + delete process.env.ZEROSHOT_CONTEXT_METRICS_LEDGER; + } else { + process.env.ZEROSHOT_CONTEXT_METRICS_LEDGER = originalLedgerEnv; + } + }); + + it('publishes metrics to the ledger and records truncation stages', function () { + const clusterId = 'cluster-metrics-1'; + const createdAt = Date.now() - 60000; + const secretMarker = 'SECRET_CONTEXT_PAYLOAD'; + const hugeText = `${secretMarker}${'x'.repeat(MAX_CONTEXT_CHARS + 200000)}`; + + messageBus.publish({ + cluster_id: clusterId, + topic: 'ISSUE_OPENED', + sender: 'user', + content: { text: hugeText }, + }); + + messageBus.publish({ + cluster_id: clusterId, + topic: 'HUGE_TOPIC', + sender: 'tester', + content: { text: 'optional-context' }, + }); + + const context = buildContext({ + id: 'worker', + role: 'implementation', + iteration: 1, + config: { + contextStrategy: { + sources: [ + { topic: 'ISSUE_OPENED', since: 'cluster_start', limit: 1 }, + { topic: 'HUGE_TOPIC', since: 'cluster_start', limit: 1 }, + ], + maxTokens: 1, + }, + }, + messageBus, + cluster: { id: clusterId, createdAt }, + triggeringMessage: { + topic: 'TASK_READY', + sender: 'planner', + content: { text: 'go' }, + }, + }); + + assert.ok(context.length > 0, 'Context should be generated'); + + const metricsMessages = ledger.query({ cluster_id: clusterId, topic: 'CONTEXT_METRICS' }); + assert.strictEqual(metricsMessages.length, 1, 'Should publish one CONTEXT_METRICS message'); + + const metrics = metricsMessages[0].content.data; + assert.strictEqual(metrics.clusterId, clusterId); + assert.strictEqual(metrics.agentId, 'worker'); + assert.strictEqual(metrics.role, 'implementation'); + assert.strictEqual(metrics.iteration, 1); + assert.strictEqual(metrics.strategy.maxTokens, 1); + assert.strictEqual(metrics.strategy.sourcesCount, 2); + + assert.strictEqual(metrics.truncation.maxContextChars.applied, true); + assert.ok( + metrics.truncation.maxContextChars.beforeChars > + metrics.truncation.maxContextChars.afterChars, + 'Max context truncation should reduce size' + ); + assert.ok( + metrics.truncation.maxContextChars.afterChars <= MAX_CONTEXT_CHARS, + 'Final context should respect max char limit' + ); + assert.strictEqual(metrics.budget.maxTokens, 1); + assert.strictEqual(metrics.total.chars, context.length); + + const metricsJson = JSON.stringify(metrics); + assert.ok(!metricsJson.includes(secretMarker), 'Metrics should not include raw context'); + assert.ok(metrics.sections.sources.chars > 0, 'Sources section should be counted'); + assert.ok( + metrics.packs.some( + (pack) => pack.id.startsWith('source:ISSUE_OPENED') && pack.status === 'included' + ), + 'Required issue pack should be included' + ); + assert.ok( + metrics.packs.some( + (pack) => pack.id.startsWith('source:HUGE_TOPIC') && pack.status === 'skipped' + ), + 'Optional pack should be skipped when over budget' + ); + }); +}); diff --git a/tests/integration/detached-stop.test.js b/tests/integration/detached-stop.test.js new file mode 100644 index 00000000..a57ab3ae --- /dev/null +++ b/tests/integration/detached-stop.test.js @@ -0,0 +1,135 @@ +const assert = require('node:assert'); +const fs = require('node:fs'); +const path = require('node:path'); +const os = require('node:os'); +const { spawn } = require('node:child_process'); + +const Orchestrator = require('../../src/orchestrator'); + +function sleep(ms) { + return new Promise((resolve) => setTimeout(resolve, ms)); +} + +async function waitForClusterRecord(storageDir, clusterId, expectedPid, timeoutMs = 10000) { + const clustersFile = path.join(storageDir, 'clusters.json'); + const deadline = Date.now() + timeoutMs; + + while (Date.now() < deadline) { + if (fs.existsSync(clustersFile)) { + try { + const data = JSON.parse(fs.readFileSync(clustersFile, 'utf8')); + const record = data[clusterId]; + if (record && record.state === 'running' && record.pid) { + if (expectedPid) { + assert.strictEqual( + record.pid, + expectedPid, + `Expected cluster pid ${expectedPid}, got ${record.pid}` + ); + } + return record; + } + } catch { + // Ignore transient parse errors while file is being written. + } + } + await sleep(100); + } + + throw new Error(`Timed out waiting for cluster ${clusterId} in ${clustersFile}`); +} + +function waitForChildExit(child, timeoutMs = 10000) { + if (child.exitCode !== null) { + return child.exitCode; + } + + return new Promise((resolve, reject) => { + const timer = setTimeout(() => { + reject(new Error(`Timed out waiting for child process ${child.pid} to exit`)); + }, timeoutMs); + + child.once('exit', (code) => { + clearTimeout(timer); + resolve(code); + }); + + child.once('error', (error) => { + clearTimeout(timer); + reject(error); + }); + }); +} + +describe('Detached daemon stop', function () { + this.timeout(20000); + + let tempDir; + let child; + + afterEach(function () { + if (child && child.exitCode === null) { + try { + process.kill(child.pid, 'SIGKILL'); + } catch { + // Ignore cleanup errors + } + } + + if (tempDir && fs.existsSync(tempDir)) { + try { + fs.rmSync(tempDir, { recursive: true, force: true }); + } catch { + // Ignore cleanup errors + } + } + }); + + it('signals remote daemon pid and halts ledger activity', async function () { + tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'zs-detached-stop-')); + const clusterId = `detached-stop-${Date.now()}`; + const fixturePath = path.join(__dirname, '..', 'fixtures', 'detached-daemon.js'); + + child = spawn(process.execPath, [fixturePath], { + env: { + ...process.env, + ZEROSHOT_TEST_STORAGE: tempDir, + ZEROSHOT_TEST_CLUSTER_ID: clusterId, + }, + stdio: ['ignore', 'pipe', 'pipe'], + }); + + await waitForClusterRecord(tempDir, clusterId, child.pid); + await sleep(200); + + const orchestrator = await Orchestrator.create({ + quiet: true, + storageDir: tempDir, + }); + + const cluster = orchestrator.getCluster(clusterId); + assert(cluster, 'Cluster should be loaded from storage'); + + const beforeCount = cluster.messageBus.count({ cluster_id: clusterId }); + + await orchestrator.stop(clusterId); + await waitForChildExit(child, 10000); + + const afterStopCount = cluster.messageBus.count({ cluster_id: clusterId }); + await sleep(250); + const afterWaitCount = cluster.messageBus.count({ cluster_id: clusterId }); + + assert.strictEqual( + afterStopCount, + afterWaitCount, + 'No new ledger messages should appear after stop' + ); + + const status = orchestrator.getStatus(clusterId); + assert.strictEqual(status.state, 'stopped'); + assert.strictEqual(status.pid, null); + assert.strictEqual(beforeCount, afterStopCount); + + orchestrator.close(); + }); +}); diff --git a/tests/integration/guidance-queue-context.test.js b/tests/integration/guidance-queue-context.test.js new file mode 100644 index 00000000..35e62876 --- /dev/null +++ b/tests/integration/guidance-queue-context.test.js @@ -0,0 +1,83 @@ +const assert = require('assert'); + +const AgentWrapper = require('../../src/agent-wrapper'); +const MessageBus = require('../../src/message-bus'); +const Ledger = require('../../src/ledger'); +const MockTaskRunner = require('../helpers/mock-task-runner'); +const { USER_GUIDANCE_AGENT } = require('../../src/guidance-topics'); + +describe('Guidance queue integration', function () { + it('injects queued guidance into next prompt only', async function () { + const ledger = new Ledger(':memory:'); + const messageBus = new MessageBus(ledger); + const mockRunner = new MockTaskRunner(); + const clusterId = 'guidance-queue-integration'; + const clusterCreatedAt = Date.now() - 5000; + + const cluster = { + id: clusterId, + createdAt: clusterCreatedAt, + agents: [], + }; + + const workerConfig = { + id: 'worker', + role: 'implementation', + modelLevel: 'level2', + timeout: 0, + maxIterations: 5, + triggers: [{ topic: 'ISSUE_OPENED', action: 'execute_task' }], + contextStrategy: { + sources: [{ topic: 'ISSUE_OPENED', since: 'cluster_start', limit: 1 }], + }, + }; + + const worker = new AgentWrapper(workerConfig, messageBus, cluster, { + testMode: true, + taskRunner: mockRunner, + }); + cluster.agents.push(worker); + + const trigger = { + cluster_id: clusterId, + topic: 'ISSUE_OPENED', + sender: 'tester', + content: { text: 'Implement feature X' }, + }; + + mockRunner.when('worker').returns({ ok: true }); + + worker.start(); + + // First run: no guidance queued + await worker._executeTask(trigger); + mockRunner.assertCalled('worker', 1); + const firstContext = mockRunner.calls[0].context; + assert(!firstContext.includes('## Guidance (Queued)'), 'no guidance block in first run'); + + // Queue guidance after first execution + messageBus.publish({ + cluster_id: clusterId, + topic: USER_GUIDANCE_AGENT, + sender: 'user', + target_agent_id: 'worker', + content: { text: 'Use approach B' }, + timestamp: Date.now() + 10, + }); + + await worker._executeTask(trigger); + mockRunner.assertCalled('worker', 2); + const secondContext = mockRunner.calls[1].context; + assert(secondContext.includes('## Guidance (Queued)'), 'guidance block appears on next run'); + assert(secondContext.includes('Use approach B'), 'guidance text is included'); + + // Third run: no new guidance + await worker._executeTask(trigger); + mockRunner.assertCalled('worker', 3); + const thirdContext = mockRunner.calls[2].context; + assert(!thirdContext.includes('## Guidance (Queued)'), 'guidance block not repeated'); + + await worker.stop(); + ledger.close(); + }); +}); diff --git a/tests/integration/orchestrator-flow.test.js b/tests/integration/orchestrator-flow.test.js index b8e56e9d..a9a26394 100644 --- a/tests/integration/orchestrator-flow.test.js +++ b/tests/integration/orchestrator-flow.test.js @@ -1,3 +1,6 @@ +// Skip gh CLI verification in integration tests (we mock the task runner, not the CLI) +process.env.ZEROSHOT_SKIP_GH_VERIFY = '1'; + /** * Integration tests for complete cluster lifecycle * @@ -334,7 +337,12 @@ function definePrModeFlowTests() { it('should stop after git-pusher completes in autoPr mode', async () => { mockRunner.when('worker').returns({ summary: 'No changes', result: 'noop' }); mockRunner.when('validator').returns({ approved: true }); - mockRunner.when('git-pusher').returns({ summary: 'PR done', result: 'Merged' }); + mockRunner.when('git-pusher').returns({ + summary: 'PR done', + result: 'Merged', + pr_number: 12345, + pr_url: 'https://github.com/test/test/pull/12345', + }); createOrchestrator(); @@ -576,6 +584,7 @@ function defineErrorHandlingTests() { role: 'implementation', timeout: 0, maxIterations: 2, + maxRetries: 1, // Fast failure for error handling test (default is now 3) triggers: [{ topic: 'ISSUE_OPENED', action: 'execute_task' }], hooks: { onError: { diff --git a/tests/integration/orchestrator-worktree.test.js b/tests/integration/orchestrator-worktree.test.js index 2d43971d..99a17db8 100644 --- a/tests/integration/orchestrator-worktree.test.js +++ b/tests/integration/orchestrator-worktree.test.js @@ -52,6 +52,26 @@ const simpleConfig = { ], }; +async function rmDirWithRetries(target, attempts = 5) { + if (!target || !fs.existsSync(target)) { + return; + } + + for (let attempt = 0; attempt < attempts; attempt += 1) { + try { + fs.rmSync(target, { recursive: true, force: true }); + return; + } catch (error) { + const retriable = + error && (error.code === 'ENOTEMPTY' || error.code === 'EBUSY' || error.code === 'EPERM'); + if (!retriable || attempt === attempts - 1) { + throw error; + } + await new Promise((resolve) => setTimeout(resolve, 50)); + } + } +} + function registerWorktreeHooks() { beforeEach(function () { tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'zs-worktree-test-')); @@ -81,16 +101,14 @@ function registerWorktreeHooks() { } } - if (tempDir && fs.existsSync(tempDir)) { - fs.rmSync(tempDir, { recursive: true, force: true }); - } + await rmDirWithRetries(tempDir); if (testRepoDir && fs.existsSync(testRepoDir)) { try { execSync('git worktree prune', { cwd: testRepoDir, stdio: 'pipe' }); } catch { // Ignore } - fs.rmSync(testRepoDir, { recursive: true, force: true }); + await rmDirWithRetries(testRepoDir); } }); } @@ -128,7 +146,7 @@ function registerWorktreePathTest() { `Worktree path should exist: ${cluster.worktree.path}` ); - const expectedRoot = fs.realpathSync(path.join(os.tmpdir(), 'zeroshot-worktrees')); + const expectedRoot = fs.realpathSync(path.join(os.homedir(), '.zeroshot', 'worktrees')); const worktreePath = fs.realpathSync(cluster.worktree.path); assert( worktreePath.startsWith(expectedRoot + path.sep), diff --git a/tests/integration/trigger-evaluation.test.js b/tests/integration/trigger-evaluation.test.js index a6e7c3dc..4bf9f15f 100644 --- a/tests/integration/trigger-evaluation.test.js +++ b/tests/integration/trigger-evaluation.test.js @@ -13,6 +13,7 @@ const os = require('os'); const LogicEngine = require('../../src/logic-engine'); const MessageBus = require('../../src/message-bus'); const Ledger = require('../../src/ledger'); +const { SHARED_TRIGGER_SCRIPT } = require('../../src/agents/git-pusher-template'); let tempDir; let ledger; @@ -61,6 +62,7 @@ describe('Trigger Evaluation Integration', function () { defineErrorHandlingTests(); defineScriptValidationTests(); defineComplexConsensusTests(); + defineGitPusherTriggerTests(); }); function defineBasicLedgerQueryTests() { @@ -476,3 +478,122 @@ function defineComplexConsensusTests() { }); }); } + +function defineGitPusherTriggerTests() { + describe('Git-pusher Trigger Evidence', () => { + it('should allow approvals with CANNOT_VALIDATE and empty output evidence', () => { + messageBus.publish({ + cluster_id: cluster.id, + topic: 'IMPLEMENTATION_READY', + sender: 'worker', + timestamp: Date.now(), + }); + + const implTime = Date.now(); + + messageBus.publish({ + cluster_id: cluster.id, + topic: 'VALIDATION_RESULT', + sender: 'validator-1', + timestamp: implTime + 100, + content: { + data: { + approved: true, + criteriaResults: [ + { + id: 'AC1', + status: 'PASS', + evidence: { command: 'npm test', exitCode: 0, output: '' }, + }, + { + id: 'AC2', + status: 'CANNOT_VALIDATE', + reason: 'Docker not available', + }, + ], + }, + }, + }); + + messageBus.publish({ + cluster_id: cluster.id, + topic: 'VALIDATION_RESULT', + sender: 'validator-2', + timestamp: implTime + 200, + content: { data: { approved: true } }, + }); + + const result = logicEngine.evaluate( + SHARED_TRIGGER_SCRIPT, + { id: 'git-pusher', cluster_id: cluster.id }, + { topic: 'VALIDATION_RESULT' } + ); + + assert.strictEqual(result, true); + }); + + it('should accept consensus-only VALIDATION_RESULT when validators do not publish directly', () => { + // Simulate staged validation (quick/heavy): validators publish stage-specific topics, + // and only a coordinator publishes a single consolidated VALIDATION_RESULT. + cluster.agents.push( + { id: 'validator-3', role: 'validator' }, + { id: 'validator-4', role: 'validator' } + ); + + const implTime = Date.now(); + messageBus.publish({ + cluster_id: cluster.id, + topic: 'IMPLEMENTATION_READY', + sender: 'worker', + timestamp: implTime, + }); + + messageBus.publish({ + cluster_id: cluster.id, + topic: 'VALIDATION_RESULT', + sender: 'consensus-coordinator', + timestamp: implTime + 100, + content: { data: { approved: true, stage: 'heavy' } }, + }); + + const result = logicEngine.evaluate( + SHARED_TRIGGER_SCRIPT, + { id: 'git-pusher', cluster_id: cluster.id }, + { topic: 'VALIDATION_RESULT' } + ); + + assert.strictEqual(result, true); + }); + + it('should not accept consensus-only VALIDATION_RESULT when rejected', () => { + cluster.agents.push( + { id: 'validator-3', role: 'validator' }, + { id: 'validator-4', role: 'validator' } + ); + + const implTime = Date.now(); + messageBus.publish({ + cluster_id: cluster.id, + topic: 'IMPLEMENTATION_READY', + sender: 'worker', + timestamp: implTime, + }); + + messageBus.publish({ + cluster_id: cluster.id, + topic: 'VALIDATION_RESULT', + sender: 'consensus-coordinator', + timestamp: implTime + 100, + content: { data: { approved: false, stage: 'quick' } }, + }); + + const result = logicEngine.evaluate( + SHARED_TRIGGER_SCRIPT, + { id: 'git-pusher', cluster_id: cluster.id }, + { topic: 'VALIDATION_RESULT' } + ); + + assert.strictEqual(result, false); + }); + }); +} diff --git a/tests/integration/worktree-isolation.test.js b/tests/integration/worktree-isolation.test.js index de95d4ee..e6281f89 100644 --- a/tests/integration/worktree-isolation.test.js +++ b/tests/integration/worktree-isolation.test.js @@ -2,7 +2,7 @@ * Test: Worktree Isolation - Lightweight git-based isolation * * Tests the worktree isolation mode that provides: - * - Git worktree creation at {os.tmpdir()}/zeroshot-worktrees/{clusterId} + * - Git worktree creation at ~/.zeroshot/worktrees/{clusterId} * - Separate branch (zeroshot/{clusterId}) without copying files * - Fast setup (<1s vs 30-60s for Docker) * - No Docker dependency @@ -73,7 +73,7 @@ function registerWorktreePathTest() { const info = manager.createWorktreeIsolation(testClusterId, testRepoDir); assert(info.path, 'Should return worktree path'); - const expectedRoot = fs.realpathSync(path.join(os.tmpdir(), 'zeroshot-worktrees')); + const expectedRoot = fs.realpathSync(path.join(os.homedir(), '.zeroshot', 'worktrees')); const worktreePath = fs.realpathSync(info.path); assert( worktreePath.startsWith(expectedRoot + path.sep), diff --git a/tests/max-model.test.js b/tests/max-model.test.js index 956453a5..0d7c90c9 100644 --- a/tests/max-model.test.js +++ b/tests/max-model.test.js @@ -219,7 +219,7 @@ function registerDynamicModelRulesTests() { function registerDefaultModelTests() { describe('Default model when unspecified', function () { it('should use maxModel ceiling when it constrains default', function () { - saveTestSettings({ maxModel: 'haiku' }); + saveTestSettings({ maxModel: 'haiku', defaultProvider: 'claude' }); const agentConfig = { id: 'default-model-agent', timeout: 0 }; @@ -256,7 +256,7 @@ function registerDefaultModelTests() { }); it('should use provider default level even when maxModel allows higher', function () { - saveTestSettings({ maxModel: 'opus' }); + saveTestSettings({ maxModel: 'opus', defaultProvider: 'claude' }); const agentConfig = { id: 'premium-default-agent', timeout: 0 }; diff --git a/tests/message-buffering-while-busy.test.js b/tests/message-buffering-while-busy.test.js new file mode 100644 index 00000000..25788c4d --- /dev/null +++ b/tests/message-buffering-while-busy.test.js @@ -0,0 +1,153 @@ +/** + * Regression test: never drop trigger-matching messages while an agent is busy. + * + * BUG: + * Agents dropped trigger-matching messages whenever state !== 'idle'. + * In real clusters this can drop VALIDATION_RESULT / QUICK_VALIDATION_RESULT + * while the worker or coordinator is executing a task, wedging the cluster. + * + * FIX: + * Buffer trigger-matching messages while busy and drain once idle. + */ + +const { expect } = require('chai'); +const crypto = require('crypto'); +const fs = require('fs'); +const os = require('os'); +const path = require('path'); + +const Orchestrator = require('../src/orchestrator'); +const MockTaskRunner = require('./helpers/mock-task-runner'); + +describe('Agent message buffering while busy', function () { + this.timeout(15000); + + let orchestrator; + let mockRunner; + let testDir; + let clusterId; + + beforeEach(() => { + testDir = path.join( + os.tmpdir(), + `zeroshot-buffering-test-${crypto.randomBytes(8).toString('hex')}` + ); + fs.mkdirSync(testDir, { recursive: true }); + + const settingsPath = path.join(testDir, 'settings.json'); + fs.writeFileSync( + settingsPath, + JSON.stringify( + { + firstRunComplete: true, + defaultProvider: 'claude', + autoCheckUpdates: false, + }, + null, + 2 + ) + ); + + mockRunner = new MockTaskRunner(); + orchestrator = new Orchestrator({ + quiet: true, + storageDir: testDir, + skipLoad: true, + taskRunner: mockRunner, + }); + }); + + afterEach(async () => { + if (clusterId) { + try { + await orchestrator.kill(clusterId); + } catch { + // Cluster may already be stopped + } + } + + try { + orchestrator.close(); + } catch { + /* ignore */ + } + + if (testDir && fs.existsSync(testDir)) { + fs.rmSync(testDir, { recursive: true, force: true }); + } + }); + + it('executes worker twice when VALIDATION_RESULT arrives during an in-flight task', async () => { + mockRunner.when('worker').delays(250, 'done'); + + const config = { + agents: [ + { + id: 'worker', + role: 'implementation', + modelLevel: 'level2', + prompt: 'do work', + triggers: [ + { topic: 'ISSUE_OPENED', action: 'execute_task' }, + { topic: 'VALIDATION_RESULT', action: 'execute_task' }, + ], + hooks: { + onComplete: { + action: 'publish_message', + config: { topic: 'WORK_DONE' }, + }, + }, + }, + { + id: 'completion-detector', + role: 'completion-detector', + modelLevel: 'level2', + prompt: 'stop after 2 work cycles', + triggers: [ + { + topic: 'WORK_DONE', + action: 'stop_cluster', + logic: { script: "return ledger.count({ topic: 'WORK_DONE' }) >= 2;" }, + }, + ], + }, + ], + completion_detector: { + type: 'topic', + config: { topic: 'CLUSTER_COMPLETE' }, + }, + }; + + const result = await orchestrator.start(config, { text: 'test' }, { cwd: process.cwd() }); + clusterId = result.id; + + const cluster = orchestrator.getCluster(clusterId); + expect(cluster).to.exist; + + // Publish a trigger-matching message while the worker is still busy with the first task. + await new Promise((resolve) => setTimeout(resolve, 50)); + cluster.messageBus.publish({ + cluster_id: clusterId, + topic: 'VALIDATION_RESULT', + sender: 'consensus-coordinator', + receiver: 'broadcast', + timestamp: Date.now(), + content: { text: 'stage 1 rejected' }, + }); + + // Wait for the cluster to stop (completion-detector publishes CLUSTER_COMPLETE after 2 WORK_DONE). + const start = Date.now(); + while (Date.now() - start < 10000) { + const current = orchestrator.getCluster(clusterId); + if (current.state === 'stopped') { + break; + } + await new Promise((resolve) => setTimeout(resolve, 50)); + } + + const finalCluster = orchestrator.getCluster(clusterId); + expect(finalCluster.state).to.equal('stopped'); + + mockRunner.assertCalled('worker', 2); + }); +}); diff --git a/tests/model-selection.test.js b/tests/model-selection.test.js index fdd0c78e..1cf6098d 100644 --- a/tests/model-selection.test.js +++ b/tests/model-selection.test.js @@ -26,6 +26,7 @@ function registerModelSelectionHooks() { } const testSettings = { maxModel: 'opus', + defaultProvider: 'claude', defaultConfig: 'conductor-bootstrap', defaultDocker: false, strictSchema: true, diff --git a/tests/nested-cluster.test.js b/tests/nested-cluster.test.js index d4f0edbd..aee0c053 100644 --- a/tests/nested-cluster.test.js +++ b/tests/nested-cluster.test.js @@ -295,6 +295,61 @@ function defineSubClusterWrapperTests() { assert.strictEqual(wrapper.role, 'orchestrator'); assert.strictEqual(wrapper.state, 'idle'); }); + + it('should select latest parent topic messages for child context', function () { + const dbPath = path.join(TEST_STORAGE, 'parent-context.db'); + const ledger = new Ledger(dbPath); + const messageBus = new MessageBus(ledger); + + const config = { + id: 'test-subcluster', + type: 'subcluster', + role: 'orchestrator', + config: { + agents: [ + { + id: 'worker', + role: 'implementation', + modelLevel: 'level1', + triggers: [{ topic: 'START' }], + }, + ], + }, + triggers: [{ topic: 'BEGIN' }], + contextStrategy: { + parentTopics: [{ topic: 'PLAN_READY', strategy: 'latest', amount: 1 }], + }, + }; + + const parentCluster = { id: 'parent-cluster' }; + const wrapper = new SubClusterWrapper(config, messageBus, parentCluster, { quiet: true }); + + const baseTime = Date.now(); + messageBus.publish({ + cluster_id: parentCluster.id, + topic: 'PLAN_READY', + sender: 'planner', + content: { text: 'Old plan' }, + timestamp: baseTime, + }); + messageBus.publish({ + cluster_id: parentCluster.id, + topic: 'PLAN_READY', + sender: 'planner', + content: { text: 'Newest plan' }, + timestamp: baseTime + 10, + }); + + const context = wrapper._buildChildContext({ + topic: 'BEGIN', + sender: 'system', + timestamp: baseTime + 20, + content: { text: 'Trigger' }, + }); + + assert(context.includes('Newest plan'), 'Should include latest parent topic message'); + assert(!context.includes('Old plan'), 'Should not include older parent topic message'); + }); }); } diff --git a/tests/openai-output-parser.test.js b/tests/openai-output-parser.test.js index bb508ef8..4e7ed1b7 100644 --- a/tests/openai-output-parser.test.js +++ b/tests/openai-output-parser.test.js @@ -58,6 +58,11 @@ describe('Codex output parser', () => { assert.strictEqual(result, null); }); + it('ignores item.started events', () => { + const result = parseEvent('{"type":"item.started","item_id":"item_123"}'); + assert.strictEqual(result, null); + }); + it('parses item.created events', () => { const line = JSON.stringify({ type: 'item.created', diff --git a/tests/orchestrator-completion-detector.test.js b/tests/orchestrator-completion-detector.test.js new file mode 100644 index 00000000..c933afaf --- /dev/null +++ b/tests/orchestrator-completion-detector.test.js @@ -0,0 +1,87 @@ +const assert = require('assert'); +const fs = require('fs'); +const os = require('os'); +const path = require('path'); +const Orchestrator = require('../src/orchestrator'); + +describe('Orchestrator completion-detector injection', function () { + const originalSettingsFile = process.env.ZEROSHOT_SETTINGS_FILE; + let tempDir; + + afterEach(function () { + if (tempDir && fs.existsSync(tempDir)) { + fs.rmSync(tempDir, { recursive: true, force: true }); + tempDir = null; + } + + if (originalSettingsFile) { + process.env.ZEROSHOT_SETTINGS_FILE = originalSettingsFile; + } else { + delete process.env.ZEROSHOT_SETTINGS_FILE; + } + }); + + function writeSettings(settings) { + tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'zeroshot-completion-detector-')); + const settingsFile = path.join(tempDir, 'settings.json'); + fs.writeFileSync(settingsFile, JSON.stringify(settings), 'utf8'); + process.env.ZEROSHOT_SETTINGS_FILE = settingsFile; + } + + it('uses modelLevel derived from claude minModel bounds', async function () { + writeSettings({ + maxModel: 'opus', + minModel: 'sonnet', + }); + + const orchestrator = new Orchestrator({ quiet: true, skipLoad: true }); + let injectedAgent = null; + orchestrator._opAddAgents = (_cluster, operation) => { + injectedAgent = operation.agents[0]; + }; + + await orchestrator._injectCompletionAgent( + { + agents: [], + config: { defaultProvider: 'claude' }, + autoPr: false, + }, + {} + ); + + assert.ok(injectedAgent, 'Expected completion-detector to be injected'); + assert.strictEqual(injectedAgent.id, 'completion-detector'); + assert.strictEqual(injectedAgent.modelLevel, 'level2'); + assert.strictEqual(injectedAgent.model, undefined); + }); + + it('uses provider-specific minLevel when forcing non-claude provider', async function () { + writeSettings({ + defaultProvider: 'codex', + providerSettings: { + codex: { + minLevel: 'level3', + }, + }, + }); + + const orchestrator = new Orchestrator({ quiet: true, skipLoad: true }); + let injectedAgent = null; + orchestrator._opAddAgents = (_cluster, operation) => { + injectedAgent = operation.agents[0]; + }; + + await orchestrator._injectCompletionAgent( + { + agents: [], + config: { forceProvider: 'codex' }, + autoPr: false, + }, + {} + ); + + assert.ok(injectedAgent, 'Expected completion-detector to be injected'); + assert.strictEqual(injectedAgent.modelLevel, 'level3'); + assert.strictEqual(injectedAgent.model, undefined); + }); +}); diff --git a/tests/orchestrator.test.js b/tests/orchestrator.test.js index f48de86f..0905a9c9 100644 --- a/tests/orchestrator.test.js +++ b/tests/orchestrator.test.js @@ -245,6 +245,50 @@ function defineLifecycleStartTests() { assert.strictEqual(callCount, 2, 'Expected SIGTERM failure to trigger one retry'); }); + // eslint-disable-next-line sonarjs/no-skipped-tests -- AGENT_RESTART_ATTEMPT feature not yet implemented (see AGENTS.md) + it.skip('should restart implementation agent after retries exhausted', async function () { + const config = { + agents: [ + { + id: 'worker', + role: 'implementation', + modelLevel: 'level2', + outputFormat: 'text', + triggers: [{ topic: 'ISSUE_OPENED', action: 'execute_task' }], + hooks: { + onComplete: { + action: 'publish_message', + config: { + topic: 'CLUSTER_COMPLETE', + content: { + data: { reason: 'restart-after-exhausted-retries-test' }, + }, + }, + }, + }, + }, + ], + }; + + let callCount = 0; + lifecycleMockRunner.when('worker').calls(() => { + callCount += 1; + if (callCount <= 3) { + return { success: false, output: '', error: 'Request timed out' }; + } + return { success: true, output: 'ok' }; + }); + + const result = await lifecycleOrchestrator.start(config, { text: 'Fix bug' }); + await waitForClusterState(lifecycleOrchestrator, result.id, 'stopped', 10000); + + const cluster = lifecycleOrchestrator.getCluster(result.id); + const ledger = new LedgerAssertions(cluster.ledger, result.id); + ledger.assertCount('AGENT_RESTART_ATTEMPT', 1); + + assert.ok(callCount >= 4, `Expected worker to be invoked at least 4 times, got ${callCount}`); + }); + it('should inject worktree cwd when worktree enabled', function () { // This test requires a real git repo - skip in test environment // The functionality is tested in integration/worktree tests @@ -290,6 +334,16 @@ function defineLifecycleStartTests() { /issue.*or text/i, 'Should reject missing input' ); + + // Regression: startup failures must be persisted for supervisor visibility (no "invisible" clusters). + const clustersFile = path.join(lifecycleStorageDir, 'clusters.json'); + assert.ok(fs.existsSync(clustersFile), 'clusters.json should exist after failed start'); + const persisted = JSON.parse(fs.readFileSync(clustersFile, 'utf8')); + const ids = Object.keys(persisted); + assert.equal(ids.length, 1, 'Expected exactly one persisted cluster entry'); + const c = persisted[ids[0]]; + assert.equal(c.state, 'failed', 'Failed start should persist state=failed'); + assert.equal(c.pid, null, 'Failed start should persist pid=null'); }); it('should auto-generate unique cluster IDs', async function () { diff --git a/tests/output-extraction.test.js b/tests/output-extraction.test.js index 93b1ae74..75401cf3 100644 --- a/tests/output-extraction.test.js +++ b/tests/output-extraction.test.js @@ -9,6 +9,7 @@ const assert = require('assert'); const { extractJsonFromOutput, + extractCliError, extractFromResultWrapper, extractFromTextEvents, extractFromMarkdown, @@ -22,6 +23,7 @@ describe('Output Extraction Module', function () { defineTextEventExtractionTests(); defineMarkdownExtractionTests(); defineDirectJsonExtractionTests(); + defineCliErrorExtractionTests(); defineFullPipelineTests(); defineRegressionTests(); }); @@ -274,6 +276,168 @@ function defineDirectJsonExtractionTests() { assert.strictEqual(extractDirectJson(' '), null); assert.strictEqual(extractDirectJson(null), null); }); + + // CLI metadata rejection tests - prevent schema validation against wrong structure + it('should reject type:result objects (CLI wrapper)', function () { + const text = '{"type":"result","subtype":"success","duration_ms":1234}'; + const result = extractDirectJson(text); + assert.strictEqual(result, null); + }); + + it('should reject CLI metadata with duration_ms and session_id', function () { + const text = + '{"duration_ms":5000,"session_id":"abc123","total_cost_usd":0.05,"usage":{"input_tokens":100}}'; + const result = extractDirectJson(text); + assert.strictEqual(result, null); + }); + + it('should reject CLI metadata with multiple metadata fields', function () { + const text = + '{"type":"result","subtype":"error","is_error":true,"duration_ms":123,"num_turns":5,"total_cost_usd":0.1,"permission_denials":[],"errors":["some error"]}'; + const result = extractDirectJson(text); + assert.strictEqual(result, null); + }); + + it('should accept normal agent output (not CLI metadata)', function () { + const text = '{"summary":"Task completed","completionStatus":{"canValidate":true}}'; + const result = extractDirectJson(text); + assert.deepStrictEqual(result, { + summary: 'Task completed', + completionStatus: { canValidate: true }, + }); + }); + + it('should accept agent output that has one CLI-like field by coincidence', function () { + // If agent happens to output a field named "errors", that's fine (< 2 CLI fields) + const text = '{"summary":"Fixed bugs","errors":[]}'; + const result = extractDirectJson(text); + assert.deepStrictEqual(result, { summary: 'Fixed bugs', errors: [] }); + }); + }); +} + +function defineCliErrorExtractionTests() { + // ============================================================================ + // CLI ERROR EXTRACTION (ALL PROVIDERS) + // ============================================================================ + describe('extractCliError', function () { + // Claude errors + it('should extract Claude error with is_error:true', function () { + const output = '{"type":"result","is_error":true,"errors":["Permission denied for tool X"]}'; + const result = extractCliError(output); + assert.deepStrictEqual(result, { + error: 'Permission denied for tool X', + provider: 'claude', + }); + }); + + it('should extract Claude error with multiple errors', function () { + const output = '{"type":"result","is_error":true,"errors":["Error 1","Error 2"]}'; + const result = extractCliError(output); + assert.deepStrictEqual(result, { + error: 'Error 1; Error 2', + provider: 'claude', + }); + }); + + it('should extract Claude error with subtype:error', function () { + const output = '{"type":"result","subtype":"error","error":"Token limit exceeded"}'; + const result = extractCliError(output); + assert.deepStrictEqual(result, { + error: 'Token limit exceeded', + provider: 'claude', + }); + }); + + // Codex errors + it('should extract Codex turn.failed error', function () { + const output = '{"type":"turn.failed","error":{"message":"API rate limit exceeded"}}'; + const result = extractCliError(output); + assert.deepStrictEqual(result, { + error: 'API rate limit exceeded', + provider: 'codex', + }); + }); + + it('should extract Codex turn.failed with string error', function () { + const output = '{"type":"turn.failed","error":"Something went wrong"}'; + const result = extractCliError(output); + assert.deepStrictEqual(result, { + error: 'Something went wrong', + provider: 'codex', + }); + }); + + // Gemini errors + it('should extract Gemini error with success:false', function () { + const output = '{"type":"result","success":false,"error":"Model unavailable"}'; + const result = extractCliError(output); + assert.deepStrictEqual(result, { + error: 'Model unavailable', + provider: 'gemini', + }); + }); + + // Opencode errors + it('should extract Opencode session.error', function () { + const output = '{"type":"session.error","error":{"message":"Connection timeout"}}'; + const result = extractCliError(output); + assert.deepStrictEqual(result, { + error: 'Connection timeout', + provider: 'opencode', + }); + }); + + it('should extract Opencode session.error with nested data', function () { + const output = + '{"type":"session.error","error":{"data":{"message":"Auth failed"},"name":"AuthError"}}'; + const result = extractCliError(output); + assert.deepStrictEqual(result, { + error: 'Auth failed', + provider: 'opencode', + }); + }); + + // No error cases + it('should return null for successful Claude output', function () { + const output = '{"type":"result","subtype":"success","result":{"foo":"bar"}}'; + const result = extractCliError(output); + assert.strictEqual(result, null); + }); + + it('should return null for successful Codex output', function () { + const output = '{"type":"turn.completed","usage":{"input_tokens":100}}'; + const result = extractCliError(output); + assert.strictEqual(result, null); + }); + + it('should return null for successful Gemini output', function () { + const output = '{"type":"result","success":true}'; + const result = extractCliError(output); + assert.strictEqual(result, null); + }); + + it('should return null for empty output', function () { + assert.strictEqual(extractCliError(''), null); + assert.strictEqual(extractCliError(null), null); + }); + + it('should return null for non-error JSON', function () { + const output = '{"foo":"bar","baz":123}'; + const result = extractCliError(output); + assert.strictEqual(result, null); + }); + + it('should find error in multi-line NDJSON output', function () { + const output = `{"type":"system","subtype":"init"} +{"type":"assistant","message":{}} +{"type":"result","is_error":true,"errors":["Task failed"]}`; + const result = extractCliError(output); + assert.deepStrictEqual(result, { + error: 'Task failed', + provider: 'claude', + }); + }); }); } @@ -493,5 +657,15 @@ Done.`; assert.strictEqual(result.complexity, 'TRIVIAL'); assert.strictEqual(result.taskType, 'TASK'); }); + + it('REGRESSION: Claude CLI error result without actual output', function () { + // When Claude has an error, it returns CLI metadata without result field. + // This MUST return null (not the CLI metadata object), so schema validation + // doesn't run against wrong structure ({duration_ms, session_id} vs {summary, completionStatus}) + const output = + '{"type":"result","subtype":"error","is_error":true,"duration_ms":1234,"duration_api_ms":1200,"num_turns":0,"session_id":"abc123","total_cost_usd":0.0,"usage":{},"modelUsage":null,"permission_denials":[],"uuid":"xyz","errors":["Permission denied"]}'; + const result = extractJsonFromOutput(output, 'claude'); + assert.strictEqual(result, null); + }); }); } diff --git a/tests/prompt-selection.test.js b/tests/prompt-selection.test.js index 32bde32c..e7f0b1e1 100644 --- a/tests/prompt-selection.test.js +++ b/tests/prompt-selection.test.js @@ -29,12 +29,10 @@ describe('Prompt Selection - Static (backward compat)', function () { }); it('should return null if no prompt configured', function () { - const agent = new AgentWrapper( - { id: 'test', timeout: 0 }, - mockMessageBus, - mockCluster, - { testMode: true, mockSpawnFn: () => {} } - ); + const agent = new AgentWrapper({ id: 'test', timeout: 0 }, mockMessageBus, mockCluster, { + testMode: true, + mockSpawnFn: () => {}, + }); assert.strictEqual(agent._selectPrompt(), null); }); diff --git a/tests/settings-providers.test.js b/tests/settings-providers.test.js index cdeb0cad..99762811 100644 --- a/tests/settings-providers.test.js +++ b/tests/settings-providers.test.js @@ -4,6 +4,7 @@ const path = require('path'); const os = require('os'); const { loadSettings, validateSetting } = require('../lib/settings'); const { validateProviderSettings, validateProviderLevel } = require('../src/config-validator'); +const { getProvider } = require('../src/providers'); describe('Provider settings', function () { const testDir = path.join(os.tmpdir(), `zeroshot-provider-settings-${Date.now()}`); @@ -83,4 +84,16 @@ describe('Provider settings', function () { assert.strictEqual(settings.providerSettings.claude.maxLevel, 'level1'); assert.strictEqual(settings.providerSettings.claude.defaultLevel, 'level1'); }); + + it('uses gpt-5.3-codex as the default codex model', function () { + const codex = getProvider('codex'); + const modelSpec = codex.resolveModelSpec(codex.getDefaultLevel(), {}); + assert.strictEqual(modelSpec.model, 'gpt-5.3-codex'); + }); + + it('maps claude level3 to opus-4.6', function () { + const claude = getProvider('claude'); + const modelSpec = claude.resolveModelSpec('level3', {}); + assert.strictEqual(modelSpec.model, 'opus-4.6'); + }); }); diff --git a/tests/state-snapshot.test.js b/tests/state-snapshot.test.js new file mode 100644 index 00000000..4f4fc0f2 --- /dev/null +++ b/tests/state-snapshot.test.js @@ -0,0 +1,381 @@ +const assert = require('assert'); +const fs = require('fs'); +const os = require('os'); +const path = require('path'); + +const Orchestrator = require('../src/orchestrator'); +const Ledger = require('../src/ledger'); +const MessageBus = require('../src/message-bus'); +const StateSnapshotter = require('../src/state-snapshotter'); +const MockTaskRunner = require('./helpers/mock-task-runner'); +const { + initStateFromIssue, + applyPlanReady, + applyWorkerProgress, + applyValidationResult, + applyInvestigationComplete, +} = require('../src/state-snapshot'); + +function createTempDir() { + return fs.mkdtempSync(path.join(os.tmpdir(), 'zeroshot-snapshot-')); +} + +function cleanupTempDir(dir) { + if (dir && fs.existsSync(dir)) { + fs.rmSync(dir, { recursive: true, force: true }); + } +} + +function buildMessage({ clusterId, topic, sender, content }) { + return { + id: `msg_${Math.random().toString(16).slice(2)}`, + timestamp: Date.now(), + cluster_id: clusterId, + topic, + sender, + receiver: 'broadcast', + content, + }; +} + +async function waitForClusterState(orchestrator, clusterId, target, timeoutMs = 5000) { + const start = Date.now(); + while (Date.now() - start < timeoutMs) { + try { + const status = orchestrator.getStatus(clusterId); + if (status.state === target) { + return; + } + } catch { + // Cluster may be removed during shutdown + } + await new Promise((resolve) => setTimeout(resolve, 50)); + } + throw new Error(`Cluster ${clusterId} did not reach ${target} within ${timeoutMs}ms`); +} + +describe('State snapshot builder', () => { + it('should replace plan text and criteria on PLAN_READY', () => { + const clusterId = 'cluster-plan'; + const issueMessage = buildMessage({ + clusterId, + topic: 'ISSUE_OPENED', + sender: 'system', + content: { text: 'Fix login bug', data: { title: 'Login bug', issue_number: 133 } }, + }); + + let state = initStateFromIssue(issueMessage); + + state = applyPlanReady( + state, + buildMessage({ + clusterId, + topic: 'PLAN_READY', + sender: 'planner', + content: { + text: 'Plan v1', + data: { + summary: 'Old plan', + acceptanceCriteria: [{ id: 'AC1', criterion: 'Old criteria', priority: 'MUST' }], + filesAffected: ['old.js'], + }, + }, + }) + ); + + state = applyPlanReady( + state, + buildMessage({ + clusterId, + topic: 'PLAN_READY', + sender: 'planner', + content: { + text: 'Plan v2', + data: { + summary: 'New plan', + acceptanceCriteria: [{ id: 'AC1', criterion: 'New criteria', priority: 'MUST' }], + filesAffected: ['new.js'], + }, + }, + }) + ); + + assert.strictEqual(state.plan.text, 'Plan v2'); + assert.deepStrictEqual(state.plan.acceptanceCriteria, ['AC1 (MUST): New criteria']); + assert.deepStrictEqual(state.plan.filesAffected, ['new.js']); + }); + + it('should update progress fields from WORKER_PROGRESS', () => { + const clusterId = 'cluster-progress'; + const issueMessage = buildMessage({ + clusterId, + topic: 'ISSUE_OPENED', + sender: 'system', + content: { text: 'Add endpoint', data: { title: 'Add endpoint' } }, + }); + + let state = initStateFromIssue(issueMessage); + state = applyWorkerProgress( + state, + buildMessage({ + clusterId, + topic: 'WORKER_PROGRESS', + sender: 'worker', + content: { + text: 'WIP', + data: { + completionStatus: { + canValidate: false, + percentComplete: 45, + nextSteps: ['Write tests', 'Run lint'], + }, + }, + }, + }) + ); + + assert.strictEqual(state.progress.percentComplete, 45); + assert.deepStrictEqual(state.progress.nextSteps, ['Write tests', 'Run lint']); + }); + + it('should update validation approval and errors', () => { + const clusterId = 'cluster-validation'; + const issueMessage = buildMessage({ + clusterId, + topic: 'ISSUE_OPENED', + sender: 'system', + content: { text: 'Fix build', data: { title: 'Build fix' } }, + }); + + let state = initStateFromIssue(issueMessage); + state = applyValidationResult( + state, + buildMessage({ + clusterId, + topic: 'VALIDATION_RESULT', + sender: 'validator', + content: { + data: { + approved: false, + errors: ['Missing tests', 'Type error'], + }, + }, + }) + ); + + assert.strictEqual(state.validation.approved, false); + assert.deepStrictEqual(state.validation.errors, ['Missing tests', 'Type error']); + }); + + it('should update debug fields from INVESTIGATION_COMPLETE', () => { + const clusterId = 'cluster-debug'; + const issueMessage = buildMessage({ + clusterId, + topic: 'ISSUE_OPENED', + sender: 'system', + content: { text: 'Debug failure', data: { title: 'Debug issue' } }, + }); + + let state = initStateFromIssue(issueMessage); + state = applyInvestigationComplete( + state, + buildMessage({ + clusterId, + topic: 'INVESTIGATION_COMPLETE', + sender: 'investigator', + content: { + text: 'Apply guard clauses and add tests', + data: { + successCriteria: 'All tests pass', + rootCauses: [{ cause: 'Null input not handled' }], + }, + }, + }) + ); + + assert.strictEqual(state.debug.fixPlan, 'Apply guard clauses and add tests'); + assert.deepStrictEqual(state.debug.rootCauses, ['Null input not handled']); + }); +}); + +describe('StateSnapshotter publishing', () => { + let tempDir; + let ledger; + let messageBus; + + beforeEach(() => { + tempDir = createTempDir(); + ledger = new Ledger(path.join(tempDir, 'ledger.db')); + messageBus = new MessageBus(ledger); + }); + + afterEach(() => { + ledger.close(); + cleanupTempDir(tempDir); + }); + + it('should publish STATE_SNAPSHOT on PLAN_READY and VALIDATION_RESULT', () => { + const clusterId = 'cluster-publish'; + const snapshotter = new StateSnapshotter({ messageBus, clusterId }); + snapshotter.start(); + + messageBus.publish({ + cluster_id: clusterId, + topic: 'ISSUE_OPENED', + sender: 'system', + content: { text: 'Issue text', data: { title: 'Issue title' } }, + metadata: { source: 'text' }, + }); + + messageBus.publish({ + cluster_id: clusterId, + topic: 'PLAN_READY', + sender: 'planner', + content: { + text: 'Plan text', + data: { + summary: 'Plan summary', + acceptanceCriteria: [{ id: 'AC1', criterion: 'Do thing', priority: 'MUST' }], + }, + }, + }); + + const planSnapshot = messageBus.findLast({ cluster_id: clusterId, topic: 'STATE_SNAPSHOT' }); + assert.ok(planSnapshot, 'STATE_SNAPSHOT should be published'); + assert.strictEqual(planSnapshot.content.data.plan.text, 'Plan text'); + + messageBus.publish({ + cluster_id: clusterId, + topic: 'VALIDATION_RESULT', + sender: 'validator', + content: { + data: { + approved: false, + errors: ['Test failure'], + }, + }, + }); + + const validationSnapshot = messageBus.findLast({ + cluster_id: clusterId, + topic: 'STATE_SNAPSHOT', + }); + assert.strictEqual(validationSnapshot.content.data.validation.approved, false); + assert.deepStrictEqual(validationSnapshot.content.data.validation.errors, ['Test failure']); + }); +}); + +describe('Snapshotter orchestration integration', () => { + let tempDir; + let orchestrator; + let mockRunner; + + beforeEach(() => { + tempDir = createTempDir(); + mockRunner = new MockTaskRunner(); + }); + + afterEach(() => { + if (orchestrator) { + orchestrator.close(); + } + cleanupTempDir(tempDir); + }); + + it('should inject STATE_SNAPSHOT into worker context', async () => { + mockRunner.when('worker').returns('{"summary":"done"}'); + + orchestrator = new Orchestrator({ + quiet: true, + storageDir: tempDir, + taskRunner: mockRunner, + }); + + const config = { + agents: [ + { + id: 'worker', + role: 'implementation', + timeout: 0, + contextStrategy: { + sources: [ + { topic: 'STATE_SNAPSHOT', priority: 'required', strategy: 'latest', amount: 1 }, + { topic: 'ISSUE_OPENED', priority: 'required', strategy: 'latest', amount: 1 }, + ], + }, + triggers: [{ topic: 'ISSUE_OPENED', action: 'execute_task' }], + prompt: 'Do work', + hooks: { + onComplete: { + action: 'publish_message', + config: { topic: 'CLUSTER_COMPLETE', content: { text: 'done' } }, + }, + }, + }, + { + id: 'completion-detector', + role: 'orchestrator', + timeout: 0, + triggers: [{ topic: 'CLUSTER_COMPLETE', action: 'stop_cluster' }], + }, + ], + }; + + const result = await orchestrator.start(config, { text: 'Do the thing' }); + await waitForClusterState(orchestrator, result.id, 'stopped', 5000); + + mockRunner.assertContextIncludes('worker', 'STATE_SNAPSHOT'); + }); + + it('should publish STATE_SNAPSHOT when loading legacy clusters', async () => { + const clusterId = 'legacy-cluster'; + const clusterDir = tempDir; + const dbPath = path.join(clusterDir, `${clusterId}.db`); + + const ledger = new Ledger(dbPath); + ledger.append({ + cluster_id: clusterId, + topic: 'ISSUE_OPENED', + sender: 'system', + content: { text: 'Legacy issue', data: { title: 'Legacy' } }, + metadata: { source: 'text' }, + }); + ledger.append({ + cluster_id: clusterId, + topic: 'PLAN_READY', + sender: 'planner', + content: { text: 'Legacy plan', data: { summary: 'Legacy summary' } }, + }); + ledger.close(); + + const fixturePath = path.join(__dirname, 'fixtures', 'single-worker.json'); + const config = JSON.parse(fs.readFileSync(fixturePath, 'utf8')); + const clustersFile = path.join(clusterDir, 'clusters.json'); + fs.writeFileSync( + clustersFile, + JSON.stringify( + { + [clusterId]: { + id: clusterId, + config, + state: 'stopped', + createdAt: Date.now(), + pid: null, + }, + }, + null, + 2 + ) + ); + + orchestrator = await Orchestrator.create({ storageDir: clusterDir, quiet: true }); + const cluster = orchestrator.getCluster(clusterId); + const snapshot = cluster.messageBus.findLast({ + cluster_id: clusterId, + topic: 'STATE_SNAPSHOT', + }); + + assert.ok(snapshot, 'STATE_SNAPSHOT should be created during load'); + assert.strictEqual(snapshot.content.data.plan.text, 'Legacy plan'); + }); +}); diff --git a/tests/structuredOutput-mapping.test.js b/tests/structuredOutput-mapping.test.js new file mode 100644 index 00000000..1439b58d --- /dev/null +++ b/tests/structuredOutput-mapping.test.js @@ -0,0 +1,108 @@ +/** + * Regression test: structuredOutput ā†’ jsonSchema mapping + * + * ROOT CAUSE (discovered 2026-02-08): + * git-pusher-template.js uses `structuredOutput` as the config key, + * but agent-config.js only recognizes `jsonSchema`. + * The structuredOutput key was silently ignored ā†’ default schema applied ā†’ + * agent never told to output pr_number ā†’ verify_github_pr hook fails with + * "VERIFICATION FAILED: git-pusher must provide pr_number in structured output" + * + * The PR was actually created and merged, but the hook couldn't extract + * pr_number because the CLI was given the wrong schema. + */ + +const assert = require('assert'); +const { generateGitPusherAgent } = require('../src/agents/git-pusher-template'); +const { validateAgentConfig } = require('../src/agent/agent-config'); + +describe('structuredOutput ā†’ jsonSchema mapping', function () { + it('should use structuredOutput as jsonSchema when both are not set', function () { + // SETUP: Generate git-pusher config (uses structuredOutput key) + const agentConfig = generateGitPusherAgent('github'); + + // VERIFY: structuredOutput is defined in the raw config + assert.ok(agentConfig.structuredOutput, 'git-pusher template must define structuredOutput'); + assert.ok( + agentConfig.structuredOutput.properties.pr_number, + 'structuredOutput must have pr_number property' + ); + + // ACTION: Pass through validateAgentConfig (this is where mapping should happen) + const normalized = validateAgentConfig({ ...agentConfig }); + + // ASSERTION: jsonSchema must be the structuredOutput schema, NOT the default + assert.ok(normalized.jsonSchema, 'jsonSchema must be set after validation'); + assert.ok( + normalized.jsonSchema.properties.pr_number, + 'jsonSchema must contain pr_number from structuredOutput (not default summary/result schema)' + ); + assert.strictEqual( + normalized.jsonSchema.properties.pr_number.type, + 'number', + 'pr_number must be type number' + ); + + // Verify it does NOT have the default schema fields + assert.strictEqual( + normalized.jsonSchema.properties.summary, + undefined, + 'jsonSchema must NOT have default "summary" field when structuredOutput is provided' + ); + }); + + it('should preserve explicit jsonSchema over structuredOutput', function () { + // If someone sets BOTH jsonSchema and structuredOutput, jsonSchema wins + const agentConfig = generateGitPusherAgent('github'); + const customSchema = { + type: 'object', + properties: { + custom_field: { type: 'string' }, + }, + required: ['custom_field'], + }; + + const normalized = validateAgentConfig({ + ...agentConfig, + jsonSchema: customSchema, + }); + + assert.strictEqual( + normalized.jsonSchema.properties.custom_field.type, + 'string', + 'explicit jsonSchema must take precedence over structuredOutput' + ); + }); + + it('should apply default schema when neither jsonSchema nor structuredOutput is set', function () { + const normalized = validateAgentConfig({ + id: 'test-agent', + role: 'test', + triggers: [], + prompt: 'test prompt', + }); + + assert.ok( + normalized.jsonSchema.properties.summary, + 'default schema must have summary when no schema provided' + ); + assert.ok( + normalized.jsonSchema.properties.result, + 'default schema must have result when no schema provided' + ); + }); + + it('should work for all git-pusher platforms', function () { + const platforms = ['github', 'gitlab', 'azure-devops']; + + for (const platform of platforms) { + const agentConfig = generateGitPusherAgent(platform); + const normalized = validateAgentConfig({ ...agentConfig }); + + assert.ok( + normalized.jsonSchema.properties.pr_number || normalized.jsonSchema.properties.mr_number, + `${platform}: jsonSchema must contain pr_number or mr_number from structuredOutput` + ); + } + }); +}); diff --git a/tests/template-resolver.test.js b/tests/template-resolver.test.js index 4c3fdb74..85df7891 100644 --- a/tests/template-resolver.test.js +++ b/tests/template-resolver.test.js @@ -32,7 +32,7 @@ function getConfig(complexity, taskType) { if (complexity === 'TRIVIAL') return 0; if (complexity === 'SIMPLE') return 1; if (complexity === 'STANDARD') return 2; - if (complexity === 'CRITICAL') return 4; + if (complexity === 'CRITICAL') return 0; // Two-stage validation with dynamic loading return 1; }; @@ -134,7 +134,7 @@ describe('TemplateResolver', function () { planner_level: 'level3', worker_level: 'level2', validator_level: 'level2', - validator_count: 4, + validator_count: 0, // CRITICAL uses two-stage validation (validators loaded dynamically) }); assert.ok(resolved.agents); @@ -142,9 +142,13 @@ describe('TemplateResolver', function () { const planner = resolved.agents.find((a) => a.id === 'planner'); assert.strictEqual(planner.modelLevel, 'level3'); - // Should have 5 validators for CRITICAL + // CRITICAL tasks use two-stage validation: meta-coordinator loads validators dynamically + // So inline validators should be filtered out, meta-coordinator should be present const validators = resolved.agents.filter((a) => a.role === 'validator'); - assert.strictEqual(validators.length, 5); + assert.strictEqual(validators.length, 0); + + const metaCoordinator = resolved.agents.find((a) => a.id === 'meta-coordinator'); + assert.ok(metaCoordinator, 'meta-coordinator should be present for CRITICAL tasks'); }); it('should fail on missing required params', function () { @@ -206,11 +210,11 @@ describe('2D Classification Routing', function () { assert.strictEqual(params.planner_level, 'level2'); }); - it('CRITICAL should use level3 planner and 4 validators', function () { + it('CRITICAL should use level3 planner and two-stage validation', function () { const { base, params } = getConfig('CRITICAL', 'TASK'); assert.strictEqual(base, 'full-workflow'); assert.strictEqual(params.planner_level, 'level3'); - assert.strictEqual(params.validator_count, 4); + assert.strictEqual(params.validator_count, 0); // Two-stage validation with dynamic loading assert.strictEqual(params.max_iterations, DEFAULT_MAX_ITERATIONS); }); diff --git a/tests/transform-sandbox-ledger.test.js b/tests/transform-sandbox-ledger.test.js new file mode 100644 index 00000000..a2bdb6fc --- /dev/null +++ b/tests/transform-sandbox-ledger.test.js @@ -0,0 +1,296 @@ +/** + * Regression test for transform sandbox ledger API + * + * ROOT CAUSE (2026-02-03): + * buildTransformSandbox() did NOT provide `ledger` API to transform scripts, + * but template transforms (e.g., heavy-validation.json:247) used `ledger.query()`. + * Result: "ledger is not defined" ā†’ hook failed ā†’ cluster deadlocked. + * + * FIX: Added ledger, cluster, and helpers APIs to buildTransformSandbox(), + * mirroring logic-engine.js _buildContext(). + */ + +const { expect } = require('chai'); +const vm = require('vm'); + +// Mock message bus +function createMockMessageBus() { + const messages = []; + return { + publish: (msg) => messages.push({ ...msg, timestamp: Date.now() }), + query: ({ topic, cluster_id, since }) => { + return messages.filter( + (m) => + m.cluster_id === cluster_id && + (!topic || m.topic === topic) && + (!since || m.timestamp > since) + ); + }, + findLast: ({ topic, cluster_id }) => { + const matching = messages.filter( + (m) => m.cluster_id === cluster_id && (!topic || m.topic === topic) + ); + return matching[matching.length - 1] || null; + }, + count: ({ topic, cluster_id, since }) => { + return messages.filter( + (m) => + m.cluster_id === cluster_id && + (!topic || m.topic === topic) && + (!since || m.timestamp > since) + ).length; + }, + since: ({ cluster_id, timestamp }) => { + return messages.filter((m) => m.cluster_id === cluster_id && m.timestamp > timestamp); + }, + _messages: messages, + }; +} + +// Mock agent +function createMockAgent(messageBus, clusterId = 'test-cluster') { + return { + id: 'test-agent', + cluster_id: clusterId, + messageBus, + cluster: { + id: clusterId, + agents: [ + { id: 'validator-1', role: 'validator' }, + { id: 'validator-2', role: 'validator' }, + { id: 'worker', role: 'implementation' }, + ], + }, + _log: () => {}, + }; +} + +describe('Transform sandbox ledger API', () => { + let messageBus; + let agent; + + beforeEach(() => { + messageBus = createMockMessageBus(); + agent = createMockAgent(messageBus); + }); + + /** + * REGRESSION TEST: Transform script can access ledger.query() + * + * This is the exact pattern that failed in heavy-validation.json:247 + */ + it('transform script can use ledger.query()', () => { + // Seed some messages + messageBus.publish({ + cluster_id: 'test-cluster', + topic: 'HEAVY_VALIDATION_RESULT', + sender: 'validator-1', + content: { data: { errors: ['error1'] } }, + }); + messageBus.publish({ + cluster_id: 'test-cluster', + topic: 'HEAVY_VALIDATION_RESULT', + sender: 'validator-2', + content: { data: { errors: ['error2'] } }, + }); + + // Build sandbox like agent-hook-executor.js does + const sandbox = buildTestSandbox(agent, { allApproved: true, summary: 'All good' }); + + // This is the EXACT script from heavy-validation.json:247 that was failing + const script = ` + return { + topic: 'VALIDATION_RESULT', + content: { + text: result.allApproved ? 'All validations passed' : 'Stage 2 rejected', + data: { + approved: result.allApproved, + stage: 'heavy', + summary: result.summary, + errors: ledger.query({ topic: 'HEAVY_VALIDATION_RESULT' }) + .flatMap(r => r.content?.data?.errors || []) + } + } + }; + `; + + const vmContext = vm.createContext(sandbox); + const wrappedScript = `(function() { ${script} })()`; + const result = vm.runInContext(wrappedScript, vmContext); + + expect(result.topic).to.equal('VALIDATION_RESULT'); + expect(result.content.data.approved).to.equal(true); + expect(result.content.data.errors).to.deep.equal(['error1', 'error2']); + }); + + /** + * REGRESSION TEST: Transform script can access ledger.findLast() + */ + it('transform script can use ledger.findLast()', () => { + messageBus.publish({ + cluster_id: 'test-cluster', + topic: 'IMPLEMENTATION_READY', + sender: 'worker', + content: { text: 'Done' }, + }); + + const sandbox = buildTestSandbox(agent, {}); + + const script = ` + const lastPush = ledger.findLast({ topic: 'IMPLEMENTATION_READY' }); + return { + topic: 'TEST_RESULT', + content: { found: !!lastPush, sender: lastPush?.sender } + }; + `; + + const vmContext = vm.createContext(sandbox); + const result = vm.runInContext(`(function() { ${script} })()`, vmContext); + + expect(result.content.found).to.equal(true); + expect(result.content.sender).to.equal('worker'); + }); + + /** + * REGRESSION TEST: Transform script can access cluster.getAgentsByRole() + * + * Pattern from full-workflow.json:255 + */ + it('transform script can use cluster.getAgentsByRole()', () => { + const sandbox = buildTestSandbox(agent, {}); + + const script = ` + const validators = cluster.getAgentsByRole('validator'); + return { + topic: 'TEST_RESULT', + content: { validatorCount: validators.length, ids: validators.map(v => v.id) } + }; + `; + + const vmContext = vm.createContext(sandbox); + const result = vm.runInContext(`(function() { ${script} })()`, vmContext); + + expect(result.content.validatorCount).to.equal(2); + expect(result.content.ids).to.deep.equal(['validator-1', 'validator-2']); + }); + + /** + * REGRESSION TEST: Transform script can use helpers.allResponded() + */ + it('transform script can use helpers.allResponded()', () => { + // Both validators responded + messageBus.publish({ + cluster_id: 'test-cluster', + topic: 'VALIDATION_RESULT', + sender: 'validator-1', + content: { data: { approved: true } }, + }); + messageBus.publish({ + cluster_id: 'test-cluster', + topic: 'VALIDATION_RESULT', + sender: 'validator-2', + content: { data: { approved: true } }, + }); + + const sandbox = buildTestSandbox(agent, {}); + + const script = ` + const validators = cluster.getAgentsByRole('validator'); + const allDone = helpers.allResponded(validators, 'VALIDATION_RESULT', 0); + return { + topic: 'TEST_RESULT', + content: { allResponded: allDone } + }; + `; + + const vmContext = vm.createContext(sandbox); + const result = vm.runInContext(`(function() { ${script} })()`, vmContext); + + expect(result.content.allResponded).to.equal(true); + }); + + /** + * REGRESSION TEST: Sandbox provides Set for validators pattern + */ + it('transform script can use Set builtin', () => { + const sandbox = buildTestSandbox(agent, {}); + + const script = ` + const ids = new Set(['a', 'b', 'a']); + return { + topic: 'TEST_RESULT', + content: { size: ids.size } + }; + `; + + const vmContext = vm.createContext(sandbox); + const result = vm.runInContext(`(function() { ${script} })()`, vmContext); + + expect(result.content.size).to.equal(2); + }); +}); + +/** + * Build sandbox matching agent-hook-executor.js buildTransformSandbox() + * This is a copy for testing - the real one is in agent-hook-executor.js + */ +function buildTestSandbox(agent, resultData) { + const clusterId = agent.cluster_id; + const messageBus = agent.messageBus; + const cluster = agent.cluster; + + const ledgerAPI = { + query: (criteria) => { + return messageBus.query({ ...criteria, cluster_id: clusterId }); + }, + findLast: (criteria) => { + return messageBus.findLast({ ...criteria, cluster_id: clusterId }); + }, + count: (criteria) => { + return messageBus.count({ ...criteria, cluster_id: clusterId }); + }, + since: (timestamp) => { + return messageBus.since({ cluster_id: clusterId, timestamp }); + }, + }; + + const clusterAPI = { + id: clusterId, + getAgents: () => (cluster ? cluster.agents || [] : []), + getAgentsByRole: (role) => + cluster ? (cluster.agents || []).filter((a) => a.role === role) : [], + getAgent: (id) => (cluster ? (cluster.agents || []).find((a) => a.id === id) : null), + }; + + const helpers = { + getConfig: () => ({}), + allResponded: (agents, topic, since) => { + const responses = ledgerAPI.query({ topic, since }); + const responders = new Set(responses.map((r) => r.sender)); + return agents.every((a) => responders.has(a.id || a)); + }, + hasConsensus: (topic, since) => { + const responses = ledgerAPI.query({ topic, since }); + if (responses.length === 0) return false; + return responses.every((r) => r.content?.data?.approved === true); + }, + }; + + return { + result: resultData, + triggeringMessage: null, + ledger: ledgerAPI, + cluster: clusterAPI, + helpers, + JSON, + Set, + Map, + Array, + Object, + console: { + log: () => {}, + error: () => {}, + warn: () => {}, + }, + }; +} diff --git a/tests/tui-layout.test.js b/tests/tui-layout.test.js deleted file mode 100644 index debc85b9..00000000 --- a/tests/tui-layout.test.js +++ /dev/null @@ -1,319 +0,0 @@ -/** - * Test suite for TUI layout - * Verifies layout creation and widget updates work correctly - */ - -const { expect } = require('chai'); -const blessed = require('blessed'); -const { - createLayout, - updateClustersTable, - updateAgentsTable, - updateStatsBox, - addLogEntry, - clearLogs, -} = require('../src/tui/layout'); - -let screen; -let layout; - -describe('TUI Layout', () => { - beforeEach(() => { - // Create a mock screen for testing - screen = blessed.screen({ mouse: true, title: 'Test Dashboard' }); - }); - - afterEach(() => { - if (screen) { - screen.destroy(); - } - }); - - defineCreateLayoutTests(); - defineUpdateClustersTableTests(); - defineUpdateAgentsTableTests(); - defineUpdateStatsBoxTests(); - defineAddLogEntryTests(); - defineClearLogsTests(); - defineFocusNavigationTests(); -}); - -function defineCreateLayoutTests() { - describe('createLayout', () => { - it('should create layout with all widgets', () => { - layout = createLayout(screen); - - expect(layout).to.exist; - expect(layout.screen).to.equal(screen); - expect(layout.grid).to.exist; - expect(layout.clustersTable).to.exist; - expect(layout.agentTable).to.exist; - expect(layout.statsBox).to.exist; - expect(layout.logsBox).to.exist; - expect(layout.helpBar).to.exist; - }); - - it('should have widgets array with 3 items', () => { - layout = createLayout(screen); - - expect(layout.widgets).to.be.an('array'); - expect(layout.widgets).to.have.lengthOf(3); - }); - - it('should initialize clusters table with empty data', () => { - layout = createLayout(screen); - - // Table should have headers and empty data initially - expect(layout.clustersTable).to.exist; - }); - - it('should provide focus control methods', () => { - layout = createLayout(screen); - - expect(layout.focus).to.be.a('function'); - expect(layout.getCurrentFocus).to.be.a('function'); - }); - - it('should set focus to clusters table initially', () => { - layout = createLayout(screen); - - expect(layout.getCurrentFocus()).to.equal(0); - }); - }); -} - -function defineUpdateClustersTableTests() { - describe('updateClustersTable', () => { - beforeEach(() => { - layout = createLayout(screen); - }); - - it('should update table with cluster data', () => { - const clusters = [ - { - id: 'cluster-1', - status: 'running', - agentCount: 5, - config: 'default', - uptime: '2h 30m', - }, - { - id: 'cluster-2', - status: 'stopped', - agentCount: 0, - config: 'simple', - uptime: '0s', - }, - ]; - - updateClustersTable(layout.clustersTable, clusters); - - // Verify the method completes without error - expect(layout.clustersTable).to.exist; - }); - - it('should handle empty cluster array', () => { - updateClustersTable(layout.clustersTable, []); - - expect(layout.clustersTable).to.exist; - }); - - it('should handle clusters with missing properties', () => { - const clusters = [ - { - id: 'cluster-1', - // missing other properties - }, - ]; - - updateClustersTable(layout.clustersTable, clusters); - - expect(layout.clustersTable).to.exist; - }); - }); -} - -function defineUpdateAgentsTableTests() { - describe('updateAgentsTable', () => { - beforeEach(() => { - layout = createLayout(screen); - }); - - it('should update table with agent data', () => { - const agents = [ - { - clusterId: 'cluster-1', - id: 'worker-1', - role: 'worker', - status: 'running', - iteration: 3, - cpu: '12.5%', - memory: '245 MB', - }, - { - clusterId: 'cluster-1', - id: 'validator-1', - role: 'validator', - status: 'idle', - iteration: 0, - cpu: '0.1%', - memory: '128 MB', - }, - ]; - - updateAgentsTable(layout.agentTable, agents); - - expect(layout.agentTable).to.exist; - }); - - it('should handle empty agent array', () => { - updateAgentsTable(layout.agentTable, []); - - expect(layout.agentTable).to.exist; - }); - - it('should handle agents with missing properties', () => { - const agents = [ - { - id: 'agent-1', - // missing other properties - }, - ]; - - updateAgentsTable(layout.agentTable, agents); - - expect(layout.agentTable).to.exist; - }); - }); -} - -function defineUpdateStatsBoxTests() { - describe('updateStatsBox', () => { - beforeEach(() => { - layout = createLayout(screen); - }); - - it('should update stats box with system metrics', () => { - const stats = { - activeClusters: 3, - totalAgents: 15, - usedMemory: '512 MB', - totalMemory: '2 GB', - totalCPU: '25.5%', - }; - - updateStatsBox(layout.statsBox, stats); - - expect(layout.statsBox).to.exist; - }); - - it('should handle missing stats properties', () => { - const stats = { - activeClusters: 2, - // missing other properties - }; - - updateStatsBox(layout.statsBox, stats); - - expect(layout.statsBox).to.exist; - }); - - it('should handle empty stats object', () => { - updateStatsBox(layout.statsBox, {}); - - expect(layout.statsBox).to.exist; - }); - }); -} - -function defineAddLogEntryTests() { - describe('addLogEntry', () => { - beforeEach(() => { - layout = createLayout(screen); - }); - - it('should add info level log entry', () => { - addLogEntry(layout.logsBox, 'Test info message', 'info'); - - expect(layout.logsBox).to.exist; - }); - - it('should add warn level log entry', () => { - addLogEntry(layout.logsBox, 'Test warning message', 'warn'); - - expect(layout.logsBox).to.exist; - }); - - it('should add error level log entry', () => { - addLogEntry(layout.logsBox, 'Test error message', 'error'); - - expect(layout.logsBox).to.exist; - }); - - it('should add debug level log entry', () => { - addLogEntry(layout.logsBox, 'Test debug message', 'debug'); - - expect(layout.logsBox).to.exist; - }); - - it('should default to info level if not specified', () => { - addLogEntry(layout.logsBox, 'Test message without level'); - - expect(layout.logsBox).to.exist; - }); - - it('should handle unknown log level', () => { - addLogEntry(layout.logsBox, 'Test unknown level', 'unknown'); - - expect(layout.logsBox).to.exist; - }); - }); -} - -function defineClearLogsTests() { - describe('clearLogs', () => { - beforeEach(() => { - layout = createLayout(screen); - }); - - it('should clear logs without error', () => { - addLogEntry(layout.logsBox, 'Test message 1'); - addLogEntry(layout.logsBox, 'Test message 2'); - - clearLogs(layout.logsBox); - - expect(layout.logsBox).to.exist; - }); - }); -} - -function defineFocusNavigationTests() { - describe('focus navigation', () => { - beforeEach(() => { - layout = createLayout(screen); - }); - - it('should cycle focus through widgets', () => { - expect(layout.getCurrentFocus()).to.equal(0); // clusters table - - layout.focus(1); - expect(layout.getCurrentFocus()).to.equal(1); // agents table - - layout.focus(2); - expect(layout.getCurrentFocus()).to.equal(2); // logs - - layout.focus(0); - expect(layout.getCurrentFocus()).to.equal(0); // back to clusters - }); - - it('should not focus on invalid indices', () => { - const initialFocus = layout.getCurrentFocus(); - - layout.focus(-1); - expect(layout.getCurrentFocus()).to.equal(initialFocus); - - layout.focus(999); - expect(layout.getCurrentFocus()).to.equal(initialFocus); - }); - }); -} diff --git a/tests/two-stage-validation.test.js b/tests/two-stage-validation.test.js new file mode 100644 index 00000000..b37c8810 --- /dev/null +++ b/tests/two-stage-validation.test.js @@ -0,0 +1,356 @@ +/** + * Tests for two-stage validation pipeline (quick ā†’ heavy) + */ + +const assert = require('assert'); +const path = require('path'); +const TemplateResolver = require('../src/template-resolver'); +const { validateConfig } = require('../src/config-validator'); +const LogicEngine = require('../src/logic-engine'); +const MessageBus = require('../src/message-bus'); +const Ledger = require('../src/ledger'); + +describe('Two-Stage Validation Pipeline', function () { + let resolver; + + before(function () { + const templatesDir = path.join(__dirname, '..', 'cluster-templates'); + resolver = new TemplateResolver(templatesDir); + }); + + describe('quick-validation template', function () { + it('should contain validator-requirements and validator-code', function () { + const resolved = resolver.resolve('quick-validation', {}); + assert.ok(resolved.agents, 'Template should have agents'); + + const requirements = resolved.agents.find((a) => a.id === 'validator-requirements'); + assert.ok(requirements, 'validator-requirements should exist'); + + const code = resolved.agents.find((a) => a.id === 'validator-code'); + assert.ok(code, 'validator-code should exist'); + }); + + it('should trigger on IMPLEMENTATION_READY', function () { + const resolved = resolver.resolve('quick-validation', {}); + + const requirements = resolved.agents.find((a) => a.id === 'validator-requirements'); + const trigger = requirements.triggers.find((t) => t.topic === 'IMPLEMENTATION_READY'); + assert.ok(trigger, 'validator-requirements should trigger on IMPLEMENTATION_READY'); + + const code = resolved.agents.find((a) => a.id === 'validator-code'); + const codeTrigger = code.triggers.find((t) => t.topic === 'IMPLEMENTATION_READY'); + assert.ok(codeTrigger, 'validator-code should trigger on IMPLEMENTATION_READY'); + }); + + it('should have consensus-coordinator publishing QUICK_VALIDATION_PASSED on success', function () { + const resolved = resolver.resolve('quick-validation', {}); + + const coordinator = resolved.agents.find((a) => a.id === 'consensus-coordinator'); + assert.ok(coordinator, 'consensus-coordinator should exist'); + + const hook = coordinator.hooks.onComplete; + assert.ok(hook, 'consensus-coordinator should have onComplete hook'); + assert.strictEqual(hook.action, 'publish_message'); + }); + + it('should publish VALIDATION_RESULT if any validator rejects', function () { + const resolved = resolver.resolve('quick-validation', {}); + + const coordinator = resolved.agents.find((a) => a.id === 'consensus-coordinator'); + const logicScript = coordinator.triggers.find((t) => t.logic)?.logic?.script; + + assert.ok(logicScript, 'consensus-coordinator should have logic script'); + assert.ok( + logicScript.includes('VALIDATION_RESULT'), + 'Logic should publish VALIDATION_RESULT on rejection' + ); + }); + }); + + describe('heavy-validation template', function () { + it('should contain validator-security and validator-tester', function () { + const resolved = resolver.resolve('heavy-validation', {}); + assert.ok(resolved.agents, 'Template should have agents'); + + const security = resolved.agents.find((a) => a.id === 'validator-security'); + assert.ok(security, 'validator-security should exist'); + + const tester = resolved.agents.find((a) => a.id === 'validator-tester'); + assert.ok(tester, 'validator-tester should exist'); + }); + + it('should trigger on QUICK_VALIDATION_PASSED', function () { + const resolved = resolver.resolve('heavy-validation', {}); + + const security = resolved.agents.find((a) => a.id === 'validator-security'); + const trigger = security.triggers.find((t) => t.topic === 'QUICK_VALIDATION_PASSED'); + assert.ok(trigger, 'validator-security should trigger on QUICK_VALIDATION_PASSED'); + + const tester = resolved.agents.find((a) => a.id === 'validator-tester'); + const testerTrigger = tester.triggers.find((t) => t.topic === 'QUICK_VALIDATION_PASSED'); + assert.ok(testerTrigger, 'validator-tester should trigger on QUICK_VALIDATION_PASSED'); + }); + + it('should have consensus-coordinator publishing VALIDATION_RESULT', function () { + const resolved = resolver.resolve('heavy-validation', {}); + + const coordinator = resolved.agents.find((a) => a.id === 'consensus-coordinator'); + assert.ok(coordinator, 'consensus-coordinator should exist'); + + const hook = coordinator.hooks.onComplete; + assert.ok(hook, 'consensus-coordinator should have onComplete hook'); + assert.strictEqual(hook.action, 'publish_message'); + + const logicScript = coordinator.triggers.find((t) => t.logic)?.logic?.script; + assert.ok( + logicScript.includes('VALIDATION_RESULT'), + 'Logic should publish VALIDATION_RESULT' + ); + }); + + it('should have contextStrategy for QUICK_VALIDATION_PASSED', function () { + const resolved = resolver.resolve('heavy-validation', {}); + + const security = resolved.agents.find((a) => a.id === 'validator-security'); + const contextSource = security.contextStrategy?.sources?.find( + (s) => s.topic === 'QUICK_VALIDATION_PASSED' + ); + + assert.ok(contextSource, 'validator-security should have QUICK_VALIDATION_PASSED context'); + assert.strictEqual(contextSource.priority, 'required'); + }); + + it('should not retrigger consensus on a late single-validator update after heavy result is published', function () { + const resolved = resolver.resolve('heavy-validation', {}); + const coordinator = resolved.agents.find((a) => a.id === 'consensus-coordinator'); + const triggerScript = coordinator?.triggers?.find( + (t) => t.topic === 'HEAVY_VALIDATION_RESULT' + )?.logic?.script; + assert.ok(triggerScript, 'heavy consensus trigger script should exist'); + + const cluster = { + id: 'heavy-regression', + agents: [ + { id: 'validator-security', role: 'validator' }, + { id: 'validator-tester', role: 'validator' }, + { id: 'consensus-coordinator', role: 'coordinator' }, + ], + }; + + const ledger = new Ledger(':memory:'); + const messageBus = new MessageBus(ledger); + const logicEngine = new LogicEngine(messageBus, cluster); + + try { + let ts = Date.now(); + const nextTs = () => ++ts; + + messageBus.publish({ + cluster_id: cluster.id, + topic: 'QUICK_VALIDATION_PASSED', + sender: 'consensus-coordinator', + timestamp: nextTs(), + }); + + messageBus.publish({ + cluster_id: cluster.id, + topic: 'HEAVY_VALIDATION_RESULT', + sender: 'validator-security', + timestamp: nextTs(), + content: { data: { approved: true } }, + }); + + let shouldTrigger = logicEngine.evaluate( + triggerScript, + { id: 'consensus-coordinator', cluster_id: cluster.id }, + { topic: 'HEAVY_VALIDATION_RESULT' } + ); + assert.strictEqual(shouldTrigger, false, 'must wait for both validators'); + + messageBus.publish({ + cluster_id: cluster.id, + topic: 'HEAVY_VALIDATION_RESULT', + sender: 'validator-tester', + timestamp: nextTs(), + content: { data: { approved: true } }, + }); + + shouldTrigger = logicEngine.evaluate( + triggerScript, + { id: 'consensus-coordinator', cluster_id: cluster.id }, + { topic: 'HEAVY_VALIDATION_RESULT' } + ); + assert.strictEqual(shouldTrigger, true, 'should trigger once both validators respond'); + + messageBus.publish({ + cluster_id: cluster.id, + topic: 'VALIDATION_RESULT', + sender: 'consensus-coordinator', + timestamp: nextTs(), + content: { data: { approved: true, stage: 'heavy' } }, + }); + + messageBus.publish({ + cluster_id: cluster.id, + topic: 'HEAVY_VALIDATION_RESULT', + sender: 'validator-security', + timestamp: nextTs(), + content: { data: { approved: false } }, + }); + + shouldTrigger = logicEngine.evaluate( + triggerScript, + { id: 'consensus-coordinator', cluster_id: cluster.id }, + { topic: 'HEAVY_VALIDATION_RESULT' } + ); + assert.strictEqual( + shouldTrigger, + false, + 'late update from one validator must not retrigger heavy consensus' + ); + + messageBus.publish({ + cluster_id: cluster.id, + topic: 'HEAVY_VALIDATION_RESULT', + sender: 'validator-tester', + timestamp: nextTs(), + content: { data: { approved: false } }, + }); + + shouldTrigger = logicEngine.evaluate( + triggerScript, + { id: 'consensus-coordinator', cluster_id: cluster.id }, + { topic: 'HEAVY_VALIDATION_RESULT' } + ); + assert.strictEqual( + shouldTrigger, + true, + 'should trigger again only after both validators publish a fresh cycle' + ); + } finally { + ledger.close(); + } + }); + }); + + describe('full-workflow integration', function () { + it('should load meta-coordinator for CRITICAL tasks', function () { + const resolved = resolver.resolve('full-workflow', { + task_type: 'TASK', + complexity: 'CRITICAL', + max_tokens: 150000, + max_iterations: 25, + planner_level: 'level3', + worker_level: 'level2', + validator_level: 'level2', + validator_count: 0, + }); + + const metaCoordinator = resolved.agents.find((a) => a.id === 'meta-coordinator'); + assert.ok(metaCoordinator, 'meta-coordinator should be present for CRITICAL tasks'); + + // Inline validators should be filtered out + const validators = resolved.agents.filter((a) => a.role === 'validator'); + assert.strictEqual(validators.length, 0, 'No inline validators for CRITICAL tasks'); + + // Regression: config-validator should NOT raise Gap 15 role-reference errors when validators are absent. + const validation = validateConfig(resolved); + const roleErrors = validation.errors.filter( + (e) => + e.includes('[Gap 15]') || + e.includes("Logic references role 'validator'") || + e.includes('Logic references role "validator"') + ); + assert.strictEqual(roleErrors.length, 0, `Unexpected role reference errors: ${roleErrors}`); + }); + + it('meta-coordinator should republish trigger topic after load_config (prevents validator deadlock)', function () { + const resolved = resolver.resolve('full-workflow', { + task_type: 'TASK', + complexity: 'CRITICAL', + max_tokens: 150000, + max_iterations: 25, + planner_level: 'level3', + worker_level: 'level2', + validator_level: 'level2', + validator_count: 0, + }); + + const metaCoordinator = resolved.agents.find((a) => a.id === 'meta-coordinator'); + assert.ok(metaCoordinator, 'meta-coordinator should be present for CRITICAL tasks'); + + const hookTransformScript = metaCoordinator.hooks?.onComplete?.transform?.script || ''; + assert.ok( + hookTransformScript.includes("action: 'publish'") || + hookTransformScript.includes('action: "publish"'), + 'meta-coordinator should publish a republished trigger topic after load_config' + ); + assert.ok( + hookTransformScript.includes('_republished'), + 'meta-coordinator republish should include _republished metadata' + ); + + const implTrigger = metaCoordinator.triggers?.find((t) => t.topic === 'IMPLEMENTATION_READY'); + assert.ok(implTrigger?.logic?.script?.includes('_republished')); + + const stage2Trigger = metaCoordinator.triggers?.find( + (t) => t.topic === 'QUICK_VALIDATION_PASSED' + ); + assert.ok(stage2Trigger?.logic?.script?.includes('_republished')); + }); + + it('should NOT load meta-coordinator for STANDARD tasks', function () { + const resolved = resolver.resolve('full-workflow', { + task_type: 'TASK', + complexity: 'STANDARD', + max_tokens: 100000, + max_iterations: 25, + planner_level: 'level2', + worker_level: 'level2', + validator_level: 'level2', + validator_count: 2, + }); + + const metaCoordinator = resolved.agents.find((a) => a.id === 'meta-coordinator'); + assert.ok(!metaCoordinator, 'meta-coordinator should NOT be present for STANDARD tasks'); + + // Inline validators should be present + const validators = resolved.agents.filter((a) => a.role === 'validator'); + assert.strictEqual(validators.length, 2, 'STANDARD tasks should have 2 inline validators'); + }); + }); + + describe('Sequential execution order', function () { + it('Stage 2 cannot trigger without QUICK_VALIDATION_PASSED', function () { + const heavyResolved = resolver.resolve('heavy-validation', {}); + + // Heavy validators ONLY trigger on QUICK_VALIDATION_PASSED + const heavySecurity = heavyResolved.agents.find((a) => a.id === 'validator-security'); + const triggers = heavySecurity.triggers.filter((t) => t.topic !== 'QUICK_VALIDATION_PASSED'); + + assert.strictEqual( + triggers.length, + 0, + 'Heavy validators should ONLY trigger on QUICK_VALIDATION_PASSED' + ); + }); + + it('Consensus-coordinator publishes VALIDATION_RESULT, not QUICK_VALIDATION_PASSED on rejection', function () { + const resolved = resolver.resolve('quick-validation', {}); + + const coordinator = resolved.agents.find((a) => a.id === 'consensus-coordinator'); + const hookLogic = coordinator.hooks.onComplete?.logic?.script; + + assert.ok(hookLogic, 'consensus-coordinator should have hook logic script'); + assert.ok( + hookLogic.includes('!result.allApproved') || + hookLogic.includes('result.approved === false'), + 'Logic should check for rejections' + ); + assert.ok( + hookLogic.includes('VALIDATION_RESULT'), + 'Should publish VALIDATION_RESULT on rejection (skips QUICK_VALIDATION_PASSED)' + ); + }); + }); +}); diff --git a/tests/unit/.test-storage/clusters.json b/tests/unit/.test-storage/clusters.json new file mode 100644 index 00000000..7b8f330a --- /dev/null +++ b/tests/unit/.test-storage/clusters.json @@ -0,0 +1,147 @@ +{ + "emerald-prism-4": { + "id": "emerald-prism-4", + "config": { + "agents": [ + { + "id": "test-conductor", + "role": "conductor", + "modelLevel": "level1", + "outputFormat": "json", + "jsonSchema": { + "type": "object", + "properties": { + "complexity": { + "type": "string", + "enum": ["TRIVIAL", "SIMPLE", "STANDARD", "CRITICAL"] + }, + "reasoning": { + "type": "string" + } + }, + "required": ["complexity", "reasoning"] + }, + "prompt": "Classify this task: {{ISSUE_OPENED.content.text}}. Return JSON with complexity and reasoning.", + "triggers": [ + { + "topic": "ISSUE_OPENED", + "action": "execute_task" + } + ], + "hooks": { + "onComplete": { + "action": "publish_message", + "config": { + "topic": "CLASSIFICATION_DONE", + "content": { + "text": "Classification complete", + "data": { + "result": "{{result}}" + } + } + } + } + }, + "cwd": "/home/ubuntu/.zeroshot/worktrees/frozen-equinox-32", + "strictSchema": true, + "timeout": 0 + }, + { + "id": "git-pusher", + "role": "completion-detector", + "modelLevel": "level2", + "triggers": [ + { + "topic": "VALIDATION_RESULT", + "logic": { + "engine": "javascript", + "script": "const validators = cluster.getAgentsByRole('validator');\nconst lastPush = ledger.findLast({ topic: 'IMPLEMENTATION_READY' });\nif (!lastPush) return false;\nif (validators.length === 0) return true;\nconst results = ledger.query({ topic: 'VALIDATION_RESULT', since: lastPush.timestamp });\nif (results.length < validators.length) return false;\nconst allApproved = results.every(r => r.content?.data?.approved === 'true' || r.content?.data?.approved === true);\nif (!allApproved) return false;\nconst hasSufficientEvidence = results.every(r => {\n const criteria = r.content?.data?.criteriaResults;\n if (!Array.isArray(criteria) || criteria.length === 0) return true;\n return criteria.every(c => {\n const status = String(c.status || '').toUpperCase();\n if (status === 'CANNOT_VALIDATE') return true;\n if (status === 'SKIPPED') return true;\n if (status === 'CANNOT_VALIDATE_YET') return false;\n const evidence = c.evidence || {};\n const hasCommand = typeof evidence.command === 'string' && evidence.command.trim().length > 0;\n const exitCode = evidence.exitCode;\n const hasExitCode =\n typeof exitCode === 'number' ||\n (typeof exitCode === 'string' && exitCode.trim() !== '' && Number.isFinite(Number(exitCode)));\n const hasOutput = evidence.output === undefined || typeof evidence.output === 'string';\n return hasCommand && hasExitCode && hasOutput;\n });\n});\nreturn hasSufficientEvidence;" + }, + "action": "execute_task" + } + ], + "prompt": "šŸšØ CRITICAL: ALL VALIDATORS APPROVED. YOU MUST CREATE A PR AND GET IT MERGED. DO NOT STOP UNTIL THE PR IS MERGED. šŸšØ\n\n## MANDATORY STEPS - EXECUTE EACH ONE IN ORDER - DO NOT SKIP ANY STEP\n\n### STEP 1: Stage ALL changes (MANDATORY)\n```bash\ngit add -A\n```\nRun this command. Do not skip it.\n\n### STEP 2: Check what's staged\n```bash\ngit status\n```\nRun this. If nothing to commit, output JSON with pr_url: null and stop.\n\n### STEP 3: Commit the changes (MANDATORY if there are changes)\n```bash\ngit commit -m \"feat: implement #unknown - Manual Input\"\n```\nRun this command. Do not skip it.\n\n### STEP 4: Push to origin (MANDATORY)\n```bash\ngit push -u origin HEAD\n```\nRun this. If it fails, check the error and fix it.\n\nāš ļø AFTER PUSH YOU ARE NOT DONE! CONTINUE TO STEP 5! āš ļø\n\n### STEP 5: CREATE THE PR (MANDATORY - YOU MUST RUN THIS COMMAND)\n```bash\ngh pr create --base dev --title \"feat: Manual Input\" --body \"Closes #unknown\"\n```\nšŸšØ YOU MUST RUN `gh pr create`! Outputting a link is NOT creating a PR! šŸšØ\nThe push output shows a \"Create a pull request\" link - IGNORE IT.\nYou MUST run the `gh pr create` command above. Save the actual PR URL from the output.\n\nāš ļø AFTER PR CREATION YOU ARE NOT DONE! CONTINUE TO STEP 6! āš ļø\n\n### STEP 6: MERGE THE PR (MANDATORY - THIS IS NOT OPTIONAL)\n```bash\ngh pr merge --merge --auto\n```\nThis sets auto-merge. If it fails (e.g., no auto-merge enabled), try:\n```bash\ngh pr merge --merge\n```\n\nšŸšØ IF MERGE FAILS DUE TO CONFLICTS - YOU MUST RESOLVE THEM:\na) Pull latest dev and rebase:\n ```bash\n git fetch origin dev\n git rebase origin/dev\n ```\nb) If conflicts appear - RESOLVE THEM IMMEDIATELY:\n - Read the conflicting files\n - Make intelligent decisions about what code to keep\n - Edit the files to resolve conflicts\n - `git add `\n - `git rebase --continue`\nc) Force push the resolved branch:\n ```bash\n git push --force-with-lease\n ```\nd) Retry merge:\n ```bash\ngh pr merge --merge\n```\n\nREPEAT UNTIL MERGED. DO NOT GIVE UP. DO NOT SKIP. THE PR MUST BE MERGED.\nIf merge is blocked by CI, wait and retry. If blocked by reviews, set auto-merge.\n\n\n\n## CRITICAL RULES\n- Execute EVERY step in order (1, 2, 3, 4, 5, 6)\n- Do NOT skip git add -A\n- Do NOT skip git commit\n- Do NOT skip gh pr create - THE TASK IS NOT DONE UNTIL PR EXISTS\n- Do NOT skip gh pr merge --merge - THE TASK IS NOT DONE UNTIL PR IS MERGED\n- If push fails, debug and fix it\n- If PR creation fails, debug and fix it\n- If merge fails, debug and fix it\n- DO NOT OUTPUT JSON UNTIL PR IS MERGED\n- A link from git push is NOT a PR - you must run gh pr create\n\n## Final Output\nONLY after the PR is MERGED, output:\n```json\n{\"pr_url\": \"https://github.com/owner/repo/pull/123\", \"pr_number\": 123, \"merged\": true}\n```\n\nIf truly no changes exist, output:\n```json\n{\"pr_url\": null, \"pr_number\": null, \"merged\": false}\n```", + "hooks": { + "onComplete": { + "action": "verify_github_pr" + } + }, + "output": { + "topic": "PR_CREATED", + "publishAfter": "CLUSTER_COMPLETE" + }, + "structuredOutput": { + "type": "object", + "properties": { + "pr_number": { + "type": "number", + "description": "MUST extract from gh pr create output - NOT from git push link" + }, + "pr_url": { + "type": "string" + }, + "merged": { + "type": "boolean" + }, + "merge_commit_sha": { + "type": "string", + "description": "MUST extract from gh pr merge output" + } + }, + "required": ["pr_number", "pr_url", "merged", "merge_commit_sha"] + }, + "cwd": "/home/ubuntu/.zeroshot/worktrees/frozen-equinox-32", + "outputFormat": "json", + "jsonSchema": { + "type": "object", + "properties": { + "summary": { + "type": "string", + "description": "Brief summary of what was done" + }, + "result": { + "type": "string", + "description": "Detailed result or output" + } + }, + "required": ["summary", "result"] + }, + "strictSchema": true, + "timeout": 0 + } + ] + }, + "state": "running", + "createdAt": 1770023875452, + "pid": 3049126, + "failureInfo": null, + "autoPr": true, + "prOptions": null, + "modelOverride": null, + "issue": null, + "isolation": { + "enabled": true, + "containerId": "e900d72bbf07", + "image": "zeroshot-cluster-base", + "workDir": "/home/ubuntu/.zeroshot/worktrees/frozen-equinox-32" + }, + "agentStates": [ + { + "id": "test-conductor", + "state": "executing_task", + "iteration": 3, + "currentTask": false, + "currentTaskId": null, + "processPid": 3051855 + }, + { + "id": "git-pusher", + "state": "idle", + "iteration": 0, + "currentTask": false, + "currentTaskId": null, + "processPid": null + } + ] + } +} diff --git a/tests/unit/.test-storage/emerald-prism-4.db b/tests/unit/.test-storage/emerald-prism-4.db new file mode 100644 index 00000000..d53f7626 Binary files /dev/null and b/tests/unit/.test-storage/emerald-prism-4.db differ diff --git a/tests/unit/.test-storage/emerald-prism-4.db-shm b/tests/unit/.test-storage/emerald-prism-4.db-shm new file mode 100644 index 00000000..3641d16a Binary files /dev/null and b/tests/unit/.test-storage/emerald-prism-4.db-shm differ diff --git a/tests/unit/.test-storage/emerald-prism-4.db-wal b/tests/unit/.test-storage/emerald-prism-4.db-wal new file mode 100644 index 00000000..8c353136 Binary files /dev/null and b/tests/unit/.test-storage/emerald-prism-4.db-wal differ diff --git a/tests/unit/attach-stdin.test.js b/tests/unit/attach-stdin.test.js new file mode 100644 index 00000000..912190cd --- /dev/null +++ b/tests/unit/attach-stdin.test.js @@ -0,0 +1,84 @@ +const assert = require('assert'); +const fs = require('fs'); +const os = require('os'); +const path = require('path'); + +const { AttachServer } = require('../../src/attach'); +const { sendInput } = require('../../src/attach/send-input'); + +describe('Attach stdin', function () { + this.timeout(10000); + + it('sendInput writes to a live PTY', async function () { + const tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'zeroshot-attach-')); + const socketPath = path.join(tmpDir, 'attach.sock'); + + const server = new AttachServer({ + id: 'attach-stdin-test', + socketPath, + command: 'cat', + args: [], + cwd: process.cwd(), + env: process.env, + cols: 80, + rows: 24, + }); + + let output = ''; + let resolved = false; + + const outputPromise = new Promise((resolve, reject) => { + const timeout = setTimeout(() => { + if (!resolved) { + resolved = true; + reject(new Error('Timed out waiting for PTY output')); + } + }, 2000); + + const onOutput = (data) => { + output += data.toString(); + if (output.includes('hello-attach')) { + if (resolved) return; + resolved = true; + clearTimeout(timeout); + server.off('output', onOutput); + resolve(); + } + }; + + server.on('output', onOutput); + }); + + let testError; + let stopError; + try { + await server.start(); + const result = await sendInput({ + socketPath, + data: 'hello-attach\n', + timeoutMs: 1000, + }); + + assert.strictEqual(result.ok, true); + await outputPromise; + assert.ok(output.includes('hello-attach')); + } catch (error) { + testError = error; + } finally { + try { + await server.stop('SIGTERM'); + } catch (error) { + console.warn('AttachServer.stop failed in attach-stdin test', error); + stopError = error; + } + fs.rmSync(tmpDir, { recursive: true, force: true }); + } + + if (stopError && !testError) { + throw stopError; + } + if (testError) { + throw testError; + } + }); +}); diff --git a/tests/unit/claude-fatal-error-detection.test.js b/tests/unit/claude-fatal-error-detection.test.js new file mode 100644 index 00000000..a7deed4c --- /dev/null +++ b/tests/unit/claude-fatal-error-detection.test.js @@ -0,0 +1,48 @@ +const { describe, it } = require('mocha'); +const { expect } = require('chai'); +const path = require('path'); +const { pathToFileURL } = require('url'); + +function loadRecoveryModule() { + const modulePath = path.resolve(__dirname, '../../task-lib/claude-recovery.js'); + return import(pathToFileURL(modulePath).href); +} + +describe('Claude fatal error detection', () => { + it('detects "No messages returned" in error output', async () => { + const { detectFatalClaudeError, NO_MESSAGES_RETURNED } = await loadRecoveryModule(); + const line = 'Error: No messages returned'; + + const detected = detectFatalClaudeError(line); + + expect(detected).to.equal(`Claude CLI error: ${NO_MESSAGES_RETURNED}`); + }); + + it('is case-insensitive', async () => { + const { detectFatalClaudeError, NO_MESSAGES_RETURNED } = await loadRecoveryModule(); + const line = 'error: NO MESSAGES RETURNED'; + + const detected = detectFatalClaudeError(line); + + expect(detected).to.equal(`Claude CLI error: ${NO_MESSAGES_RETURNED}`); + }); + + it('returns null for unrelated output', async () => { + const { detectFatalClaudeError } = await loadRecoveryModule(); + + expect(detectFatalClaudeError('All good')).to.equal(null); + expect(detectFatalClaudeError('')).to.equal(null); + }); + + it('does not flag valid JSON output that contains the message', async () => { + const { detectFatalClaudeError } = await loadRecoveryModule(); + const jsonLine = JSON.stringify({ + type: 'result', + structured_output: { + summary: 'No messages returned in issue description', + }, + }); + + expect(detectFatalClaudeError(jsonLine)).to.equal(null); + }); +}); diff --git a/tests/unit/cli-default-entry.test.js b/tests/unit/cli-default-entry.test.js new file mode 100644 index 00000000..8d09d764 --- /dev/null +++ b/tests/unit/cli-default-entry.test.js @@ -0,0 +1,46 @@ +/** + * Test: CLI Default Entry Behavior (no args) + * + * Verifies zeroshot with no args launches TUI on interactive TTY + * and prints help on non-interactive input. + */ + +const assert = require('assert'); + +function resolveDefaultEntry(args, { isInteractiveTty }) { + let workingArgs = [...args]; + let shouldOutputHelp = false; + + if (workingArgs.length === 0) { + if (isInteractiveTty) { + workingArgs = ['tui']; + } else { + shouldOutputHelp = true; + } + } + + return { args: workingArgs, shouldOutputHelp }; +} + +describe('CLI Default Entry (no args)', function () { + it('routes to tui when no args and interactive TTY', function () { + const result = resolveDefaultEntry([], { isInteractiveTty: true }); + + assert.deepStrictEqual(result.args, ['tui']); + assert.strictEqual(result.shouldOutputHelp, false); + }); + + it('prints help when no args and non-interactive', function () { + const result = resolveDefaultEntry([], { isInteractiveTty: false }); + + assert.deepStrictEqual(result.args, []); + assert.strictEqual(result.shouldOutputHelp, true); + }); + + it('does not change args when already provided', function () { + const result = resolveDefaultEntry(['list'], { isInteractiveTty: true }); + + assert.deepStrictEqual(result.args, ['list']); + assert.strictEqual(result.shouldOutputHelp, false); + }); +}); diff --git a/tests/unit/cli-invalid-command.test.js b/tests/unit/cli-invalid-command.test.js index 5a38f213..fa9d8260 100644 --- a/tests/unit/cli-invalid-command.test.js +++ b/tests/unit/cli-invalid-command.test.js @@ -37,6 +37,11 @@ describe('CLI Invalid Command Handling', function () { 'purge', 'export', 'watch', + 'tui', + 'codex', + 'claude', + 'gemini', + 'opencode', 'attach', 'agents', 'config', @@ -101,6 +106,22 @@ describe('CLI Invalid Command Handling', function () { it('should not prepend run for "settings" command', function () { assert.strictEqual(shouldPrependRun(['settings']), false); }); + + it('should not prepend run for "codex" command', function () { + assert.strictEqual(shouldPrependRun(['codex']), false); + }); + + it('should not prepend run for "claude" command', function () { + assert.strictEqual(shouldPrependRun(['claude']), false); + }); + + it('should not prepend run for "gemini" command', function () { + assert.strictEqual(shouldPrependRun(['gemini']), false); + }); + + it('should not prepend run for "opencode" command', function () { + assert.strictEqual(shouldPrependRun(['opencode']), false); + }); }); describe('Flags should not trigger run prepending', function () { diff --git a/tests/unit/cli-pr-base-env.test.js b/tests/unit/cli-pr-base-env.test.js new file mode 100644 index 00000000..4adcdd9a --- /dev/null +++ b/tests/unit/cli-pr-base-env.test.js @@ -0,0 +1,137 @@ +/** + * Test: PR config env fallback + * + * Ensures detached/daemon runs can read PR config from env vars. + */ + +const assert = require('assert'); +const { buildStartOptions } = require('../../lib/start-cluster'); + +const ENV_VARS = [ + 'ZEROSHOT_CWD', + 'ZEROSHOT_RUN_OPTIONS', + 'ZEROSHOT_PR_BASE', + 'ZEROSHOT_MERGE_QUEUE', + 'ZEROSHOT_CLOSE_ISSUE', +]; + +const originalEnv = ENV_VARS.reduce((acc, key) => { + acc[key] = process.env[key]; + return acc; +}, {}); + +function restoreEnv() { + for (const key of ENV_VARS) { + if (originalEnv[key] === undefined) { + delete process.env[key]; + } else { + process.env[key] = originalEnv[key]; + } + } +} + +describe('CLI PR config env fallback', function () { + afterEach(function () { + restoreEnv(); + }); + + it('uses ZEROSHOT_PR_BASE when options.prBase is missing', function () { + process.env.ZEROSHOT_CWD = '/tmp/zeroshot-test'; + process.env.ZEROSHOT_PR_BASE = 'dev'; + + const result = buildStartOptions({ clusterId: 'test', options: {}, settings: {} }); + + assert.strictEqual(result.prBase, 'dev'); + }); + + it('uses ZEROSHOT_RUN_OPTIONS when options are missing', function () { + process.env.ZEROSHOT_CWD = '/tmp/zeroshot-test'; + process.env.ZEROSHOT_RUN_OPTIONS = JSON.stringify({ + prBase: 'dev', + mergeQueue: true, + closeIssue: 'always', + }); + + const result = buildStartOptions({ clusterId: 'test', options: {}, settings: {} }); + + assert.strictEqual(result.prBase, 'dev'); + assert.strictEqual(result.mergeQueue, true); + assert.strictEqual(result.closeIssue, 'always'); + }); + + it('prefers explicit options over ZEROSHOT_RUN_OPTIONS', function () { + process.env.ZEROSHOT_CWD = '/tmp/zeroshot-test'; + process.env.ZEROSHOT_RUN_OPTIONS = JSON.stringify({ + prBase: 'dev', + mergeQueue: false, + closeIssue: 'always', + }); + + const result = buildStartOptions({ + clusterId: 'test', + options: { prBase: 'main', mergeQueue: true, closeIssue: 'never' }, + settings: {}, + }); + + assert.strictEqual(result.prBase, 'main'); + assert.strictEqual(result.mergeQueue, true); + assert.strictEqual(result.closeIssue, 'never'); + }); + + it('prefers options.prBase over ZEROSHOT_PR_BASE', function () { + process.env.ZEROSHOT_CWD = '/tmp/zeroshot-test'; + process.env.ZEROSHOT_PR_BASE = 'dev'; + + const result = buildStartOptions({ + clusterId: 'test', + options: { prBase: 'main' }, + settings: {}, + }); + + assert.strictEqual(result.prBase, 'main'); + }); + + it('uses ZEROSHOT_MERGE_QUEUE when options.mergeQueue is missing', function () { + process.env.ZEROSHOT_CWD = '/tmp/zeroshot-test'; + process.env.ZEROSHOT_MERGE_QUEUE = '1'; + + const result = buildStartOptions({ clusterId: 'test', options: {}, settings: {} }); + + assert.strictEqual(result.mergeQueue, true); + }); + + it('prefers options.mergeQueue over ZEROSHOT_MERGE_QUEUE', function () { + process.env.ZEROSHOT_CWD = '/tmp/zeroshot-test'; + process.env.ZEROSHOT_MERGE_QUEUE = '0'; + + const result = buildStartOptions({ + clusterId: 'test', + options: { mergeQueue: true }, + settings: {}, + }); + + assert.strictEqual(result.mergeQueue, true); + }); + + it('uses ZEROSHOT_CLOSE_ISSUE when options.closeIssue is missing', function () { + process.env.ZEROSHOT_CWD = '/tmp/zeroshot-test'; + process.env.ZEROSHOT_CLOSE_ISSUE = 'always'; + + const result = buildStartOptions({ clusterId: 'test', options: {}, settings: {} }); + + assert.strictEqual(result.closeIssue, 'always'); + }); + + it('prefers options.closeIssue over ZEROSHOT_CLOSE_ISSUE', function () { + process.env.ZEROSHOT_CWD = '/tmp/zeroshot-test'; + process.env.ZEROSHOT_CLOSE_ISSUE = 'always'; + + const result = buildStartOptions({ + clusterId: 'test', + options: { closeIssue: 'never' }, + settings: {}, + }); + + assert.strictEqual(result.closeIssue, 'never'); + }); +}); diff --git a/tests/unit/cli-provider-override.test.js b/tests/unit/cli-provider-override.test.js new file mode 100644 index 00000000..59bc0752 --- /dev/null +++ b/tests/unit/cli-provider-override.test.js @@ -0,0 +1,74 @@ +/** + * Test: CLI Provider Override + * + * Verifies that provider override is only applied when explicitly set + * via --provider or ZEROSHOT_PROVIDER. + */ + +const assert = require('assert'); + +function normalizeProviderName(name) { + if (!name || typeof name !== 'string') return name; + const normalized = name.toLowerCase(); + if (normalized === 'anthropic') return 'claude'; + if (normalized === 'openai') return 'codex'; + if (normalized === 'google') return 'gemini'; + return normalized; +} + +// Mirrors resolveProviderOverride in cli/index.js +function resolveProviderOverride(options) { + const override = options.provider || process.env.ZEROSHOT_PROVIDER; + if (!override || (typeof override === 'string' && !override.trim())) { + return null; + } + return normalizeProviderName(override); +} + +describe('CLI Provider Override', function () { + const originalEnv = process.env.ZEROSHOT_PROVIDER; + + afterEach(function () { + if (originalEnv === undefined) { + delete process.env.ZEROSHOT_PROVIDER; + } else { + process.env.ZEROSHOT_PROVIDER = originalEnv; + } + }); + + it('returns null when no override is set', function () { + delete process.env.ZEROSHOT_PROVIDER; + const result = resolveProviderOverride({}); + assert.strictEqual(result, null); + }); + + it('uses --provider when provided', function () { + delete process.env.ZEROSHOT_PROVIDER; + const result = resolveProviderOverride({ provider: 'claude' }); + assert.strictEqual(result, 'claude'); + }); + + it('normalizes provider aliases', function () { + delete process.env.ZEROSHOT_PROVIDER; + const result = resolveProviderOverride({ provider: 'Anthropic' }); + assert.strictEqual(result, 'claude'); + }); + + it('uses ZEROSHOT_PROVIDER when --provider is missing', function () { + process.env.ZEROSHOT_PROVIDER = 'codex'; + const result = resolveProviderOverride({}); + assert.strictEqual(result, 'codex'); + }); + + it('ignores empty ZEROSHOT_PROVIDER', function () { + process.env.ZEROSHOT_PROVIDER = ' '; + const result = resolveProviderOverride({}); + assert.strictEqual(result, null); + }); + + it('prefers --provider over ZEROSHOT_PROVIDER', function () { + process.env.ZEROSHOT_PROVIDER = 'gemini'; + const result = resolveProviderOverride({ provider: 'claude' }); + assert.strictEqual(result, 'claude'); + }); +}); diff --git a/tests/unit/cli-resume-loads-clusters.test.js b/tests/unit/cli-resume-loads-clusters.test.js new file mode 100644 index 00000000..3516bb93 --- /dev/null +++ b/tests/unit/cli-resume-loads-clusters.test.js @@ -0,0 +1,24 @@ +const assert = require('assert'); +const fs = require('fs'); +const path = require('path'); + +describe('CLI resume command', function () { + it('should load clusters before checking for a task fallback', function () { + const cliPath = path.join(__dirname, '..', '..', 'cli', 'index.js'); + const cliCode = fs.readFileSync(cliPath, 'utf8'); + + const resumeStart = cliCode.indexOf(".command('resume"); + assert(resumeStart !== -1, 'resume command not found in cli/index.js'); + + const resumeEnd = cliCode.indexOf(".command('finish", resumeStart); + const resumeBlock = cliCode.slice(resumeStart, resumeEnd === -1 ? cliCode.length : resumeEnd); + + const usesCreate = + resumeBlock.includes('OrchestratorModule.create') || resumeBlock.includes('getOrchestrator('); + + assert( + usesCreate, + 'resume command should load clusters via Orchestrator.create or getOrchestrator' + ); + }); +}); diff --git a/tests/unit/cli-tui-binary-resolution.test.js b/tests/unit/cli-tui-binary-resolution.test.js new file mode 100644 index 00000000..9baeb0f3 --- /dev/null +++ b/tests/unit/cli-tui-binary-resolution.test.js @@ -0,0 +1,103 @@ +/** + * Test: CLI TUI Binary Resolution + * + * Verifies local Rust builds are preferred over installed libexec binaries. + */ + +const assert = require('assert'); +const fs = require('fs'); +const path = require('path'); +const { resolveRustTuiBinary } = require('../../lib/tui-launcher'); + +const DEFAULT_RUST_BIN_NAME = process.platform === 'win32' ? 'zeroshot-tui.exe' : 'zeroshot-tui'; +const DEBUG_SUFFIX = path.join('tui-rs', 'target', 'debug', DEFAULT_RUST_BIN_NAME); +const RELEASE_SUFFIX = path.join('tui-rs', 'target', 'release', DEFAULT_RUST_BIN_NAME); +const LIBEXEC_SUFFIX = path.join('libexec', DEFAULT_RUST_BIN_NAME); + +describe('CLI TUI binary resolution', function () { + function withPatchedExistsSync(mock, callback) { + const originalExistsSync = fs.existsSync; + fs.existsSync = mock; + try { + callback(); + } finally { + fs.existsSync = originalExistsSync; + } + } + + function withCleanBinaryEnv(callback) { + const keys = ['ZEROSHOT_TUI_BINARY_PATH', 'ZEROSHOT_TUI_PATH', 'ZEROSHOT_TUI_BIN']; + const previous = Object.fromEntries(keys.map((key) => [key, process.env[key]])); + + for (const key of keys) { + delete process.env[key]; + } + + try { + callback(); + } finally { + for (const key of keys) { + if (previous[key] === undefined) { + delete process.env[key]; + } else { + process.env[key] = previous[key]; + } + } + } + } + + it('prefers local debug build over release and installed libexec binaries', function () { + withCleanBinaryEnv(() => { + withPatchedExistsSync( + (candidate) => { + if (candidate.endsWith(DEBUG_SUFFIX)) { + return true; + } + if (candidate.endsWith(RELEASE_SUFFIX)) { + return true; + } + if (candidate.endsWith(LIBEXEC_SUFFIX)) { + return true; + } + return false; + }, + () => { + const resolved = resolveRustTuiBinary(); + assert(resolved.endsWith(DEBUG_SUFFIX)); + } + ); + }); + }); + + it('falls back to local release build when debug build is unavailable', function () { + withCleanBinaryEnv(() => { + withPatchedExistsSync( + (candidate) => { + if (candidate.endsWith(RELEASE_SUFFIX)) { + return true; + } + if (candidate.endsWith(LIBEXEC_SUFFIX)) { + return true; + } + return false; + }, + () => { + const resolved = resolveRustTuiBinary(); + assert(resolved.endsWith(RELEASE_SUFFIX)); + } + ); + }); + }); + + it('falls back to installed libexec binary when local build is unavailable', function () { + withCleanBinaryEnv(() => { + withPatchedExistsSync( + (candidate) => candidate.endsWith(LIBEXEC_SUFFIX), + () => { + const resolved = resolveRustTuiBinary(); + assert(resolved.endsWith(LIBEXEC_SUFFIX)); + } + ); + }); + }); +}); diff --git a/tests/unit/cli-tui-entrypoints.test.js b/tests/unit/cli-tui-entrypoints.test.js new file mode 100644 index 00000000..a8b5227c --- /dev/null +++ b/tests/unit/cli-tui-entrypoints.test.js @@ -0,0 +1,23 @@ +/** + * Test: CLI TUI Entrypoints + * + * Verifies provider-specific entrypoints map to TUI provider override. + */ + +const assert = require('assert'); +const { resolveTuiProviderOverride } = require('../../lib/tui-launcher'); + +function buildEntrypointOptions(providerName) { + return { providerOverride: resolveTuiProviderOverride({ provider: providerName }) }; +} + +describe('CLI TUI Entrypoints', function () { + const entrypoints = ['codex', 'claude', 'gemini', 'opencode']; + + for (const provider of entrypoints) { + it(`sets providerOverride for ${provider}`, function () { + const result = buildEntrypointOptions(provider); + assert.strictEqual(result.providerOverride, provider); + }); + } +}); diff --git a/tests/unit/cli-tui-launcher.test.js b/tests/unit/cli-tui-launcher.test.js new file mode 100644 index 00000000..7ee9a2e7 --- /dev/null +++ b/tests/unit/cli-tui-launcher.test.js @@ -0,0 +1,39 @@ +/** + * Test: CLI TUI Launcher + * + * Verifies Rust TUI spawn is default. + */ + +const assert = require('assert'); +const { launchTuiSession } = require('../../lib/tui-launcher'); + +describe('CLI TUI Launcher', function () { + it('spawns Rust TUI by default with initial screen + provider override', function () { + const spawnCalls = []; + const spawnStub = (command, args, options) => { + spawnCalls.push({ command, args, options }); + return { on: () => {} }; + }; + + launchTuiSession({ + initialView: 'monitor', + provider: 'codex', + spawn: spawnStub, + binaryPath: '/tmp/zeroshot-tui', + cwd: '/tmp', + }); + + assert.strictEqual(spawnCalls.length, 1); + assert.strictEqual(spawnCalls[0].command, '/tmp/zeroshot-tui'); + assert.deepStrictEqual(spawnCalls[0].args, [ + '--initial-screen', + 'monitor', + '--provider-override', + 'codex', + ]); + assert.strictEqual(spawnCalls[0].options.cwd, '/tmp'); + assert.strictEqual(spawnCalls[0].options.stdio, 'inherit'); + assert.strictEqual(spawnCalls[0].options.env.ZEROSHOT_TUI_INITIAL_SCREEN, 'monitor'); + assert.strictEqual(spawnCalls[0].options.env.ZEROSHOT_TUI_PROVIDER_OVERRIDE, 'codex'); + }); +}); diff --git a/tests/unit/cli-tui-provider-override.test.js b/tests/unit/cli-tui-provider-override.test.js new file mode 100644 index 00000000..9b66db36 --- /dev/null +++ b/tests/unit/cli-tui-provider-override.test.js @@ -0,0 +1,31 @@ +/** + * Test: CLI TUI Provider Override + * + * Verifies that --provider parsing for `zeroshot tui` normalizes and validates + * the provider override passed to the TUI bootstrap. + */ + +const assert = require('assert'); +const { buildRustTuiCommand, resolveTuiProviderOverride } = require('../../lib/tui-launcher'); + +describe('CLI TUI Provider Override', function () { + it('returns null when no override is set', function () { + const result = resolveTuiProviderOverride({}); + assert.strictEqual(result, null); + }); + + it('passes provider override into Rust TUI command', function () { + const result = buildRustTuiCommand({ provider: 'codex', binaryPath: '/tmp/zeroshot-tui' }); + assert.deepStrictEqual(result.args, ['--provider-override', 'codex']); + assert.strictEqual(result.env.ZEROSHOT_TUI_PROVIDER_OVERRIDE, 'codex'); + }); + + it('normalizes provider aliases', function () { + const result = resolveTuiProviderOverride({ provider: 'OpenAI' }); + assert.strictEqual(result, 'codex'); + }); + + it('throws on unknown provider', function () { + assert.throws(() => resolveTuiProviderOverride({ provider: 'invalid' }), /Unknown provider:/); + }); +}); diff --git a/tests/unit/cli-tui-ui-variant.test.js b/tests/unit/cli-tui-ui-variant.test.js new file mode 100644 index 00000000..017f35a7 --- /dev/null +++ b/tests/unit/cli-tui-ui-variant.test.js @@ -0,0 +1,25 @@ +/** + * Test: CLI TUI UI Variant + * + * Verifies UI variant parsing and Rust TUI args/env plumbing. + */ + +const assert = require('assert'); +const { buildRustTuiCommand, resolveUiVariant } = require('../../lib/tui-launcher'); + +describe('CLI TUI UI Variant', function () { + it('returns null when no variant is set', function () { + const result = resolveUiVariant({}); + assert.strictEqual(result, null); + }); + + it('passes normalized ui variant into Rust TUI command', function () { + const result = buildRustTuiCommand({ ui: 'Disruptive', binaryPath: '/tmp/zeroshot-tui' }); + assert.deepStrictEqual(result.args, ['--ui', 'disruptive']); + assert.strictEqual(result.env.ZEROSHOT_TUI_UI, 'disruptive'); + }); + + it('throws on unknown ui variant', function () { + assert.throws(() => resolveUiVariant({ ui: 'weird' }), /Unknown UI variant/); + }); +}); diff --git a/tests/unit/context-metrics.test.js b/tests/unit/context-metrics.test.js new file mode 100644 index 00000000..b78a03de --- /dev/null +++ b/tests/unit/context-metrics.test.js @@ -0,0 +1,107 @@ +const assert = require('assert'); +const { buildContextMetrics, estimateTokensFromChars } = require('../../src/agent/context-metrics'); + +describe('Context Metrics', function () { + it('estimates tokens using ceil(chars / 4)', function () { + assert.strictEqual(estimateTokensFromChars(0), 0); + assert.strictEqual(estimateTokensFromChars(1), 1); + assert.strictEqual(estimateTokensFromChars(4), 1); + assert.strictEqual(estimateTokensFromChars(5), 2); + }); + + it('builds section breakdown with totals', function () { + const metrics = buildContextMetrics({ + clusterId: 'cluster-1', + agentId: 'agent-1', + role: 'worker', + iteration: 2, + triggeringMessage: { topic: 'TASK', sender: 'user' }, + strategy: { sources: [{ topic: 'A' }, { topic: 'B' }], maxTokens: 10 }, + packs: [ + { id: 'header', section: 'header', status: 'included', chars: 4, estimatedTokens: 1 }, + { + id: 'instructions', + section: 'instructions', + status: 'included', + chars: 5, + estimatedTokens: 2, + }, + { + id: 'jsonSchema', + section: 'jsonSchema', + status: 'included', + chars: 2, + estimatedTokens: 1, + }, + { + id: 'validatorSkip', + section: 'validatorSkip', + status: 'included', + chars: 1, + estimatedTokens: 1, + }, + { + id: 'triggeringMessage', + section: 'triggeringMessage', + status: 'included', + chars: 1, + estimatedTokens: 1, + }, + { + id: 'sources-a', + section: 'sources', + status: 'included', + chars: 3, + estimatedTokens: 1, + }, + { + id: 'sources-skipped', + section: 'sources', + status: 'skipped', + chars: 100, + estimatedTokens: 25, + }, + ], + budget: { + maxTokens: 10, + remainingTokens: 2, + overBudgetTokens: 0, + finalTokens: 6, + }, + }); + + assert.strictEqual(metrics.sections.header.chars, 4); + assert.strictEqual(metrics.sections.header.estimatedTokens, 1); + assert.strictEqual(metrics.sections.instructions.chars, 5); + assert.strictEqual(metrics.sections.instructions.estimatedTokens, 2); + assert.strictEqual(metrics.sections.jsonSchema.chars, 2); + assert.strictEqual(metrics.sections.validatorSkip.chars, 1); + assert.strictEqual(metrics.sections.triggeringMessage.chars, 1); + assert.strictEqual(metrics.sections.sources.chars, 3); + assert.strictEqual(metrics.total.chars, 4 + 5 + 2 + 1 + 1 + 3); + assert.strictEqual(metrics.total.estimatedTokens, Math.ceil(metrics.total.chars / 4)); + + assert.strictEqual(metrics.strategy.maxTokens, 10); + assert.strictEqual(metrics.strategy.sourcesCount, 2); + assert.strictEqual(metrics.triggeredBy, 'TASK'); + assert.strictEqual(metrics.triggerFrom, 'user'); + assert.strictEqual(metrics.budget.maxTokens, 10); + assert.strictEqual(metrics.budget.remainingTokens, 2); + assert.strictEqual(metrics.truncation.maxContextChars.beforeChars, metrics.total.chars); + }); + + it('defaults maxTokens when not provided', function () { + const metrics = buildContextMetrics({ + clusterId: 'cluster-1', + agentId: 'agent-1', + role: 'worker', + iteration: 1, + triggeringMessage: { topic: 'TASK', sender: 'user' }, + strategy: { sources: [] }, + packs: [{ id: 'header', section: 'header', status: 'included', chars: 1 }], + }); + + assert.strictEqual(metrics.strategy.maxTokens, 100000); + assert.strictEqual(metrics.budget.maxTokens, 100000); + }); +}); diff --git a/tests/unit/github-provider-parse-identifier.test.js b/tests/unit/github-provider-parse-identifier.test.js new file mode 100644 index 00000000..b4746f1f --- /dev/null +++ b/tests/unit/github-provider-parse-identifier.test.js @@ -0,0 +1,67 @@ +/** + * Regression test: GitHub provider must ALWAYS extract repo from identifier + * + * Bug: _extractIssueNumber discarded repo, causing `gh issue view` to guess + * from CWD git remote, which failed when CWD was in a different repo. + */ + +const { expect } = require('chai'); +const GitHubProvider = require('../../src/issue-providers/github-provider'); + +describe('GitHubProvider._parseIdentifier', () => { + let provider; + + beforeEach(() => { + provider = new GitHubProvider(); + }); + + describe('extracts repo from identifier', () => { + it('org/repo#123 format returns both repo and number', () => { + const result = provider._parseIdentifier('covibes/covibes#1172'); + expect(result).to.deep.equal({ repo: 'covibes/covibes', number: '1172' }); + }); + + it('org-with-dash/repo-with-dash#123 format', () => { + const result = provider._parseIdentifier('my-org/my-repo#456'); + expect(result).to.deep.equal({ repo: 'my-org/my-repo', number: '456' }); + }); + + it('org.with.dots/repo.with.dots#123 format', () => { + const result = provider._parseIdentifier('my.org/my.repo#789'); + expect(result).to.deep.equal({ repo: 'my.org/my.repo', number: '789' }); + }); + + it('GitHub URL extracts repo and number', () => { + const result = provider._parseIdentifier('https://github.com/covibes/covibes/issues/1172'); + expect(result).to.deep.equal({ repo: 'covibes/covibes', number: '1172' }); + }); + + it('bare number with gitContext uses context repo', () => { + const gitContext = { owner: 'covibes', repo: 'covibes' }; + const result = provider._parseIdentifier('1172', gitContext); + expect(result).to.deep.equal({ repo: 'covibes/covibes', number: '1172' }); + }); + + it('bare number without gitContext returns null repo', () => { + const result = provider._parseIdentifier('1172', null); + expect(result).to.deep.equal({ repo: null, number: '1172' }); + }); + }); + + describe('never loses repo information', () => { + it('explicit repo takes precedence over gitContext', () => { + const gitContext = { owner: 'other', repo: 'repo' }; + const result = provider._parseIdentifier('covibes/covibes#1172', gitContext); + expect(result.repo).to.equal('covibes/covibes'); + }); + + it('URL repo takes precedence over gitContext', () => { + const gitContext = { owner: 'other', repo: 'repo' }; + const result = provider._parseIdentifier( + 'https://github.com/covibes/covibes/issues/1172', + gitContext + ); + expect(result.repo).to.equal('covibes/covibes'); + }); + }); +}); diff --git a/tests/unit/guidance-delivery.test.js b/tests/unit/guidance-delivery.test.js new file mode 100644 index 00000000..f9fc5d67 --- /dev/null +++ b/tests/unit/guidance-delivery.test.js @@ -0,0 +1,276 @@ +const fs = require('fs'); +const os = require('os'); +const path = require('path'); + +const tempHome = fs.mkdtempSync(path.join(os.tmpdir(), 'zeroshot-guidance-')); +process.env.ZEROSHOT_HOME = tempHome; + +const assert = require('assert'); + +const Orchestrator = require('../../src/orchestrator'); +const AgentWrapper = require('../../src/agent-wrapper'); +const MessageBus = require('../../src/message-bus'); +const Ledger = require('../../src/ledger'); +const { USER_GUIDANCE_AGENT, USER_GUIDANCE_CLUSTER } = require('../../src/guidance-topics'); +const { AttachServer } = require('../../src/attach'); + +describe('Guidance delivery', function () { + this.timeout(10000); + + let orchestrator; + let cluster; + let agent; + let ledger; + + beforeEach(() => { + ledger = new Ledger(':memory:'); + const messageBus = new MessageBus(ledger); + cluster = { + id: 'guidance-cluster', + messageBus, + ledger, + agents: [], + config: {}, + }; + + orchestrator = new Orchestrator({ + quiet: true, + skipLoad: true, + storageDir: path.join(tempHome, 'clusters'), + }); + + agent = new AgentWrapper( + { + id: 'agent-1', + role: 'implementation', + modelLevel: 'level1', + prompt: 'noop', + triggers: [], + }, + messageBus, + cluster, + { testMode: true } + ); + + cluster.agents.push(agent); + orchestrator.clusters.set(cluster.id, cluster); + }); + + afterEach(() => { + ledger.close(); + orchestrator.close(); + orchestrator.clusters.clear(); + }); + + after(() => { + fs.rmSync(tempHome, { recursive: true, force: true }); + }); + + it('publishes unsupported delivery metadata when no live socket is available', async () => { + const result = await orchestrator.sendGuidanceToAgent(cluster.id, agent.id, 'Use approach A'); + + assert.strictEqual(result.status, 'unsupported'); + + const messages = cluster.messageBus.query({ + cluster_id: cluster.id, + topic: USER_GUIDANCE_AGENT, + }); + + assert.strictEqual(messages.length, 1); + assert.strictEqual(messages[0].metadata.delivery.status, 'unsupported'); + }); + + it('publishes injected delivery metadata when attach socket is live', async () => { + const { addTask, ensureDirs, removeTask } = await import('../../task-lib/store.js'); + ensureDirs(); + + const taskId = 'task-guidance-1'; + const socketDir = fs.mkdtempSync(path.join(os.tmpdir(), 'zeroshot-guidance-sock-')); + const socketPath = path.join(socketDir, 'attach.sock'); + + const server = new AttachServer({ + id: taskId, + socketPath, + command: 'cat', + args: [], + cwd: process.cwd(), + env: process.env, + cols: 80, + rows: 24, + }); + + let testError; + let stopError; + try { + await server.start(); + + addTask({ + id: taskId, + prompt: 'test', + fullPrompt: 'test', + cwd: process.cwd(), + status: 'running', + pid: server.pid, + sessionId: null, + logFile: null, + createdAt: new Date().toISOString(), + updatedAt: new Date().toISOString(), + exitCode: null, + error: null, + provider: 'claude', + model: null, + scheduleId: null, + socketPath, + attachable: true, + }); + + agent.currentTaskId = taskId; + + const result = await orchestrator.sendGuidanceToAgent( + cluster.id, + agent.id, + 'Injected guidance' + ); + + assert.strictEqual(result.status, 'injected'); + + const messages = cluster.messageBus.query({ + cluster_id: cluster.id, + topic: USER_GUIDANCE_AGENT, + }); + + assert.strictEqual(messages.length, 1); + assert.strictEqual(messages[0].metadata.delivery.status, 'injected'); + assert.strictEqual(messages[0].metadata.delivery.method, 'pty'); + assert.strictEqual(messages[0].metadata.delivery.taskId, taskId); + } catch (error) { + testError = error; + } finally { + removeTask(taskId); + try { + await server.stop('SIGTERM'); + } catch (error) { + console.warn('AttachServer.stop failed in guidance-delivery test', error); + stopError = error; + } + fs.rmSync(socketDir, { recursive: true, force: true }); + } + + if (stopError && !testError) { + throw stopError; + } + if (testError) { + throw testError; + } + }); + + it('broadcasts cluster guidance with per-agent delivery results', async () => { + const { addTask, ensureDirs, removeTask } = await import('../../task-lib/store.js'); + ensureDirs(); + + const taskId = 'task-guidance-cluster-1'; + const socketDir = fs.mkdtempSync(path.join(os.tmpdir(), 'zeroshot-guidance-cluster-sock-')); + const socketPath = path.join(socketDir, 'attach.sock'); + + const server = new AttachServer({ + id: taskId, + socketPath, + command: 'cat', + args: [], + cwd: process.cwd(), + env: process.env, + cols: 80, + rows: 24, + }); + + const agentTwo = new AgentWrapper( + { + id: 'agent-2', + role: 'implementation', + modelLevel: 'level1', + prompt: 'noop', + triggers: [], + }, + cluster.messageBus, + cluster, + { testMode: true } + ); + + cluster.agents.push(agentTwo); + + let testError; + let stopError; + try { + await server.start(); + + addTask({ + id: taskId, + prompt: 'test', + fullPrompt: 'test', + cwd: process.cwd(), + status: 'running', + pid: server.pid, + sessionId: null, + logFile: null, + createdAt: new Date().toISOString(), + updatedAt: new Date().toISOString(), + exitCode: null, + error: null, + provider: 'claude', + model: null, + scheduleId: null, + socketPath, + attachable: true, + }); + + agent.currentTaskId = taskId; + + const result = await orchestrator.sendGuidanceToCluster( + cluster.id, + 'Cluster guidance message' + ); + + assert.deepStrictEqual(result.summary, { injected: 1, queued: 1, total: 2 }); + assert.strictEqual(result.agents[agent.id].status, 'injected'); + assert.strictEqual(result.agents[agent.id].method, 'pty'); + assert.strictEqual(result.agents[agent.id].taskId, taskId); + assert.strictEqual(result.agents[agentTwo.id].status, 'unsupported'); + assert.ok(result.agents[agentTwo.id].reason); + + const clusterMessages = cluster.messageBus.query({ + cluster_id: cluster.id, + topic: USER_GUIDANCE_CLUSTER, + }); + const agentMessages = cluster.messageBus.query({ + cluster_id: cluster.id, + topic: USER_GUIDANCE_AGENT, + }); + + assert.strictEqual(clusterMessages.length, 1); + assert.strictEqual(agentMessages.length, 0); + + const deliveryMeta = clusterMessages[0].metadata.delivery; + assert.deepStrictEqual(deliveryMeta.summary, { injected: 1, queued: 1, total: 2 }); + assert.strictEqual(deliveryMeta.agents[agent.id].status, 'injected'); + assert.strictEqual(deliveryMeta.agents[agentTwo.id].status, 'unsupported'); + } catch (error) { + testError = error; + } finally { + removeTask(taskId); + try { + await server.stop('SIGTERM'); + } catch (error) { + console.warn('AttachServer.stop failed in guidance-delivery test', error); + stopError = error; + } + fs.rmSync(socketDir, { recursive: true, force: true }); + } + + if (stopError && !testError) { + throw stopError; + } + if (testError) { + throw testError; + } + }); +}); diff --git a/tests/unit/guidance-mailbox.test.js b/tests/unit/guidance-mailbox.test.js new file mode 100644 index 00000000..367fff28 --- /dev/null +++ b/tests/unit/guidance-mailbox.test.js @@ -0,0 +1,132 @@ +const assert = require('assert'); + +const Ledger = require('../../src/ledger'); +const { USER_GUIDANCE_AGENT, USER_GUIDANCE_CLUSTER } = require('../../src/guidance-topics'); + +describe('Guidance mailbox', function () { + it('persists guidance messages with target_agent_id mapped to receiver', function () { + const ledger = new Ledger(':memory:'); + const clusterId = 'guidance-mailbox-1'; + + const published = ledger.append({ + cluster_id: clusterId, + topic: USER_GUIDANCE_AGENT, + sender: 'user', + target_agent_id: 'agent-1', + content: { text: 'Use approach A' }, + }); + + assert.strictEqual(published.receiver, 'agent-1'); + + const rows = ledger.query({ cluster_id: clusterId, topic: USER_GUIDANCE_AGENT }); + assert.strictEqual(rows.length, 1); + assert.strictEqual(rows[0].receiver, 'agent-1'); + + ledger.close(); + }); + + it('returns cluster + agent guidance since last delivered in deterministic order', function () { + const ledger = new Ledger(':memory:'); + const clusterId = 'guidance-mailbox-2'; + const now = Date.now(); + + ledger.append({ + cluster_id: clusterId, + topic: USER_GUIDANCE_CLUSTER, + sender: 'user', + content: { text: 'Old cluster guidance' }, + timestamp: now + 10, + }); + + ledger.append({ + cluster_id: clusterId, + topic: USER_GUIDANCE_AGENT, + sender: 'user', + target_agent_id: 'agent-a', + content: { text: 'Old agent guidance' }, + timestamp: now + 20, + }); + + ledger.append({ + cluster_id: clusterId, + topic: USER_GUIDANCE_CLUSTER, + sender: 'user', + content: { text: 'Cluster guidance 1' }, + timestamp: now + 200, + }); + + ledger.append({ + cluster_id: clusterId, + topic: USER_GUIDANCE_AGENT, + sender: 'user', + target_agent_id: 'agent-a', + content: { text: 'Agent guidance' }, + timestamp: now + 210, + }); + + ledger.append({ + cluster_id: clusterId, + topic: USER_GUIDANCE_CLUSTER, + sender: 'user', + content: { text: 'Cluster guidance 2' }, + timestamp: now + 220, + }); + + const mailbox = ledger.queryGuidanceMailbox({ + cluster_id: clusterId, + target_agent_id: 'agent-a', + lastDeliveredAt: now + 100, + }); + + const topics = mailbox.map((message) => message.topic); + assert.deepStrictEqual(topics, [ + USER_GUIDANCE_CLUSTER, + USER_GUIDANCE_AGENT, + USER_GUIDANCE_CLUSTER, + ]); + + const timestamps = mailbox.map((message) => message.timestamp); + const sorted = [...timestamps].sort((a, b) => a - b); + assert.deepStrictEqual(timestamps, sorted); + + const texts = mailbox.map((message) => message.content?.text); + assert.deepStrictEqual(texts, ['Cluster guidance 1', 'Agent guidance', 'Cluster guidance 2']); + + ledger.close(); + }); + + it('excludes guidance for other agents', function () { + const ledger = new Ledger(':memory:'); + const clusterId = 'guidance-mailbox-3'; + const now = Date.now(); + + ledger.append({ + cluster_id: clusterId, + topic: USER_GUIDANCE_AGENT, + sender: 'user', + target_agent_id: 'agent-a', + content: { text: 'Target agent' }, + timestamp: now + 10, + }); + + ledger.append({ + cluster_id: clusterId, + topic: USER_GUIDANCE_AGENT, + sender: 'user', + target_agent_id: 'agent-b', + content: { text: 'Other agent' }, + timestamp: now + 20, + }); + + const mailbox = ledger.queryGuidanceMailbox({ + cluster_id: clusterId, + target_agent_id: 'agent-a', + lastDeliveredAt: now - 1, + }); + + assert.strictEqual(mailbox.length, 1); + assert.strictEqual(mailbox[0].content.text, 'Target agent'); + + ledger.close(); + }); +}); diff --git a/tests/unit/guidance-queue.test.js b/tests/unit/guidance-queue.test.js new file mode 100644 index 00000000..9dfcf7be --- /dev/null +++ b/tests/unit/guidance-queue.test.js @@ -0,0 +1,193 @@ +const assert = require('assert'); + +const MessageBus = require('../../src/message-bus'); +const Ledger = require('../../src/ledger'); +const { + GUIDANCE_BLOCK_START, + GUIDANCE_BLOCK_END, + formatGuidanceBlock, + collectQueuedGuidance, +} = require('../../src/agent/guidance-queue'); +const { USER_GUIDANCE_AGENT, USER_GUIDANCE_CLUSTER } = require('../../src/guidance-topics'); + +describe('Guidance queue formatting', () => { + it('formats a single delimited guidance block in timestamp order', () => { + const now = Date.now(); + const messages = [ + { + topic: USER_GUIDANCE_CLUSTER, + sender: 'user', + timestamp: now + 20, + content: { text: 'Second' }, + }, + { + topic: USER_GUIDANCE_AGENT, + sender: 'user', + timestamp: now + 10, + content: { text: 'First' }, + }, + ]; + + const block = formatGuidanceBlock(messages); + + assert(block.includes('## Guidance (Queued)'), 'block includes header'); + assert(block.includes(GUIDANCE_BLOCK_START), 'block includes start marker'); + assert(block.includes(GUIDANCE_BLOCK_END), 'block includes end marker'); + + const firstIndex = block.indexOf('First'); + const secondIndex = block.indexOf('Second'); + assert(firstIndex > -1 && secondIndex > -1, 'block includes both messages'); + assert(firstIndex < secondIndex, 'messages are ordered by timestamp'); + }); + + it('returns empty string when no guidance exists', () => { + assert.strictEqual(formatGuidanceBlock([]), ''); + assert.strictEqual(formatGuidanceBlock(null), ''); + }); +}); + +describe('Guidance queue collection', () => { + it('deduplicates across iterations via lastDeliveredAt cursor', () => { + const ledger = new Ledger(':memory:'); + const messageBus = new MessageBus(ledger); + const clusterId = 'guidance-queue-1'; + const now = Date.now(); + + ledger.append({ + cluster_id: clusterId, + topic: USER_GUIDANCE_CLUSTER, + sender: 'user', + content: { text: 'Initial' }, + timestamp: now + 5, + }); + + const first = collectQueuedGuidance({ + messageBus, + clusterId, + agentId: 'agent-a', + lastDeliveredAt: null, + }); + + assert.strictEqual(first.messages.length, 1); + assert.strictEqual(first.latestTimestamp, now + 5); + assert(first.guidanceBlock.includes('Initial')); + + const second = collectQueuedGuidance({ + messageBus, + clusterId, + agentId: 'agent-a', + lastDeliveredAt: first.latestTimestamp, + }); + + assert.strictEqual(second.messages.length, 0); + assert.strictEqual(second.guidanceBlock, ''); + + ledger.close(); + }); + + it('collects cluster and agent guidance in order', () => { + const ledger = new Ledger(':memory:'); + const messageBus = new MessageBus(ledger); + const clusterId = 'guidance-queue-2'; + const now = Date.now(); + + ledger.append({ + cluster_id: clusterId, + topic: USER_GUIDANCE_CLUSTER, + sender: 'user', + content: { text: 'Cluster 1' }, + timestamp: now + 10, + }); + + ledger.append({ + cluster_id: clusterId, + topic: USER_GUIDANCE_AGENT, + sender: 'user', + target_agent_id: 'agent-a', + content: { text: 'Agent 1' }, + timestamp: now + 20, + }); + + ledger.append({ + cluster_id: clusterId, + topic: USER_GUIDANCE_CLUSTER, + sender: 'user', + content: { text: 'Cluster 2' }, + timestamp: now + 30, + }); + + const result = collectQueuedGuidance({ + messageBus, + clusterId, + agentId: 'agent-a', + lastDeliveredAt: now, + }); + + assert.deepStrictEqual( + result.messages.map((message) => message.content.text), + ['Cluster 1', 'Agent 1', 'Cluster 2'] + ); + + ledger.close(); + }); +}); + +describe('Guidance queue placement in context', () => { + it('injects guidance between instructions and JSON output schema', () => { + const ledger = new Ledger(':memory:'); + const messageBus = new MessageBus(ledger); + const clusterId = 'guidance-queue-context'; + const cluster = { id: clusterId, createdAt: Date.now() - 1000 }; + + const config = { + id: 'worker', + role: 'implementation', + outputFormat: 'json', + jsonSchema: { + type: 'object', + properties: { ok: { type: 'boolean' } }, + required: ['ok'], + }, + contextStrategy: { sources: [] }, + prompt: 'Do the work.', + }; + + const { buildContext } = require('../../src/agent/agent-context-builder'); + const guidanceBlock = formatGuidanceBlock([ + { + topic: USER_GUIDANCE_CLUSTER, + sender: 'user', + timestamp: Date.now(), + content: { text: 'Queued guidance' }, + }, + ]); + + const context = buildContext({ + id: 'worker', + role: 'implementation', + iteration: 1, + config, + messageBus, + cluster, + triggeringMessage: { + cluster_id: clusterId, + topic: 'ISSUE_OPENED', + sender: 'tester', + content: { text: 'Task' }, + }, + queuedGuidance: guidanceBlock, + }); + + const instructionsIndex = context.indexOf('## Instructions'); + const guidanceIndex = context.indexOf('## Guidance (Queued)'); + const outputSchemaIndex = context.indexOf('## šŸ”“ OUTPUT FORMAT - JSON ONLY'); + + assert(instructionsIndex !== -1, 'instructions present'); + assert(guidanceIndex !== -1, 'guidance present'); + assert(outputSchemaIndex !== -1, 'json schema present'); + assert(instructionsIndex < guidanceIndex, 'guidance after instructions'); + assert(guidanceIndex < outputSchemaIndex, 'guidance before json schema'); + + ledger.close(); + }); +}); diff --git a/tests/unit/orchestrator-agent-error-stop.test.js b/tests/unit/orchestrator-agent-error-stop.test.js new file mode 100644 index 00000000..115245ea --- /dev/null +++ b/tests/unit/orchestrator-agent-error-stop.test.js @@ -0,0 +1,85 @@ +const assert = require('node:assert'); +const fs = require('node:fs'); +const os = require('node:os'); +const path = require('node:path'); + +const sinon = require('sinon'); + +const Ledger = require('../../src/ledger'); +const MessageBus = require('../../src/message-bus'); +const Orchestrator = require('../../src/orchestrator'); + +describe('Orchestrator critical agent error handling', function () { + this.timeout(10_000); + + let tempDir; + let ledger; + let messageBus; + let orchestrator; + + beforeEach(() => { + tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'zeroshot-orchestrator-agent-error-')); + ledger = new Ledger(path.join(tempDir, 'test.db')); + messageBus = new MessageBus(ledger); + + orchestrator = new Orchestrator({ quiet: true, skipLoad: true, storageDir: tempDir }); + sinon.stub(orchestrator, '_saveClusters').resolves(); + }); + + afterEach(() => { + sinon.restore(); + if (ledger) ledger.close(); + if (tempDir && fs.existsSync(tempDir)) { + fs.rmSync(tempDir, { recursive: true, force: true }); + } + }); + + it('stops cluster when coordinator fails after retries', async () => { + const stopSpy = sinon.stub(orchestrator, 'stop').resolves(); + orchestrator._registerAgentErrorHandler(messageBus, 'c1'); + + messageBus.publish({ + cluster_id: 'c1', + topic: 'AGENT_ERROR', + sender: 'consensus-coordinator', + content: { data: { role: 'coordinator', attempts: 3, error: 'boom' } }, + }); + + await new Promise((r) => setTimeout(r, 10)); + assert.equal(stopSpy.calledOnce, true); + assert.equal(stopSpy.firstCall.args[0], 'c1'); + }); + + it('stops cluster immediately when hookFailure is true (even with attempts=1)', async () => { + const stopSpy = sinon.stub(orchestrator, 'stop').resolves(); + orchestrator._registerAgentErrorHandler(messageBus, 'c2'); + + messageBus.publish({ + cluster_id: 'c2', + topic: 'AGENT_ERROR', + sender: 'consensus-coordinator', + content: { + data: { role: 'coordinator', attempts: 1, hookFailure: true, error: 'hook died' }, + }, + }); + + await new Promise((r) => setTimeout(r, 10)); + assert.equal(stopSpy.calledOnce, true); + assert.equal(stopSpy.firstCall.args[0], 'c2'); + }); + + it('does not stop cluster for validator errors by default', async () => { + const stopSpy = sinon.stub(orchestrator, 'stop').resolves(); + orchestrator._registerAgentErrorHandler(messageBus, 'c3'); + + messageBus.publish({ + cluster_id: 'c3', + topic: 'AGENT_ERROR', + sender: 'validator-1', + content: { data: { role: 'validator', attempts: 3, error: 'nope' } }, + }); + + await new Promise((r) => setTimeout(r, 10)); + assert.equal(stopSpy.called, false); + }); +}); diff --git a/tests/unit/orchestrator-issue-duplicate-check.test.js b/tests/unit/orchestrator-issue-duplicate-check.test.js new file mode 100644 index 00000000..00b2301d --- /dev/null +++ b/tests/unit/orchestrator-issue-duplicate-check.test.js @@ -0,0 +1,38 @@ +const assert = require('node:assert'); +const fs = require('node:fs'); +const os = require('node:os'); +const path = require('node:path'); + +const Orchestrator = require('../../src/orchestrator'); + +describe('Orchestrator duplicate issue check', function () { + this.timeout(5000); + + let tempDir; + let orchestrator; + + beforeEach(() => { + tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'zeroshot-orchestrator-dup-')); + orchestrator = new Orchestrator({ quiet: true, skipLoad: true, storageDir: tempDir }); + }); + + afterEach(() => { + if (tempDir && fs.existsSync(tempDir)) { + fs.rmSync(tempDir, { recursive: true, force: true }); + } + }); + + it('does not treat the current cluster as a duplicate of itself', () => { + const clusterId = 'self'; + orchestrator.clusters.set(clusterId, { + id: clusterId, + issue: 1172, + state: 'initializing', + pid: process.pid, + createdAt: Date.now(), + }); + + const active = orchestrator._getActiveClustersForIssue(1172, clusterId); + assert.deepEqual(active, []); + }); +}); diff --git a/tests/unit/orchestrator-provider-reload.test.js b/tests/unit/orchestrator-provider-reload.test.js new file mode 100644 index 00000000..0e215797 --- /dev/null +++ b/tests/unit/orchestrator-provider-reload.test.js @@ -0,0 +1,100 @@ +/** + * Orchestrator Provider Reload Test + * + * Regression test for provider resolution when reloading clusters from storage. + * Ensures forceProvider is honored in status output after reload. + */ + +const assert = require('assert'); +const fs = require('fs'); +const path = require('path'); +const os = require('os'); +const Orchestrator = require('../../src/orchestrator.js'); +const Ledger = require('../../src/ledger.js'); + +function createTempDir(prefix) { + return fs.mkdtempSync(path.join(os.tmpdir(), prefix)); +} + +function cleanupDir(dirPath) { + if (dirPath && fs.existsSync(dirPath)) { + fs.rmSync(dirPath, { recursive: true, force: true }); + } +} + +describe('Orchestrator provider reload', function () { + const originalSettingsFile = process.env.ZEROSHOT_SETTINGS_FILE; + + afterEach(function () { + if (originalSettingsFile === undefined) { + delete process.env.ZEROSHOT_SETTINGS_FILE; + } else { + process.env.ZEROSHOT_SETTINGS_FILE = originalSettingsFile; + } + }); + + it('uses config.forceProvider after reload', async function () { + const storageDir = createTempDir('zeroshot-provider-reload-'); + const settingsDir = createTempDir('zeroshot-provider-settings-'); + const settingsFile = path.join(settingsDir, 'settings.json'); + + fs.writeFileSync(settingsFile, JSON.stringify({ defaultProvider: 'claude' })); + process.env.ZEROSHOT_SETTINGS_FILE = settingsFile; + + const clusterId = 'provider-reload-test'; + const clusterData = { + id: clusterId, + config: { + forceProvider: 'codex', + agents: [ + { + id: 'worker', + role: 'implementation', + modelLevel: 'level2', + outputFormat: 'text', + triggers: [{ topic: 'ISSUE_OPENED', action: 'execute_task' }], + prompt: 'Worker', + }, + ], + }, + state: 'stopped', + createdAt: Date.now(), + autoPr: false, + prOptions: null, + issue: null, + isolation: null, + worktree: null, + agentStates: null, + }; + + const clustersFile = path.join(storageDir, 'clusters.json'); + fs.writeFileSync(clustersFile, JSON.stringify({ [clusterId]: clusterData }, null, 2)); + + const dbPath = path.join(storageDir, `${clusterId}.db`); + const ledger = new Ledger(dbPath); + ledger.append({ + topic: 'TEST', + sender: 'tester', + receiver: 'tester', + content_text: 'ok', + cluster_id: clusterId, + }); + ledger.close(); + + let orchestrator; + try { + orchestrator = await Orchestrator.create({ storageDir, quiet: true }); + const status = orchestrator.getStatus(clusterId); + const providers = status.agents.map((agent) => agent.provider); + for (const provider of providers) { + assert.strictEqual(provider, 'codex'); + } + } finally { + if (orchestrator) { + orchestrator.close(); + } + cleanupDir(storageDir); + cleanupDir(settingsDir); + } + }); +}); diff --git a/tests/unit/output-extraction-fatal-strings.test.js b/tests/unit/output-extraction-fatal-strings.test.js new file mode 100644 index 00000000..ae790a86 --- /dev/null +++ b/tests/unit/output-extraction-fatal-strings.test.js @@ -0,0 +1,30 @@ +const assert = require('assert'); +const { + extractJsonFromOutput, + hasFatalStandaloneOutput, +} = require('../../src/agent/output-extraction'); + +describe('Output Extraction - fatal strings handling', () => { + it('extracts JSON when fatal substrings appear inside JSON events', () => { + const output = [ + '{"type":"item.completed","item":{"id":"item_1","type":"command_execution","aggregated_output":"Task not found"}}', + '{"type":"item.completed","item":{"id":"item_2","type":"agent_message","text":"{\\"summary\\":\\"ok\\",\\"completionStatus\\":{\\"canValidate\\":true,\\"percentComplete\\":100}}"}}', + '{"type":"turn.completed","usage":{"input_tokens":1,"output_tokens":1}}', + ].join('\n'); + + const parsed = extractJsonFromOutput(output, 'codex'); + + assert.deepStrictEqual(parsed, { + summary: 'ok', + completionStatus: { + canValidate: true, + percentComplete: 100, + }, + }); + }); + + it('detects standalone fatal output lines', () => { + const output = 'Task not found\n'; + assert.strictEqual(hasFatalStandaloneOutput(output), true); + }); +}); diff --git a/tests/unit/provider-retryable-errors.test.js b/tests/unit/provider-retryable-errors.test.js new file mode 100644 index 00000000..52c83a82 --- /dev/null +++ b/tests/unit/provider-retryable-errors.test.js @@ -0,0 +1,35 @@ +const assert = require('assert'); +const { getProvider } = require('../../src/providers'); + +describe('Provider error retry classification', () => { + it('retries on rate limit (all providers)', () => { + const err = new Error('Rate limit exceeded. Retry after 60 seconds.'); + for (const name of ['claude', 'codex', 'gemini']) { + const provider = getProvider(name); + assert.strictEqual(provider.isRetryableError(err), true, `${name} should retry rate limit`); + } + }); + + it('does not retry on invalid_api_key (all providers)', () => { + const err = new Error('invalid_api_key: key revoked'); + for (const name of ['claude', 'codex', 'gemini']) { + const provider = getProvider(name); + assert.strictEqual(provider.isRetryableError(err), false, `${name} should not retry auth`); + } + }); + + it('Claude: retries on "No messages returned"', () => { + const provider = getProvider('claude'); + assert.strictEqual(provider.isRetryableError(new Error('No messages returned')), true); + }); + + it('Codex: retries on "server_error"', () => { + const provider = getProvider('codex'); + assert.strictEqual(provider.isRetryableError(new Error('server_error')), true); + }); + + it('Gemini: retries on "RESOURCE_EXHAUSTED"', () => { + const provider = getProvider('gemini'); + assert.strictEqual(provider.isRetryableError(new Error('RESOURCE_EXHAUSTED')), true); + }); +}); diff --git a/tests/unit/rate-limit-backoff.test.js b/tests/unit/rate-limit-backoff.test.js new file mode 100644 index 00000000..b285399d --- /dev/null +++ b/tests/unit/rate-limit-backoff.test.js @@ -0,0 +1,103 @@ +const assert = require('assert'); +const { + calculateRateLimitDelay, + isRateLimitError, + parseRetryAfter, +} = require('../../src/agent/rate-limit-backoff'); + +describe('rate-limit-backoff', () => { + describe('isRateLimitError', () => { + it('detects HTTP 429 errors', () => { + assert.strictEqual(isRateLimitError(new Error('HTTP 429: Rate limit exceeded')), true); + assert.strictEqual(isRateLimitError(new Error('Error 429 - Too many requests')), true); + }); + + it('detects "rate limit" text', () => { + assert.strictEqual(isRateLimitError(new Error('Rate limit exceeded')), true); + assert.strictEqual(isRateLimitError(new Error('rate-limit error')), true); + }); + + it('detects Gemini "No capacity available" errors', () => { + assert.strictEqual( + isRateLimitError(new Error('No capacity available for model gemini-3-pro')), + true + ); + }); + + it('detects quota exceeded errors', () => { + assert.strictEqual(isRateLimitError(new Error('Quota exceeded for this project')), true); + }); + + it('detects resource exhausted errors', () => { + assert.strictEqual(isRateLimitError(new Error('RESOURCE_EXHAUSTED: No capacity')), true); + }); + + it('returns false for non-rate-limit errors', () => { + assert.strictEqual(isRateLimitError(new Error('Network timeout')), false); + assert.strictEqual(isRateLimitError(new Error('No messages returned')), false); + assert.strictEqual(isRateLimitError(new Error('SIGTERM')), false); + }); + + it('handles null/undefined', () => { + assert.strictEqual(isRateLimitError(null), false); + assert.strictEqual(isRateLimitError(undefined), false); + }); + }); + + describe('parseRetryAfter', () => { + it('parses "Retry-After: N" header format', () => { + assert.strictEqual(parseRetryAfter(new Error('Rate limit. Retry-After: 120')), 120); + assert.strictEqual(parseRetryAfter(new Error('Retry-After:60')), 60); + }); + + it('returns null when not found', () => { + assert.strictEqual(parseRetryAfter(new Error('Rate limit exceeded')), null); + assert.strictEqual(parseRetryAfter(null), null); + }); + }); + + describe('calculateRateLimitDelay', () => { + let originalRandom; + beforeEach(() => { + originalRandom = Math.random; + Math.random = () => 0.5; // Neutralizes jitter + }); + afterEach(() => { + Math.random = originalRandom; + }); + + it('uses 30s base for rate limit errors', () => { + const error = new Error('HTTP 429: Rate limit'); + const delay = calculateRateLimitDelay(error, 1, {}); + assert.strictEqual(delay, 30000); + }); + + it('uses exponential backoff for rate limit errors', () => { + const error = new Error('HTTP 429: Rate limit'); + assert.strictEqual(calculateRateLimitDelay(error, 1, {}), 30000); + assert.strictEqual(calculateRateLimitDelay(error, 2, {}), 60000); + assert.strictEqual(calculateRateLimitDelay(error, 3, {}), 120000); + }); + + it('caps rate limit delays at 5 minutes', () => { + const error = new Error('HTTP 429: Rate limit'); + assert.strictEqual(calculateRateLimitDelay(error, 5, {}), 300000); + }); + + it('uses 2s base for non-rate-limit errors', () => { + const error = new Error('No messages returned'); + const delay = calculateRateLimitDelay(error, 1, {}); + assert.strictEqual(delay, 2000); + }); + + it('honors Retry-After header', () => { + const error = new Error('Rate limit. Retry-After: 120'); + assert.strictEqual(calculateRateLimitDelay(error, 1, {}), 120000); + }); + + it('caps Retry-After at 5 minutes', () => { + const error = new Error('Rate limit. Retry-After: 600'); + assert.strictEqual(calculateRateLimitDelay(error, 1, {}), 300000); + }); + }); +}); diff --git a/tests/unit/stream-json-parser-codex.test.js b/tests/unit/stream-json-parser-codex.test.js new file mode 100644 index 00000000..3bcea16d --- /dev/null +++ b/tests/unit/stream-json-parser-codex.test.js @@ -0,0 +1,51 @@ +const assert = require('assert'); +const { parseChunk } = require('../../lib/stream-json-parser'); + +describe('stream-json parser (Codex)', () => { + it('maps command_execution start/completed into tool_call/tool_result', () => { + const chunk = [ + JSON.stringify({ + type: 'item.started', + item: { id: 'item_1', type: 'command_execution', command: 'ls -la' }, + }), + JSON.stringify({ + type: 'item.completed', + item: { + id: 'item_1', + type: 'command_execution', + aggregated_output: 'file1.txt\nfile2.txt\n', + exit_code: 0, + }, + }), + ].join('\n'); + + const events = parseChunk(chunk); + assert.deepStrictEqual(events[0], { + type: 'tool_call', + toolName: 'Bash', + toolId: 'item_1', + input: { command: 'ls -la' }, + }); + assert.deepStrictEqual(events[1], { + type: 'tool_result', + toolId: 'item_1', + content: 'file1.txt\nfile2.txt\n', + isError: false, + }); + }); + + it('maps reasoning items into thinking', () => { + const chunk = JSON.stringify({ + type: 'item.completed', + item: { id: 'r1', type: 'reasoning', text: 'thinking...' }, + }); + const events = parseChunk(chunk); + assert.deepStrictEqual(events, [{ type: 'thinking', text: 'thinking...' }]); + }); + + it('maps top-level errors into result errors', () => { + const chunk = JSON.stringify({ type: 'error', error: { message: 'boom' } }); + const events = parseChunk(chunk); + assert.deepStrictEqual(events, [{ type: 'result', success: false, error: 'boom' }]); + }); +}); diff --git a/tests/unit/template-simulation.test.js b/tests/unit/template-simulation.test.js new file mode 100644 index 00000000..aa0dd358 --- /dev/null +++ b/tests/unit/template-simulation.test.js @@ -0,0 +1,84 @@ +const assert = require('node:assert'); + +const { + simulateConsensusGates, +} = require('../../src/template-validation/simulate-consensus-gates'); + +describe('Template micro-simulation (consensus gates)', function () { + it('flags consensus gates that fire on duplicate sender', function () { + const config = { + name: 'Bad consensus template', + agents: [ + { + id: 'validator-a', + role: 'validator', + triggers: [{ topic: 'START', action: 'execute_task' }], + hooks: { onComplete: { action: 'publish_message', config: { topic: 'X' } } }, + }, + { + id: 'validator-b', + role: 'validator', + triggers: [{ topic: 'START', action: 'execute_task' }], + hooks: { onComplete: { action: 'publish_message', config: { topic: 'X' } } }, + }, + { + id: 'consensus-coordinator', + role: 'coordinator', + triggers: [ + { + topic: 'X', + logic: { + engine: 'javascript', + // BUGGY: counts messages, doesn't require distinct senders. + script: + "const results = ledger.query({ topic: 'X' }); return results.length === 2;", + }, + action: 'execute_task', + }, + ], + }, + ], + }; + + const failures = simulateConsensusGates(config); + assert.ok(failures.length >= 1); + assert.ok(failures.some((f) => f.includes('fires early'))); + }); + + it('accepts consensus gates that require distinct producers via helpers.allResponded', function () { + const config = { + name: 'Good consensus template', + agents: [ + { + id: 'validator-a', + role: 'validator', + triggers: [{ topic: 'START', action: 'execute_task' }], + hooks: { onComplete: { action: 'publish_message', config: { topic: 'X' } } }, + }, + { + id: 'validator-b', + role: 'validator', + triggers: [{ topic: 'START', action: 'execute_task' }], + hooks: { onComplete: { action: 'publish_message', config: { topic: 'X' } } }, + }, + { + id: 'consensus-coordinator', + role: 'coordinator', + triggers: [ + { + topic: 'X', + logic: { + engine: 'javascript', + script: "return helpers.allResponded(['validator-a','validator-b'], 'X', 0);", + }, + action: 'execute_task', + }, + ], + }, + ], + }; + + const failures = simulateConsensusGates(config); + assert.deepStrictEqual(failures, []); + }); +}); diff --git a/tests/unit/template-validation-deep.test.js b/tests/unit/template-validation-deep.test.js new file mode 100644 index 00000000..30a095ba --- /dev/null +++ b/tests/unit/template-validation-deep.test.js @@ -0,0 +1,14 @@ +const assert = require('node:assert'); +const path = require('node:path'); + +const { validateTemplates } = require('../../src/template-validation'); + +describe('Template validation (deep)', function () { + this.timeout(10000); + + it('passes deep sim for base templates', async function () { + const templatesDir = path.join(__dirname, '..', '..', 'cluster-templates', 'base-templates'); + const report = await validateTemplates({ templatesDir, deep: true }); + assert.strictEqual(report.valid, true); + }); +}); diff --git a/tests/unit/tui-backend-cluster-launcher.test.js b/tests/unit/tui-backend-cluster-launcher.test.js new file mode 100644 index 00000000..78c14a59 --- /dev/null +++ b/tests/unit/tui-backend-cluster-launcher.test.js @@ -0,0 +1,91 @@ +const assert = require('assert'); +const fs = require('fs'); +const path = require('path'); +const { execSync } = require('child_process'); + +const buildOutput = path.join( + __dirname, + '..', + '..', + 'lib', + 'tui-backend', + 'services', + 'cluster-launcher.js' +); +const sourcePath = path.join( + __dirname, + '..', + '..', + 'src', + 'tui-backend', + 'services', + 'cluster-launcher.ts' +); + +function ensureBackendBuild() { + if (!fs.existsSync(buildOutput)) { + execSync('npm run build:tui-backend', { stdio: 'inherit' }); + return; + } + if (fs.existsSync(sourcePath)) { + const buildMtime = fs.statSync(buildOutput).mtimeMs; + const sourceMtime = fs.statSync(sourcePath).mtimeMs; + if (sourceMtime > buildMtime) { + execSync('npm run build:tui-backend', { stdio: 'inherit' }); + } + } +} + +ensureBackendBuild(); + +const { + launchClusterFromIssue, + InvalidIssueReferenceError, +} = require('../../lib/tui-backend/services/cluster-launcher'); + +describe('tui-backend cluster launcher', function () { + it('throws InvalidIssueReferenceError for invalid issue refs', async function () { + await assert.rejects( + () => + launchClusterFromIssue({ + ref: 'not-an-issue', + deps: { + detectRunInput: () => ({ text: 'not-an-issue' }), + }, + }), + (error) => { + assert.ok(error instanceof InvalidIssueReferenceError); + assert.ok(error.message.includes('Invalid issue reference: not-an-issue')); + return true; + } + ); + }); + + it('forwards providerOverride and clusterId to startClusterFromIssue', async function () { + const calls = []; + const deps = { + getOrchestrator: () => ({ id: 'orch' }), + loadSettings: () => ({ defaultConfig: 'conductor-bootstrap', providerSettings: {} }), + resolveConfigPath: () => '/tmp/config.json', + loadClusterConfig: () => ({ name: 'config' }), + detectRunInput: () => ({ issue: '123' }), + startClusterFromIssue: (args) => { + calls.push(args); + }, + generateClusterId: () => 'generated', + }; + + const result = await launchClusterFromIssue({ + ref: '123', + providerOverride: 'codex', + clusterId: 'cluster-789', + deps, + }); + + assert.deepStrictEqual(result, { clusterId: 'cluster-789' }); + assert.strictEqual(calls.length, 1); + assert.strictEqual(calls[0].issue, '123'); + assert.strictEqual(calls[0].providerOverride, 'codex'); + assert.strictEqual(calls[0].clusterId, 'cluster-789'); + }); +}); diff --git a/tests/unit/tui-backend-cluster-registry.test.js b/tests/unit/tui-backend-cluster-registry.test.js new file mode 100644 index 00000000..87b6f1f6 --- /dev/null +++ b/tests/unit/tui-backend-cluster-registry.test.js @@ -0,0 +1,113 @@ +const assert = require('assert'); +const fs = require('fs'); +const path = require('path'); +const { execSync } = require('child_process'); + +const buildOutput = path.join( + __dirname, + '..', + '..', + 'lib', + 'tui-backend', + 'services', + 'cluster-registry.js' +); +const sourcePath = path.join( + __dirname, + '..', + '..', + 'src', + 'tui-backend', + 'services', + 'cluster-registry.ts' +); + +function ensureBackendBuild() { + if (!fs.existsSync(buildOutput)) { + execSync('npm run build:tui-backend', { stdio: 'inherit' }); + return; + } + if (fs.existsSync(sourcePath)) { + const buildMtime = fs.statSync(buildOutput).mtimeMs; + const sourceMtime = fs.statSync(sourcePath).mtimeMs; + if (sourceMtime > buildMtime) { + execSync('npm run build:tui-backend', { stdio: 'inherit' }); + } + } +} + +ensureBackendBuild(); + +const { + listClusters, + getClusterSummary, + ClusterNotFoundError, +} = require('../../lib/tui-backend/services/cluster-registry'); + +function createOrchestrator(summaries, clustersById) { + return { + listClusters() { + return summaries; + }, + getCluster(id) { + return clustersById[id]; + }, + }; +} + +describe('tui-backend cluster registry', function () { + it('resolves provider with forceProvider -> defaultProvider -> settings default', async function () { + const summaries = [ + { id: 'cluster-force', state: 'running', createdAt: 1, agentCount: 1, messageCount: 1 }, + { id: 'cluster-default', state: 'running', createdAt: 2, agentCount: 1, messageCount: 1 }, + { id: 'cluster-settings', state: 'running', createdAt: 3, agentCount: 1, messageCount: 1 }, + { id: 'cluster-empty', state: 'running', createdAt: 4, agentCount: 1, messageCount: 1 }, + ]; + const clustersById = { + 'cluster-force': { config: { forceProvider: 'openai', defaultProvider: 'claude' } }, + 'cluster-default': { config: { defaultProvider: 'google' } }, + 'cluster-settings': { config: {} }, + 'cluster-empty': {}, + }; + const orchestrator = createOrchestrator(summaries, clustersById); + + const result = await listClusters({ + deps: { + getOrchestrator: () => Promise.resolve(orchestrator), + loadSettings: () => ({ defaultProvider: 'opencode' }), + }, + }); + + const providerById = Object.fromEntries( + result.map((cluster) => [cluster.id, cluster.provider]) + ); + assert.strictEqual(providerById['cluster-force'], 'codex'); + assert.strictEqual(providerById['cluster-default'], 'gemini'); + assert.strictEqual(providerById['cluster-settings'], 'opencode'); + assert.strictEqual(providerById['cluster-empty'], 'opencode'); + }); + + it('throws ClusterNotFoundError for missing cluster id', async function () { + const summaries = [ + { id: 'cluster-1', state: 'running', createdAt: 1, agentCount: 1, messageCount: 1 }, + ]; + const clustersById = { + 'cluster-1': { config: { defaultProvider: 'claude' } }, + }; + const orchestrator = createOrchestrator(summaries, clustersById); + + try { + await getClusterSummary({ + clusterId: 'missing-cluster', + deps: { + getOrchestrator: () => Promise.resolve(orchestrator), + loadSettings: () => ({ defaultProvider: 'claude' }), + }, + }); + assert.fail('Expected ClusterNotFoundError'); + } catch (error) { + assert.ok(error instanceof ClusterNotFoundError); + assert.strictEqual(error.clusterId, 'missing-cluster'); + } + }); +}); diff --git a/tests/unit/tui-backend-smoke.test.js b/tests/unit/tui-backend-smoke.test.js new file mode 100644 index 00000000..5e90615f --- /dev/null +++ b/tests/unit/tui-backend-smoke.test.js @@ -0,0 +1,49 @@ +const assert = require('assert'); +const fs = require('fs'); +const path = require('path'); +const { execSync } = require('child_process'); + +const buildOutput = path.join( + __dirname, + '..', + '..', + 'lib', + 'tui-backend', + 'services', + 'cluster-registry.js' +); +const sourcePath = path.join( + __dirname, + '..', + '..', + 'src', + 'tui-backend', + 'services', + 'cluster-registry.ts' +); + +function ensureBackendBuild() { + if (!fs.existsSync(buildOutput)) { + execSync('npm run build:tui-backend', { stdio: 'inherit' }); + return; + } + if (fs.existsSync(sourcePath)) { + const buildMtime = fs.statSync(buildOutput).mtimeMs; + const sourceMtime = fs.statSync(sourcePath).mtimeMs; + if (sourceMtime > buildMtime) { + execSync('npm run build:tui-backend', { stdio: 'inherit' }); + } + } +} + +ensureBackendBuild(); + +describe('TUI backend build', function () { + it('exposes cluster registry services', function () { + const registry = require('../../lib/tui-backend/services/cluster-registry'); + assert.ok(registry); + assert.strictEqual(typeof registry.listClusters, 'function'); + assert.strictEqual(typeof registry.getClusterSummary, 'function'); + assert.strictEqual(typeof registry.listClusterMetrics, 'function'); + }); +}); diff --git a/tests/unit/tui-backend-stdio.test.js b/tests/unit/tui-backend-stdio.test.js new file mode 100644 index 00000000..ce75f573 --- /dev/null +++ b/tests/unit/tui-backend-stdio.test.js @@ -0,0 +1,707 @@ +const assert = require('assert'); +const fs = require('fs'); +const os = require('os'); +const path = require('path'); +const { execSync, spawn } = require('child_process'); +const Ledger = require('../../src/ledger'); + +const PROJECT_ROOT = path.resolve(__dirname, '../..'); +const SERVER_PATH = path.join(PROJECT_ROOT, 'lib', 'tui-backend', 'server.js'); +const SERVER_SOURCE_PATH = path.join(PROJECT_ROOT, 'src', 'tui-backend', 'server.ts'); +const CLUSTER_LOGS_SOURCE_PATH = path.join( + PROJECT_ROOT, + 'src', + 'tui-backend', + 'services', + 'cluster-logs.ts' +); +const CLUSTER_LOGS_BUILD_PATH = path.join( + PROJECT_ROOT, + 'lib', + 'tui-backend', + 'services', + 'cluster-logs.js' +); + +function isBuildStale(sourcePath, buildPath) { + if (!fs.existsSync(buildPath)) { + return true; + } + if (!fs.existsSync(sourcePath)) { + return false; + } + return fs.statSync(sourcePath).mtimeMs > fs.statSync(buildPath).mtimeMs; +} + +const encodeFrame = (payload) => { + const body = Buffer.from(typeof payload === 'string' ? payload : JSON.stringify(payload), 'utf8'); + const header = Buffer.from(`Content-Length: ${body.length}\r\n\r\n`, 'utf8'); + return Buffer.concat([header, body]); +}; + +const createFrameCollector = () => { + let buffer = Buffer.alloc(0); + return (chunk) => { + buffer = Buffer.concat([buffer, chunk]); + const frames = []; + while (true) { + const headerIndex = buffer.indexOf('\r\n\r\n'); + if (headerIndex === -1) { + break; + } + const headerText = buffer.slice(0, headerIndex).toString('utf8'); + const match = headerText.match(/Content-Length:\s*(\d+)/i); + if (!match) { + throw new Error('Missing Content-Length header in response'); + } + const length = Number.parseInt(match[1], 10); + const totalLength = headerIndex + 4 + length; + if (buffer.length < totalLength) { + break; + } + const payload = buffer.slice(headerIndex + 4, totalLength).toString('utf8'); + frames.push(payload); + buffer = buffer.slice(totalLength); + } + return frames; + }; +}; + +const createMessageQueue = () => { + const queue = []; + const waiters = []; + return { + push(message) { + if (waiters.length) { + const waiter = waiters.shift(); + clearTimeout(waiter.timer); + waiter.resolve(message); + return; + } + queue.push(message); + }, + next(timeoutMs = 2000) { + if (queue.length) { + return Promise.resolve(queue.shift()); + } + return new Promise((resolve, reject) => { + const timer = setTimeout(() => { + const index = waiters.findIndex((waiter) => waiter.resolve === resolve); + if (index !== -1) { + waiters.splice(index, 1); + } + reject(new Error('Timed out waiting for response')); + }, timeoutMs); + waiters.push({ resolve, reject, timer }); + }); + }, + }; +}; + +describe('tui-backend stdio JSON-RPC', function () { + this.timeout(15000); + + let server; + let queue; + const originalHome = process.env.HOME; + let tempHome; + let MAX_LOG_LINES; + const topologyClusterId = 'cluster-stdio-topology'; + const metricsClusterId = 'cluster-stdio-metrics'; + + const waitForMessage = async (predicate, timeoutMs = 2000) => { + const deadline = Date.now() + timeoutMs; + while (Date.now() < deadline) { + const remaining = Math.max(deadline - Date.now(), 10); + try { + const message = await queue.next(remaining); + if (predicate(message)) { + return message; + } + } catch (error) { + if (Date.now() >= deadline) { + throw error; + } + } + } + throw new Error('Timed out waiting for matching response'); + }; + + const expectNoMessage = async (predicate, timeoutMs = 300) => { + const deadline = Date.now() + timeoutMs; + while (Date.now() < deadline) { + const remaining = Math.max(deadline - Date.now(), 10); + try { + const message = await queue.next(remaining); + if (predicate(message)) { + assert.fail(`Unexpected message: ${JSON.stringify(message)}`); + } + } catch (error) { + if (error && String(error.message).includes('Timed out waiting for response')) { + return; + } + throw error; + } + } + }; + + before(function () { + tempHome = fs.mkdtempSync(path.join(os.tmpdir(), 'zeroshot-tui-backend-')); + process.env.HOME = tempHome; + const zeroshotDir = path.join(tempHome, '.zeroshot'); + fs.mkdirSync(zeroshotDir, { recursive: true }); + + const clustersFile = path.join(zeroshotDir, 'clusters.json'); + const now = Date.now(); + const baseConfig = { + agents: [ + { + id: 'worker', + role: 'implementation', + triggers: [{ topic: 'ISSUE_OPENED', action: 'execute_task' }], + hooks: { onComplete: { config: { topic: 'IMPLEMENTATION_READY' } } }, + }, + { + id: 'validator', + role: 'validator', + triggers: [{ topic: 'IMPLEMENTATION_READY', action: 'execute_task' }], + }, + ], + }; + const clustersData = { + [topologyClusterId]: { + id: topologyClusterId, + config: baseConfig, + state: 'stopped', + createdAt: now - 1000, + pid: null, + }, + [metricsClusterId]: { + id: metricsClusterId, + config: baseConfig, + state: 'stopped', + createdAt: now, + pid: null, + }, + }; + fs.writeFileSync(clustersFile, JSON.stringify(clustersData, null, 2)); + + for (const clusterId of [topologyClusterId, metricsClusterId]) { + const dbPath = path.join(zeroshotDir, `${clusterId}.db`); + const ledger = new Ledger(dbPath); + ledger.append({ + cluster_id: clusterId, + topic: 'SYSTEM', + sender: 'test', + content: { text: `seed ${clusterId}`, data: { line: `seed ${clusterId}` } }, + }); + ledger.close(); + } + + if ( + isBuildStale(SERVER_SOURCE_PATH, SERVER_PATH) || + isBuildStale(CLUSTER_LOGS_SOURCE_PATH, CLUSTER_LOGS_BUILD_PATH) + ) { + execSync('npm run build:tui-backend', { cwd: PROJECT_ROOT, stdio: 'inherit' }); + } + + ({ MAX_LOG_LINES } = require('../../lib/tui-backend/services/cluster-logs')); + + server = spawn('node', [SERVER_PATH], { + cwd: PROJECT_ROOT, + stdio: ['pipe', 'pipe', 'pipe'], + env: { + ...process.env, + ZEROSHOT_TUI_BACKEND_MOCK_LAUNCH: '1', + ZEROSHOT_TUI_BACKEND_MOCK_GUIDANCE: '1', + ZEROSHOT_TUI_BACKEND_METRICS_PLATFORM: 'sunos', + HOME: tempHome, + }, + }); + + queue = createMessageQueue(); + const collectFrames = createFrameCollector(); + + server.stdout.on('data', (chunk) => { + const frames = collectFrames(chunk); + for (const frame of frames) { + queue.push(JSON.parse(frame)); + } + }); + + server.stderr.on('data', () => {}); + }); + + after(async function () { + if (!server) return; + server.stdin.end(); + await new Promise((resolve) => server.on('exit', resolve)); + process.env.HOME = originalHome; + fs.rmSync(tempHome, { recursive: true, force: true }); + }); + + it('responds to initialize with capabilities', async function () { + const request = { + jsonrpc: '2.0', + id: 1, + method: 'initialize', + params: { + protocolVersion: 1, + client: { name: 'test-client', version: '0.1.0' }, + }, + }; + + server.stdin.write(encodeFrame(request)); + const response = await queue.next(); + + assert.strictEqual(response.id, 1); + assert.strictEqual(response.jsonrpc, '2.0'); + assert.strictEqual(response.result.protocolVersion, 1); + assert.ok(response.result.server.name); + assert.ok(response.result.server.version); + assert.ok(response.result.capabilities.methods.includes('initialize')); + assert.ok(response.result.capabilities.methods.includes('ping')); + assert.ok(response.result.capabilities.methods.includes('listClusters')); + assert.ok(response.result.capabilities.methods.includes('getClusterSummary')); + assert.ok(response.result.capabilities.methods.includes('listClusterMetrics')); + assert.ok(response.result.capabilities.methods.includes('getClusterTopology')); + assert.ok(response.result.capabilities.methods.includes('unsubscribe')); + assert.deepStrictEqual(response.result.capabilities.notifications, [ + 'clusterLogLines', + 'clusterTimelineEvents', + ]); + }); + + it('responds to ping', async function () { + const request = { + jsonrpc: '2.0', + id: 2, + method: 'ping', + params: {}, + }; + + server.stdin.write(encodeFrame(request)); + const response = await queue.next(); + + assert.strictEqual(response.id, 2); + assert.deepStrictEqual(response.result, { ok: true }); + }); + + it('responds to listClusters with provider and cwd fields', async function () { + const request = { + jsonrpc: '2.0', + id: 7, + method: 'listClusters', + params: {}, + }; + + server.stdin.write(encodeFrame(request)); + const response = await queue.next(); + + assert.strictEqual(response.id, 7); + assert.strictEqual(response.jsonrpc, '2.0'); + assert.ok(response.result); + assert.ok(Array.isArray(response.result.clusters)); + for (const cluster of response.result.clusters) { + assert.ok(Object.prototype.hasOwnProperty.call(cluster, 'provider')); + assert.ok(Object.prototype.hasOwnProperty.call(cluster, 'cwd')); + } + }); + + it('returns cluster not found for unknown cluster id', async function () { + const request = { + jsonrpc: '2.0', + id: 8, + method: 'getClusterSummary', + params: { clusterId: 'missing-cluster-stdio' }, + }; + + server.stdin.write(encodeFrame(request)); + const response = await queue.next(); + + assert.strictEqual(response.id, 8); + assert.strictEqual(response.error.code, -32002); + }); + + it('responds to getClusterSummary for a known cluster when available', async function () { + const listRequest = { + jsonrpc: '2.0', + id: 9, + method: 'listClusters', + params: {}, + }; + + server.stdin.write(encodeFrame(listRequest)); + const listResponse = await queue.next(); + const clusters = listResponse.result?.clusters ?? []; + if (clusters.length === 0) { + return; + } + + const request = { + jsonrpc: '2.0', + id: 10, + method: 'getClusterSummary', + params: { clusterId: clusters[0].id }, + }; + + server.stdin.write(encodeFrame(request)); + const response = await queue.next(); + + assert.strictEqual(response.id, 10); + assert.strictEqual(response.result.summary.id, clusters[0].id); + assert.ok(Object.prototype.hasOwnProperty.call(response.result.summary, 'provider')); + assert.ok(Object.prototype.hasOwnProperty.call(response.result.summary, 'cwd')); + }); + + it('responds to getClusterTopology for a seeded cluster', async function () { + const request = { + jsonrpc: '2.0', + id: 17, + method: 'getClusterTopology', + params: { clusterId: topologyClusterId }, + }; + + server.stdin.write(encodeFrame(request)); + const response = await queue.next(); + + assert.strictEqual(response.id, 17); + assert.ok(response.result.topology); + assert.ok(Array.isArray(response.result.topology.agents)); + assert.ok(Array.isArray(response.result.topology.edges)); + assert.ok(Array.isArray(response.result.topology.topics)); + assert.ok(response.result.topology.topics.includes('ISSUE_OPENED')); + assert.ok(response.result.topology.topics.includes('IMPLEMENTATION_READY')); + assert.ok( + response.result.topology.edges.some( + (edge) => edge.from === 'system' && edge.to === 'ISSUE_OPENED' && edge.kind === 'source' + ) + ); + assert.ok( + response.result.topology.edges.some( + (edge) => edge.from === 'ISSUE_OPENED' && edge.to === 'worker' + ) + ); + assert.ok( + response.result.topology.edges.some( + (edge) => edge.from === 'worker' && edge.to === 'IMPLEMENTATION_READY' + ) + ); + }); + + it('responds to listClusterMetrics with filtered cluster ids', async function () { + const request = { + jsonrpc: '2.0', + id: 18, + method: 'listClusterMetrics', + params: { + clusterIds: [metricsClusterId, 'missing-metrics', topologyClusterId], + }, + }; + + server.stdin.write(encodeFrame(request)); + const response = await queue.next(); + + assert.strictEqual(response.id, 18); + assert.ok(Array.isArray(response.result.metrics)); + assert.strictEqual(response.result.metrics.length, 2); + assert.strictEqual(response.result.metrics[0].id, metricsClusterId); + assert.strictEqual(response.result.metrics[1].id, topologyClusterId); + for (const metric of response.result.metrics) { + assert.strictEqual(metric.supported, false); + assert.strictEqual(metric.cpuPercent, null); + assert.strictEqual(metric.memoryMB, null); + } + }); + + it('responds to startClusterFromText with a cluster id', async function () { + const request = { + jsonrpc: '2.0', + id: 11, + method: 'startClusterFromText', + params: { text: 'Launch from text', clusterId: 'cluster-stdio' }, + }; + + server.stdin.write(encodeFrame(request)); + const response = await queue.next(); + + assert.strictEqual(response.id, 11); + assert.deepStrictEqual(response.result, { clusterId: 'cluster-stdio' }); + }); + + it('returns invalid params for invalid issue ref', async function () { + const request = { + jsonrpc: '2.0', + id: 12, + method: 'startClusterFromIssue', + params: { ref: 'not-an-issue' }, + }; + + server.stdin.write(encodeFrame(request)); + const response = await queue.next(); + + assert.strictEqual(response.id, 12); + assert.strictEqual(response.error.code, -32602); + assert.ok(response.error.data.detail.includes('Invalid issue reference:')); + }); + + it('responds to sendGuidanceToAgent with delivery details', async function () { + const request = { + jsonrpc: '2.0', + id: 13, + method: 'sendGuidanceToAgent', + params: { + clusterId: 'cluster-guidance', + agentId: 'agent-1', + text: 'Use approach A', + timeoutMs: 250, + }, + }; + + server.stdin.write(encodeFrame(request)); + const response = await queue.next(); + + assert.strictEqual(response.id, 13); + assert.strictEqual(response.result.result.status, 'injected'); + assert.strictEqual(response.result.result.reason, null); + assert.strictEqual(response.result.result.method, 'pty'); + assert.strictEqual(response.result.result.taskId, 'task-agent-1'); + }); + + it('responds to sendGuidanceToCluster with summary and agents', async function () { + const request = { + jsonrpc: '2.0', + id: 14, + method: 'sendGuidanceToCluster', + params: { + clusterId: 'cluster-guidance', + text: 'Use approach B', + timeoutMs: 500, + }, + }; + + server.stdin.write(encodeFrame(request)); + const response = await queue.next(); + + assert.strictEqual(response.id, 14); + assert.strictEqual(response.result.result.summary.total, 2); + assert.strictEqual(response.result.result.summary.injected, 1); + assert.strictEqual(response.result.result.summary.queued, 1); + assert.strictEqual(response.result.result.agents['mock-agent-1'].status, 'injected'); + assert.strictEqual(response.result.result.agents['mock-agent-2'].status, 'queued'); + assert.strictEqual(response.result.result.timestamp, 1700000000000); + }); + + it('returns invalid params for empty agent guidance text', async function () { + const request = { + jsonrpc: '2.0', + id: 15, + method: 'sendGuidanceToAgent', + params: { + clusterId: 'cluster-guidance', + agentId: 'agent-1', + text: ' ', + }, + }; + + server.stdin.write(encodeFrame(request)); + const response = await queue.next(); + + assert.strictEqual(response.id, 15); + assert.strictEqual(response.error.code, -32602); + }); + + it('returns invalid params for empty cluster guidance text', async function () { + const request = { + jsonrpc: '2.0', + id: 16, + method: 'sendGuidanceToCluster', + params: { + clusterId: 'cluster-guidance', + text: '', + }, + }; + + server.stdin.write(encodeFrame(request)); + const response = await queue.next(); + + assert.strictEqual(response.id, 16); + assert.strictEqual(response.error.code, -32602); + }); + + it('returns parse error for invalid JSON', async function () { + server.stdin.write(encodeFrame('{ not-json')); + const response = await queue.next(); + + assert.strictEqual(response.id, null); + assert.strictEqual(response.error.code, -32700); + }); + + it('returns method not found for unknown method', async function () { + const request = { + jsonrpc: '2.0', + id: 3, + method: 'nope', + params: {}, + }; + + server.stdin.write(encodeFrame(request)); + const response = await queue.next(); + + assert.strictEqual(response.id, 3); + assert.strictEqual(response.error.code, -32601); + }); + + it('returns invalid params for malformed initialize', async function () { + const request = { + jsonrpc: '2.0', + id: 4, + method: 'initialize', + params: { protocolVersion: 1 }, + }; + + server.stdin.write(encodeFrame(request)); + const response = await queue.next(); + + assert.strictEqual(response.id, 4); + assert.strictEqual(response.error.code, -32602); + }); + + it('returns protocol mismatch with supported versions', async function () { + const request = { + jsonrpc: '2.0', + id: 5, + method: 'initialize', + params: { + protocolVersion: 999, + client: { name: 'test-client', version: '0.1.0' }, + }, + }; + + server.stdin.write(encodeFrame(request)); + const response = await queue.next(); + + assert.strictEqual(response.id, 5); + assert.strictEqual(response.error.code, -32000); + assert.deepStrictEqual(response.error.data.supportedVersions, [1]); + }); + + it('reassembles partial frames across chunks', async function () { + const request = { + jsonrpc: '2.0', + id: 6, + method: 'ping', + params: {}, + }; + const frame = encodeFrame(request); + const splitIndex = Math.floor(frame.length / 2); + server.stdin.write(frame.slice(0, splitIndex)); + server.stdin.write(frame.slice(splitIndex)); + + const response = await queue.next(); + assert.strictEqual(response.id, 6); + assert.deepStrictEqual(response.result, { ok: true }); + }); + + it('streams cluster logs and stops after unsubscribe', async function () { + const clusterId = 'cluster-stdio-logs'; + const zeroshotDir = path.join(tempHome, '.zeroshot'); + fs.mkdirSync(zeroshotDir, { recursive: true }); + + const dbPath = path.join(zeroshotDir, `${clusterId}.db`); + const clustersFile = path.join(zeroshotDir, 'clusters.json'); + fs.writeFileSync( + clustersFile, + JSON.stringify( + { + [clusterId]: { + config: { + dbPath, + }, + }, + }, + null, + 2 + ) + ); + + const seedLedger = new Ledger(dbPath); + seedLedger.close(); + + const subscribeRequest = { + jsonrpc: '2.0', + id: 20, + method: 'subscribeClusterLogs', + params: { clusterId }, + }; + + server.stdin.write(encodeFrame(subscribeRequest)); + const subscribeResponse = await waitForMessage((msg) => msg.id === 20); + const subscriptionId = subscribeResponse.result.subscriptionId; + + const writer = new Ledger(dbPath); + const payloadCount = MAX_LOG_LINES + 5; + const messages = Array.from({ length: payloadCount }, (_, index) => ({ + cluster_id: clusterId, + topic: 'AGENT_OUTPUT', + sender: 'worker', + content: { + text: `line ${index}`, + data: { + agent: 'worker', + role: 'implementation', + line: `line ${index}`, + }, + }, + })); + writer.batchAppend(messages); + writer.close(); + + const notification = await waitForMessage( + (msg) => + msg.method === 'clusterLogLines' && + msg.params && + msg.params.subscriptionId === subscriptionId, + 4000 + ); + + assert.strictEqual(notification.params.clusterId, clusterId); + assert.strictEqual(notification.params.lines.length, MAX_LOG_LINES); + assert.strictEqual(notification.params.droppedCount, 5); + + const unsubscribeRequest = { + jsonrpc: '2.0', + id: 21, + method: 'unsubscribe', + params: { subscriptionId }, + }; + + server.stdin.write(encodeFrame(unsubscribeRequest)); + const unsubscribeResponse = await waitForMessage((msg) => msg.id === 21); + assert.deepStrictEqual(unsubscribeResponse.result, { removed: true }); + + const writer2 = new Ledger(dbPath); + writer2.append({ + cluster_id: clusterId, + topic: 'AGENT_OUTPUT', + sender: 'worker', + content: { + text: 'after unsubscribe', + data: { + agent: 'worker', + role: 'implementation', + line: 'after unsubscribe', + }, + }, + }); + writer2.close(); + + await expectNoMessage( + (msg) => + msg.method === 'clusterLogLines' && + msg.params && + msg.params.subscriptionId === subscriptionId, + 600 + ); + }); +}); diff --git a/tests/unit/tui-backend-subscriptions.test.js b/tests/unit/tui-backend-subscriptions.test.js new file mode 100644 index 00000000..69a12f5d --- /dev/null +++ b/tests/unit/tui-backend-subscriptions.test.js @@ -0,0 +1,64 @@ +const assert = require('assert'); +const fs = require('fs'); +const path = require('path'); +const { execSync } = require('child_process'); + +const PROJECT_ROOT = path.resolve(__dirname, '../..'); +const SOURCE_PATH = path.join(PROJECT_ROOT, 'src', 'tui-backend', 'subscriptions', 'index.ts'); +const BUILD_PATH = path.join(PROJECT_ROOT, 'lib', 'tui-backend', 'subscriptions', 'index.js'); + +function isBuildStale(sourcePath, buildPath) { + if (!fs.existsSync(buildPath)) { + return true; + } + if (!fs.existsSync(sourcePath)) { + return false; + } + return fs.statSync(sourcePath).mtimeMs > fs.statSync(buildPath).mtimeMs; +} + +if (isBuildStale(SOURCE_PATH, BUILD_PATH)) { + execSync('npm run build:tui-backend', { cwd: PROJECT_ROOT, stdio: 'inherit' }); +} + +const { createSubscriptionRegistry } = require('../../lib/tui-backend/subscriptions'); + +describe('tui-backend subscription registry', function () { + it('closes subscriptions exactly once', function () { + const registry = createSubscriptionRegistry(); + let closes = 0; + + const id = registry.add('logs', () => { + closes += 1; + }); + + const first = registry.unsubscribe(id); + assert.deepStrictEqual(first, { removed: true }); + const second = registry.unsubscribe(id); + assert.deepStrictEqual(second, { removed: false }); + + assert.strictEqual(closes, 1); + assert.strictEqual(registry.size(), 0); + }); + + it('closeAll closes remaining subscriptions and clears registry', function () { + const registry = createSubscriptionRegistry(); + let closes = 0; + + registry.add('logs', () => { + closes += 1; + }); + registry.add('timeline', () => { + closes += 1; + }); + + const count = registry.closeAll(); + assert.strictEqual(count, 2); + assert.strictEqual(closes, 2); + assert.strictEqual(registry.size(), 0); + + const again = registry.closeAll(); + assert.strictEqual(again, 0); + assert.strictEqual(closes, 2); + }); +}); diff --git a/tests/unit/tui-backend/protocol-validation.test.js b/tests/unit/tui-backend/protocol-validation.test.js new file mode 100644 index 00000000..efac1f0b --- /dev/null +++ b/tests/unit/tui-backend/protocol-validation.test.js @@ -0,0 +1,83 @@ +const fs = require('fs'); +const path = require('path'); +const { expect } = require('chai'); + +require.extensions['.ts'] = require.extensions['.js']; + +const { + createValidator, + RPC_ERROR_CODES, +} = require('../../../src/tui-backend/protocol/validator.ts'); + +const fixturesDir = path.join(__dirname, '..', '..', 'fixtures', 'tui-v2', 'protocol'); + +const readFixture = (file) => { + const raw = fs.readFileSync(path.join(fixturesDir, file), 'utf8'); + return JSON.parse(raw); +}; + +const listFixtures = () => + fs + .readdirSync(fixturesDir) + .filter((file) => file.endsWith('.json')) + .sort(); + +const extractMethod = (file) => { + const parts = file.split('.'); + if (parts.length < 3) { + return null; + } + return parts[1]; +}; + +describe('tui-v2 protocol validation', () => { + const validator = createValidator(); + + it('accepts all valid request fixtures', () => { + for (const file of listFixtures()) { + if (!file.startsWith('request.')) continue; + const method = extractMethod(file); + const payload = readFixture(file); + const result = validator.validateRequest(payload); + expect(result.ok, `${file} failed: ${JSON.stringify(result.error)}`).to.equal(true); + expect(payload.method).to.equal(method); + } + }); + + it('accepts all valid response fixtures', () => { + for (const file of listFixtures()) { + if (!file.startsWith('response.')) continue; + const method = extractMethod(file); + const payload = readFixture(file); + const result = validator.validateResponse(payload, method); + expect(result.ok, `${file} failed: ${JSON.stringify(result.error)}`).to.equal(true); + } + }); + + it('accepts all valid notification fixtures', () => { + for (const file of listFixtures()) { + if (!file.startsWith('notification.')) continue; + const method = extractMethod(file); + const payload = readFixture(file); + const result = validator.validateNotification(payload); + expect(result.ok, `${file} failed: ${JSON.stringify(result.error)}`).to.equal(true); + expect(payload.method).to.equal(method); + } + }); + + it('rejects invalid fixtures with structured RPC errors', () => { + for (const file of listFixtures()) { + if (!file.startsWith('invalid.')) continue; + const payload = readFixture(file); + const result = validator.validateRequest(payload); + expect(result.ok, `${file} unexpectedly ok`).to.equal(false); + expect(result.error).to.be.an('object'); + + if (file.startsWith('invalid.request.')) { + expect(result.error.code).to.equal(RPC_ERROR_CODES.INVALID_REQUEST); + } else if (file.startsWith('invalid.params.')) { + expect(result.error.code).to.equal(RPC_ERROR_CODES.INVALID_PARAMS); + } + } + }); +}); diff --git a/tests/unit/tui-binary.test.js b/tests/unit/tui-binary.test.js new file mode 100644 index 00000000..239331dc --- /dev/null +++ b/tests/unit/tui-binary.test.js @@ -0,0 +1,82 @@ +/** + * Test: TUI Binary helpers + * + * Validates platform/arch mapping, asset naming, download URL building, + * and env overrides used by the install script and launcher. + */ + +const assert = require('assert'); +const fs = require('fs'); +const os = require('os'); +const path = require('path'); + +const { + ENV_BINARY_PATH, + ENV_BINARY_SKIP, + getAssetName, + getInstalledBinaryPath, + resolveBinaryPathOverride, + resolveDownloadUrl, + resolveTarget, + shouldSkipBinaryInstall, +} = require('../../lib/tui-binary'); + +describe('TUI Binary helpers', function () { + const originalEnv = { ...process.env }; + + afterEach(function () { + process.env = { ...originalEnv }; + }); + + it('builds asset names for supported targets', function () { + assert.strictEqual(getAssetName('darwin', 'arm64'), 'zeroshot-tui-darwin-arm64.tar.gz'); + assert.strictEqual(getAssetName('linux', 'x64'), 'zeroshot-tui-linux-x64.tar.gz'); + }); + + it('returns null for unsupported targets', function () { + assert.strictEqual(resolveTarget('win32', 'x64'), null); + assert.strictEqual(resolveTarget('darwin', 'ia32'), null); + }); + + it('builds release URLs with version overrides', function () { + const url = resolveDownloadUrl({ + version: '1.2.3', + platform: 'darwin', + arch: 'arm64', + }); + assert.strictEqual( + url, + 'https://github.com/covibes/zeroshot/releases/download/v1.2.3/zeroshot-tui-darwin-arm64.tar.gz' + ); + }); + + it('uses binary path override when provided', function () { + const tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'zeroshot-tui-test-')); + const tempBin = path.join(tempDir, 'zeroshot-tui'); + fs.writeFileSync(tempBin, 'binary'); + + process.env[ENV_BINARY_PATH] = tempBin; + + assert.strictEqual(resolveBinaryPathOverride(), tempBin); + }); + + it('throws when binary override path is missing', function () { + process.env[ENV_BINARY_PATH] = '/tmp/missing-zeroshot-tui'; + assert.throws(() => resolveBinaryPathOverride(), /Rust TUI binary not found/); + }); + + it('returns installed binary path', function () { + const expected = path.join(path.resolve(__dirname, '..', '..'), 'libexec', 'zeroshot-tui'); + assert.strictEqual(getInstalledBinaryPath(), expected); + }); + + it('honors skip env values', function () { + assert.strictEqual(shouldSkipBinaryInstall(), false); + + process.env[ENV_BINARY_SKIP] = '1'; + assert.strictEqual(shouldSkipBinaryInstall(), true); + + process.env[ENV_BINARY_SKIP] = 'false'; + assert.strictEqual(shouldSkipBinaryInstall(), false); + }); +}); diff --git a/tests/unit/tui-start-cluster.test.js b/tests/unit/tui-start-cluster.test.js new file mode 100644 index 00000000..cf5e1814 --- /dev/null +++ b/tests/unit/tui-start-cluster.test.js @@ -0,0 +1,82 @@ +const assert = require('assert'); + +const { startClusterFromText, startClusterFromIssue } = require('../../lib/start-cluster'); + +function createOrchestrator() { + const calls = { loadConfig: [], start: [] }; + return { + calls, + loadConfig(configPath) { + calls.loadConfig.push(configPath); + return { agents: [] }; + }, + start(config, input, options) { + calls.start.push({ config, input, options }); + return { id: 'cluster-test' }; + }, + }; +} + +describe('TUI start cluster helper', function () { + const originalCwd = process.env.ZEROSHOT_CWD; + + afterEach(function () { + if (originalCwd === undefined) { + delete process.env.ZEROSHOT_CWD; + } else { + process.env.ZEROSHOT_CWD = originalCwd; + } + }); + + it('startClusterFromText builds text input and forwards providerOverride', async function () { + process.env.ZEROSHOT_CWD = '/tmp'; + const orchestrator = createOrchestrator(); + const settings = { defaultProvider: 'claude', providerSettings: {} }; + const configPath = '/tmp/config.json'; + + const result = await startClusterFromText({ + orchestrator, + text: 'Launch cluster', + configPath, + settings, + providerOverride: 'codex', + modelOverride: 'gpt-4o', + forceProvider: 'github', + clusterId: 'cluster-123', + options: { docker: false, worktree: false, pr: false, mounts: false }, + }); + + assert.strictEqual(orchestrator.calls.loadConfig[0], configPath); + assert.strictEqual(orchestrator.calls.start.length, 1); + const call = orchestrator.calls.start[0]; + assert.deepStrictEqual(call.input, { text: 'Launch cluster' }); + assert.strictEqual(call.options.providerOverride, 'codex'); + assert.strictEqual(call.options.clusterId, 'cluster-123'); + assert.strictEqual(call.options.noMounts, true); + assert.strictEqual(result.id, 'cluster-test'); + }); + + it('startClusterFromIssue builds issue input and forwards providerOverride', async function () { + process.env.ZEROSHOT_CWD = '/tmp'; + const orchestrator = createOrchestrator(); + const settings = { defaultProvider: 'claude', providerSettings: {} }; + const configPath = '/tmp/config.json'; + + await startClusterFromIssue({ + orchestrator, + issue: '123', + configPath, + settings, + providerOverride: 'codex', + clusterId: 'cluster-456', + options: { docker: false, worktree: false, pr: false }, + }); + + assert.strictEqual(orchestrator.calls.loadConfig[0], configPath); + assert.strictEqual(orchestrator.calls.start.length, 1); + const call = orchestrator.calls.start[0]; + assert.deepStrictEqual(call.input, { issue: '123' }); + assert.strictEqual(call.options.providerOverride, 'codex'); + assert.strictEqual(call.options.clusterId, 'cluster-456'); + }); +}); diff --git a/tests/validation-platform.test.js b/tests/validation-platform.test.js new file mode 100644 index 00000000..d47d0723 --- /dev/null +++ b/tests/validation-platform.test.js @@ -0,0 +1,49 @@ +const assert = require('assert'); +const { + isPlatformMismatchReason, + findPlatformMismatchReason, +} = require('../src/agent/validation-platform'); + +describe('Validation platform mismatch detection', function () { + it('detects platform mismatch from reason strings', function () { + assert.ok(isPlatformMismatchReason('EBADPLATFORM @esbuild/linux-x64')); + assert.ok(isPlatformMismatchReason('Unsupported platform for @esbuild/linux-x64')); + assert.ok(isPlatformMismatchReason('darwin-arm64 vs linux-x64')); + assert.ok(!isPlatformMismatchReason('kubectl not installed')); + }); + + it('finds platform mismatch in criteriaResults', function () { + const result = { + criteriaResults: [ + { id: 'AC1', status: 'PASS' }, + { + id: 'AC2', + status: 'CANNOT_VALIDATE', + reason: 'npm install fails on darwin-arm64 (EBADPLATFORM for @esbuild/linux-x64)', + }, + ], + }; + + const reason = findPlatformMismatchReason(result); + assert.ok(reason, 'Should return a platform mismatch reason'); + assert.ok(reason.includes('EBADPLATFORM'), 'Should keep original reason'); + }); + + it('finds platform mismatch in errors array', function () { + const result = { + errors: ['EBADPLATFORM for @esbuild/linux-x64'], + }; + + const reason = findPlatformMismatchReason(result); + assert.ok(reason, 'Should return a platform mismatch reason'); + }); + + it('returns null when no mismatch found', function () { + const result = { + criteriaResults: [{ id: 'AC1', status: 'CANNOT_VALIDATE', reason: 'kubectl not installed' }], + errors: ['No SSH access'], + }; + + assert.strictEqual(findPlatformMismatchReason(result), null); + }); +}); diff --git a/tests/verify-github-pr-hook.test.js b/tests/verify-github-pr-hook.test.js new file mode 100644 index 00000000..0abecf52 --- /dev/null +++ b/tests/verify-github-pr-hook.test.js @@ -0,0 +1,327 @@ +/** + * verify_github_pr Hook Action Test Suite + * + * Tests for the verify_github_pr hook action that validates PR existence and merge status + * Part of issue #340 - Prevent git-pusher hallucination + */ + +const assert = require('assert'); +const path = require('path'); + +// Mock agent with required methods +function createMockAgent(workingDirectory = process.cwd()) { + return { + id: 'test-agent', + role: 'test', + workingDirectory, + _log: () => {}, + _publish: function (message) { + this.lastPublished = message; + }, + lastPublished: null, + }; +} + +describe('verify_github_pr hook action', function () { + this.timeout(60000); + + let executeHook; + let mockExecSyncFn; + + beforeEach(() => { + // Clear module cache + const hookExecutorPath = path.join(__dirname, '../src/agent/agent-hook-executor.js'); + delete require.cache[require.resolve(hookExecutorPath)]; + + const safeExecPath = path.join(__dirname, '../src/lib/safe-exec.js'); + delete require.cache[require.resolve(safeExecPath)]; + + // Mock safe-exec module + require.cache[require.resolve(safeExecPath)] = { + exports: { + execSync: function (...args) { + if (mockExecSyncFn) { + return mockExecSyncFn(...args); + } + throw new Error('Mock execSync not configured'); + }, + }, + }; + + // Reload executeHook with mocked safe-exec + executeHook = require('../src/agent/agent-hook-executor').executeHook; + mockExecSyncFn = null; + }); + + afterEach(() => { + mockExecSyncFn = null; + }); + + it('should not require pr_number in structured output', async function () { + const agent = createMockAgent(); + const hook = { action: 'verify_github_pr' }; + const result = { + output: JSON.stringify({ + summary: 'Merged', + result: + 'PR merged: {"pr_url":"https://github.com/org/repo/pull/123","pr_number":123,"merged":true}', + }), + }; + + mockExecSyncFn = () => { + return JSON.stringify({ + number: 123, + state: 'MERGED', + mergedAt: '2026-01-15T10:30:00Z', + url: 'https://github.com/org/repo/pull/123', + }); + }; + + await executeHook({ hook, agent, result }); + assert(agent.lastPublished, 'Expected message to be published'); + assert.strictEqual(agent.lastPublished.topic, 'CLUSTER_COMPLETE'); + }); + + it('should throw when PR does not exist in GitHub', async function () { + const agent = createMockAgent(); + const hook = { action: 'verify_github_pr' }; + const result = { + output: JSON.stringify({ + pr_url: 'https://github.com/org/repo/pull/9999', + merged: true, + }), + }; + + mockExecSyncFn = () => { + const error = new Error('Could not resolve to a PullRequest'); + error.status = 1; + throw error; + }; + + try { + await executeHook({ hook, agent, result }); + assert.fail('Expected error to be thrown'); + } catch (err) { + assert.match(err.message, /DOES NOT EXIST/); + assert.match(err.message, /HALLUCINATED/); + } + }); + + it('should throw when PR exists but genuinely not merged after all polls', async function () { + const agent = createMockAgent(); + const hook = { action: 'verify_github_pr' }; + const result = { + output: JSON.stringify({ + pr_url: 'https://github.com/org/repo/pull/123', + pr_number: 123, + merged: true, + }), + }; + + // Always returns OPEN ā€” genuinely not merged + mockExecSyncFn = () => { + return JSON.stringify({ + number: 123, + state: 'OPEN', + mergedAt: null, + url: 'https://github.com/org/repo/pull/123', + }); + }; + + try { + await executeHook({ hook, agent, result }); + assert.fail('Expected error to be thrown'); + } catch (err) { + assert.match(err.message, /LIED/i); + assert.match(err.message, /polls/i); + } + }); + + // REGRESSION: gentle-hydra-56 (2026-02-11) + // GitHub API returned state="OPEN" immediately after gh pr merge, but PR was actually merged. + // Old code had no merge propagation polling ā€” killed the cluster after 3s. + it('should succeed when GitHub API shows OPEN initially then MERGED after propagation delay', async function () { + const agent = createMockAgent(); + agent._log = () => {}; // suppress log noise + const hook = { action: 'verify_github_pr' }; + const result = { + output: JSON.stringify({ + pr_url: 'https://github.com/org/repo/pull/1411', + pr_number: 1411, + merged: true, + }), + }; + + // Simulate GitHub eventual consistency: OPEN for first 3 calls, then MERGED + let callCount = 0; + mockExecSyncFn = () => { + callCount++; + if (callCount <= 3) { + return JSON.stringify({ + number: 1411, + state: 'OPEN', + mergedAt: null, + url: 'https://github.com/org/repo/pull/1411', + }); + } + return JSON.stringify({ + number: 1411, + state: 'MERGED', + mergedAt: '2026-02-11T10:08:37Z', + url: 'https://github.com/org/repo/pull/1411', + }); + }; + + await executeHook({ hook, agent, result }); + + assert(agent.lastPublished, 'Expected CLUSTER_COMPLETE to be published'); + assert.strictEqual(agent.lastPublished.topic, 'CLUSTER_COMPLETE'); + assert.strictEqual(agent.lastPublished.content.data.pr_number, 1411); + assert(callCount >= 4, `Expected at least 4 gh calls (got ${callCount})`); + }); + + it('should use explicit PR number in gh command when available in agent output', async function () { + const agent = createMockAgent(); + const hook = { action: 'verify_github_pr' }; + const result = { + output: JSON.stringify({ + pr_url: 'https://github.com/org/repo/pull/555', + pr_number: 555, + merged: true, + }), + }; + + let capturedCmd; + mockExecSyncFn = (cmd) => { + capturedCmd = cmd; + return JSON.stringify({ + number: 555, + state: 'MERGED', + mergedAt: '2026-02-11T10:00:00Z', + url: 'https://github.com/org/repo/pull/555', + }); + }; + + await executeHook({ hook, agent, result }); + assert(capturedCmd.includes('gh pr view 555'), `Expected PR number in command, got: ${capturedCmd}`); + }); + + it('should fall back to branch-based resolution when pr_number not in output', async function () { + const agent = createMockAgent(); + const hook = { action: 'verify_github_pr' }; + const result = { + output: JSON.stringify({ + summary: 'Done', + }), + }; + + let capturedCmd; + mockExecSyncFn = (cmd) => { + capturedCmd = cmd; + return JSON.stringify({ + number: 100, + state: 'MERGED', + mergedAt: '2026-02-11T10:00:00Z', + url: 'https://github.com/org/repo/pull/100', + }); + }; + + await executeHook({ hook, agent, result }); + assert.strictEqual(capturedCmd, 'gh pr view --json state,mergedAt,url,number'); + }); + + it('should publish CLUSTER_COMPLETE when PR verified merged', async function () { + const agent = createMockAgent(); + const hook = { action: 'verify_github_pr' }; + const result = { + output: JSON.stringify({ + pr_url: 'https://github.com/org/repo/pull/456', + merged: true, + }), + }; + + mockExecSyncFn = () => { + return JSON.stringify({ + number: 456, + state: 'MERGED', + mergedAt: '2026-01-15T10:30:00Z', + url: 'https://github.com/org/repo/pull/456', + }); + }; + + await executeHook({ hook, agent, result }); + + assert(agent.lastPublished, 'Expected message to be published'); + assert.strictEqual(agent.lastPublished.topic, 'CLUSTER_COMPLETE'); + assert.strictEqual(agent.lastPublished.content.data.pr_number, 456); + }); + + it('should pass correct workingDirectory to gh CLI', async function () { + const agent = createMockAgent('/custom/work/dir'); + const hook = { action: 'verify_github_pr' }; + const result = { + output: JSON.stringify({ + pr_url: 'https://github.com/org/repo/pull/789', + merged: true, + }), + }; + + let capturedCwd; + mockExecSyncFn = (cmd, opts) => { + capturedCwd = opts.cwd; + return JSON.stringify({ + number: 789, + state: 'MERGED', + mergedAt: '2026-01-15T10:30:00Z', + url: 'https://github.com/org/repo/pull/789', + }); + }; + + await executeHook({ hook, agent, result }); + assert.strictEqual(capturedCwd, '/custom/work/dir'); + }); + + it('should propagate non-hallucination errors', async function () { + const agent = createMockAgent(); + const hook = { action: 'verify_github_pr' }; + const result = { + output: JSON.stringify({ + pr_url: 'https://github.com/org/repo/pull/999', + merged: true, + }), + }; + + mockExecSyncFn = () => { + throw new Error('Network error: timeout'); + }; + + try { + await executeHook({ hook, agent, result }); + assert.fail('Expected error to be thrown'); + } catch (err) { + assert.match(err.message, /Network error: timeout/); + } + }); + + it('should throw when claimed pr_url does not match the branch PR', async function () { + const agent = createMockAgent(); + const hook = { action: 'verify_github_pr' }; + const result = { + output: JSON.stringify({ + pr_url: 'https://github.com/org/repo/pull/111', + merged: true, + }), + }; + + mockExecSyncFn = () => { + return JSON.stringify({ + number: 222, + state: 'MERGED', + mergedAt: '2026-01-15T10:30:00Z', + url: 'https://github.com/org/repo/pull/222', + }); + }; + + await assert.rejects(() => executeHook({ hook, agent, result }), /claimed PR URL/i); + }); +}); diff --git a/tests/worktree-compose-cleanup.test.js b/tests/worktree-compose-cleanup.test.js new file mode 100644 index 00000000..9898ffab --- /dev/null +++ b/tests/worktree-compose-cleanup.test.js @@ -0,0 +1,266 @@ +/** + * Worktree Docker Compose Cleanup Test Suite + * + * Regression test for bug where `docker compose down` was never called when + * cleaning up zeroshot worktrees. Agents could run `docker compose up` inside + * worktrees, and those containers would keep running after session end, + * hogging host ports (5433, 6379, 3001, etc.) and blocking the main project. + * + * Root cause: removeWorktree() only did `git worktree remove` + `fs.rmSync`, + * never tore down Docker Compose services. The orchestrator stop() path + * (Ctrl+C / SIGINT) preserved worktrees entirely, including running containers. + * + * Fix: + * - isolation-manager.js: removeWorktree() now calls `docker compose down` first + * - orchestrator.js: stop() calls _teardownWorktreeCompose() even when preserving worktree + */ + +const assert = require('assert'); +const fs = require('fs'); +const path = require('path'); +const os = require('os'); + +describe('Worktree Docker Compose Cleanup', function () { + this.timeout(10000); + + let tmpDir; + + beforeEach(function () { + tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'zeroshot-compose-cleanup-')); + }); + + afterEach(function () { + fs.rmSync(tmpDir, { recursive: true, force: true }); + }); + + describe('IsolationManager.removeWorktree', function () { + it('should call docker compose down before removing worktree with docker-compose.yml', function () { + // Mock safe-exec to intercept execSync calls from isolation-manager + const safeExec = require('../src/lib/safe-exec'); + const origExecSync = safeExec.execSync; + const calls = []; + + safeExec.execSync = function (cmd, opts) { + calls.push({ cmd, cwd: opts?.cwd }); + if (cmd.includes('docker compose down')) { + return ''; + } + if (cmd.includes('git worktree') || cmd.includes('git branch')) { + throw new Error('not a git repo'); + } + return ''; + }; + + try { + // Re-require to get fresh module with mocked execSync + // (isolation-manager destructures execSync at import, but from safe-exec module object) + // Since IsolationManager binds execSync at require-time, we need to clear cache + delete require.cache[require.resolve('../src/isolation-manager')]; + const IsolationManager = require('../src/isolation-manager'); + const manager = new IsolationManager(); + + // Create a fake worktree directory with docker-compose.yml + const fakeWorktreePath = path.join(tmpDir, 'test-worktree'); + fs.mkdirSync(fakeWorktreePath, { recursive: true }); + fs.writeFileSync(path.join(fakeWorktreePath, 'docker-compose.yml'), 'version: "3"'); + + // Register worktree in manager's internal map + manager.worktrees.set('test-cluster', { + path: fakeWorktreePath, + branch: 'zeroshot/test-cluster', + repoRoot: tmpDir, + }); + + manager.cleanupWorktreeIsolation('test-cluster'); + + // Verify docker compose down was called + const composeDownCall = calls.find((c) => c.cmd.includes('docker compose down')); + assert.ok(composeDownCall, 'docker compose down should be called during cleanup'); + assert.strictEqual( + composeDownCall.cwd, + fakeWorktreePath, + 'docker compose down should run in the worktree directory' + ); + assert.ok( + composeDownCall.cmd.includes('--remove-orphans'), + 'docker compose down should use --remove-orphans' + ); + assert.ok( + composeDownCall.cmd.includes('--volumes'), + 'docker compose down should use --volumes to free disk' + ); + + // Verify ordering: compose down appears before git worktree remove + const composeIdx = calls.findIndex((c) => c.cmd.includes('docker compose down')); + const gitIdx = calls.findIndex((c) => c.cmd.includes('git worktree remove')); + assert.ok( + composeIdx < gitIdx, + `docker compose down (idx ${composeIdx}) should run before git worktree remove (idx ${gitIdx})` + ); + } finally { + safeExec.execSync = origExecSync; + delete require.cache[require.resolve('../src/isolation-manager')]; + } + }); + + it('should skip docker compose down when no docker-compose.yml exists', function () { + const safeExec = require('../src/lib/safe-exec'); + const origExecSync = safeExec.execSync; + const calls = []; + + safeExec.execSync = function (cmd, opts) { + calls.push({ cmd, cwd: opts?.cwd }); + if (cmd.includes('git worktree') || cmd.includes('git branch')) { + throw new Error('not a git repo'); + } + return ''; + }; + + try { + delete require.cache[require.resolve('../src/isolation-manager')]; + const IsolationManager = require('../src/isolation-manager'); + const manager = new IsolationManager(); + + // Create a fake worktree directory WITHOUT docker-compose.yml + const fakeWorktreePath = path.join(tmpDir, 'test-worktree-no-compose'); + fs.mkdirSync(fakeWorktreePath, { recursive: true }); + + manager.worktrees.set('test-no-compose', { + path: fakeWorktreePath, + branch: 'zeroshot/test-no-compose', + repoRoot: tmpDir, + }); + + manager.cleanupWorktreeIsolation('test-no-compose'); + + const composeCall = calls.find((c) => c.cmd.includes('docker compose')); + assert.strictEqual( + composeCall, + undefined, + 'docker compose should NOT be called when no docker-compose.yml exists' + ); + } finally { + safeExec.execSync = origExecSync; + delete require.cache[require.resolve('../src/isolation-manager')]; + } + }); + + it('should not fail when docker compose down throws', function () { + const safeExec = require('../src/lib/safe-exec'); + const origExecSync = safeExec.execSync; + + safeExec.execSync = function (cmd) { + if (cmd.includes('docker compose down')) { + throw new Error('Docker daemon not running'); + } + if (cmd.includes('git')) { + throw new Error('not a git repo'); + } + return ''; + }; + + try { + delete require.cache[require.resolve('../src/isolation-manager')]; + const IsolationManager = require('../src/isolation-manager'); + const manager = new IsolationManager(); + + const fakeWorktreePath = path.join(tmpDir, 'test-worktree-compose-fail'); + fs.mkdirSync(fakeWorktreePath, { recursive: true }); + fs.writeFileSync(path.join(fakeWorktreePath, 'docker-compose.yml'), 'version: "3"'); + + manager.worktrees.set('test-compose-fail', { + path: fakeWorktreePath, + branch: 'zeroshot/test-compose-fail', + repoRoot: tmpDir, + }); + + // Should not throw ā€” compose down failure is best-effort + assert.doesNotThrow(() => { + manager.cleanupWorktreeIsolation('test-compose-fail'); + }); + } finally { + safeExec.execSync = origExecSync; + delete require.cache[require.resolve('../src/isolation-manager')]; + } + }); + }); + + describe('Orchestrator._teardownWorktreeCompose', function () { + // _teardownWorktreeCompose uses safe-exec (required lazily inside the method). + // Mock via the safe-exec module object since the method re-requires it each call. + let safeExec, origExecSync; + + beforeEach(function () { + safeExec = require('../src/lib/safe-exec'); + origExecSync = safeExec.execSync; + }); + + afterEach(function () { + safeExec.execSync = origExecSync; + }); + + it('should tear down compose services during stop (Ctrl+C path)', function () { + const Orchestrator = require('../src/orchestrator'); + const orchestrator = new Orchestrator({ dataDir: tmpDir }); + + const fakeWorktreePath = path.join(tmpDir, 'stop-test-worktree'); + fs.mkdirSync(fakeWorktreePath, { recursive: true }); + fs.writeFileSync(path.join(fakeWorktreePath, 'docker-compose.yml'), 'version: "3"'); + + const calls = []; + safeExec.execSync = function (cmd, opts) { + calls.push({ cmd, cwd: opts?.cwd }); + return ''; + }; + + orchestrator._teardownWorktreeCompose(fakeWorktreePath); + + const composeCall = calls.find((c) => c.cmd.includes('docker compose down')); + assert.ok(composeCall, '_teardownWorktreeCompose should call docker compose down'); + assert.strictEqual(composeCall.cwd, fakeWorktreePath); + assert.ok(composeCall.cmd.includes('--remove-orphans')); + assert.ok(composeCall.cmd.includes('--volumes')); + }); + + it('should skip when no docker-compose.yml exists', function () { + const Orchestrator = require('../src/orchestrator'); + const orchestrator = new Orchestrator({ dataDir: tmpDir }); + + const fakeWorktreePath = path.join(tmpDir, 'no-compose-worktree'); + fs.mkdirSync(fakeWorktreePath, { recursive: true }); + + const calls = []; + safeExec.execSync = function (cmd, opts) { + calls.push({ cmd, cwd: opts?.cwd }); + return ''; + }; + + orchestrator._teardownWorktreeCompose(fakeWorktreePath); + assert.strictEqual( + calls.length, + 0, + 'No commands should be called without docker-compose.yml' + ); + }); + + it('should not throw when docker compose down fails', function () { + const Orchestrator = require('../src/orchestrator'); + const orchestrator = new Orchestrator({ dataDir: tmpDir }); + + const fakeWorktreePath = path.join(tmpDir, 'fail-compose-worktree'); + fs.mkdirSync(fakeWorktreePath, { recursive: true }); + fs.writeFileSync(path.join(fakeWorktreePath, 'docker-compose.yml'), 'version: "3"'); + + safeExec.execSync = function (cmd) { + if (cmd.includes('docker compose')) { + throw new Error('no such service'); + } + return ''; + }; + + assert.doesNotThrow(() => { + orchestrator._teardownWorktreeCompose(fakeWorktreePath); + }); + }); + }); +}); diff --git a/tests/worktree-cwd-injection.test.js b/tests/worktree-cwd-injection.test.js index 546d2d13..eaae0bd9 100644 --- a/tests/worktree-cwd-injection.test.js +++ b/tests/worktree-cwd-injection.test.js @@ -17,7 +17,7 @@ describe('Worktree CWD Injection', function () { describe('_opAddAgents cwd injection', function () { it('should inject worktree path into dynamically added agents', function () { // Simulate a cluster with worktree enabled - const worktreePath = '/tmp/zeroshot-worktrees/test-cluster-123'; + const worktreePath = '/home/eivind/.zeroshot/worktrees/test-cluster-123'; const cluster = { id: 'test-cluster-123', worktree: { @@ -63,7 +63,7 @@ describe('Worktree CWD Injection', function () { }); it('should not override agent cwd if already set', function () { - const worktreePath = '/tmp/zeroshot-worktrees/test-cluster-456'; + const worktreePath = '/home/eivind/.zeroshot/worktrees/test-cluster-456'; const customCwd = '/custom/path/for/agent'; const cluster = { @@ -136,7 +136,7 @@ describe('Worktree CWD Injection', function () { describe('resume path cwd fix', function () { it('should fix agents saved without cwd on resume', function () { - const worktreePath = '/tmp/zeroshot-worktrees/old-cluster'; + const worktreePath = '/home/eivind/.zeroshot/worktrees/old-cluster'; // Simulate cluster data saved BEFORE the bugfix (agents have cwd: null) const clusterData = { diff --git a/tsconfig.json b/tsconfig.json index 722cbadb..1caf592d 100644 --- a/tsconfig.json +++ b/tsconfig.json @@ -29,6 +29,7 @@ "resolveJsonModule": true, "esModuleInterop": true, "allowSyntheticDefaultImports": true, + "typeRoots": ["./node_modules/@types"], // Target Node.js 18+ "target": "ES2022", @@ -52,10 +53,5 @@ "cluster-scripts/**/*", "test/**/*" ], - "exclude": [ - "node_modules", - "dist", - "coverage", - "task-lib" - ] + "exclude": ["node_modules", "dist", "coverage", "task-lib"] } diff --git a/tsconfig.tui-backend.json b/tsconfig.tui-backend.json new file mode 100644 index 00000000..1e6de1fe --- /dev/null +++ b/tsconfig.tui-backend.json @@ -0,0 +1,18 @@ +{ + "compilerOptions": { + "target": "ES2022", + "module": "commonjs", + "moduleResolution": "node", + "jsx": "react", + "esModuleInterop": true, + "allowSyntheticDefaultImports": true, + "typeRoots": ["./node_modules/@types"], + "strict": false, + "skipLibCheck": true, + "moduleDetection": "force", + "outDir": "lib/tui-backend", + "rootDir": "src/tui-backend" + }, + "include": ["src/tui-backend/**/*.ts"], + "exclude": ["node_modules", "lib"] +} diff --git a/tui-rs/Cargo.lock b/tui-rs/Cargo.lock new file mode 100644 index 00000000..ea85e762 --- /dev/null +++ b/tui-rs/Cargo.lock @@ -0,0 +1,651 @@ +# This file is automatically @generated by Cargo. +# It is not intended for manual editing. +version = 4 + +[[package]] +name = "allocator-api2" +version = "0.2.21" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "683d7910e743518b0e34f1186f92494becacb047c7b6bf616c96772180fef923" + +[[package]] +name = "bitflags" +version = "2.10.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "812e12b5285cc515a9c72a5c1d3b6d46a19dac5acfef5265968c166106e31dd3" + +[[package]] +name = "cassowary" +version = "0.3.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "df8670b8c7b9dae1793364eafadf7239c40d669904660c5960d74cfd80b46a53" + +[[package]] +name = "castaway" +version = "0.2.4" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "dec551ab6e7578819132c713a93c022a05d60159dc86e7a7050223577484c55a" +dependencies = [ + "rustversion", +] + +[[package]] +name = "cfg-if" +version = "1.0.4" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "9330f8b2ff13f34540b44e946ef35111825727b38d33286ef986142615121801" + +[[package]] +name = "compact_str" +version = "0.8.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "3b79c4069c6cad78e2e0cdfcbd26275770669fb39fd308a752dc110e83b9af32" +dependencies = [ + "castaway", + "cfg-if", + "itoa", + "rustversion", + "ryu", + "static_assertions", +] + +[[package]] +name = "crossterm" +version = "0.28.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "829d955a0bb380ef178a640b91779e3987da38c9aea133b20614cfed8cdea9c6" +dependencies = [ + "bitflags", + "crossterm_winapi", + "mio", + "parking_lot", + "rustix", + "signal-hook", + "signal-hook-mio", + "winapi", +] + +[[package]] +name = "crossterm_winapi" +version = "0.9.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "acdd7c62a3665c7f6830a51635d9ac9b23ed385797f70a83bb8bafe9c572ab2b" +dependencies = [ + "winapi", +] + +[[package]] +name = "darling" +version = "0.23.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "25ae13da2f202d56bd7f91c25fba009e7717a1e4a1cc98a76d844b65ae912e9d" +dependencies = [ + "darling_core", + "darling_macro", +] + +[[package]] +name = "darling_core" +version = "0.23.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "9865a50f7c335f53564bb694ef660825eb8610e0a53d3e11bf1b0d3df31e03b0" +dependencies = [ + "ident_case", + "proc-macro2", + "quote", + "strsim", + "syn", +] + +[[package]] +name = "darling_macro" +version = "0.23.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "ac3984ec7bd6cfa798e62b4a642426a5be0e68f9401cfc2a01e3fa9ea2fcdb8d" +dependencies = [ + "darling_core", + "quote", + "syn", +] + +[[package]] +name = "either" +version = "1.15.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "48c757948c5ede0e46177b7add2e67155f70e33c07fea8284df6576da70b3719" + +[[package]] +name = "equivalent" +version = "1.0.2" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "877a4ace8713b0bcf2a4e7eec82529c029f1d0619886d18145fea96c3ffe5c0f" + +[[package]] +name = "errno" +version = "0.3.14" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "39cab71617ae0d63f51a36d69f866391735b51691dbda63cf6f96d042b63efeb" +dependencies = [ + "libc", + "windows-sys 0.61.2", +] + +[[package]] +name = "foldhash" +version = "0.1.5" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "d9c4f5dac5e15c24eb999c26181a6ca40b39fe946cbe4c263c7209467bc83af2" + +[[package]] +name = "hashbrown" +version = "0.15.5" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "9229cfe53dfd69f0609a49f65461bd93001ea1ef889cd5529dd176593f5338a1" +dependencies = [ + "allocator-api2", + "equivalent", + "foldhash", +] + +[[package]] +name = "heck" +version = "0.5.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "2304e00983f87ffb38b55b444b5e3b60a884b5d30c0fca7d82fe33449bbe55ea" + +[[package]] +name = "ident_case" +version = "1.0.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "b9e0384b61958566e926dc50660321d12159025e767c18e043daf26b70104c39" + +[[package]] +name = "indoc" +version = "2.0.7" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "79cf5c93f93228cf8efb3ba362535fb11199ac548a09ce117c9b1adc3030d706" +dependencies = [ + "rustversion", +] + +[[package]] +name = "instability" +version = "0.3.11" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "357b7205c6cd18dd2c86ed312d1e70add149aea98e7ef72b9fdf0270e555c11d" +dependencies = [ + "darling", + "indoc", + "proc-macro2", + "quote", + "syn", +] + +[[package]] +name = "itertools" +version = "0.13.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "413ee7dfc52ee1a4949ceeb7dbc8a33f2d6c088194d9f922fb8318faf1f01186" +dependencies = [ + "either", +] + +[[package]] +name = "itoa" +version = "1.0.17" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "92ecc6618181def0457392ccd0ee51198e065e016d1d527a7ac1b6dc7c1f09d2" + +[[package]] +name = "libc" +version = "0.2.180" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "bcc35a38544a891a5f7c865aca548a982ccb3b8650a5b06d0fd33a10283c56fc" + +[[package]] +name = "linux-raw-sys" +version = "0.4.15" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "d26c52dbd32dccf2d10cac7725f8eae5296885fb5703b261f7d0a0739ec807ab" + +[[package]] +name = "lock_api" +version = "0.4.14" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "224399e74b87b5f3557511d98dff8b14089b3dadafcab6bb93eab67d3aace965" +dependencies = [ + "scopeguard", +] + +[[package]] +name = "log" +version = "0.4.29" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "5e5032e24019045c762d3c0f28f5b6b8bbf38563a65908389bf7978758920897" + +[[package]] +name = "lru" +version = "0.12.5" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "234cf4f4a04dc1f57e24b96cc0cd600cf2af460d4161ac5ecdd0af8e1f3b2a38" +dependencies = [ + "hashbrown", +] + +[[package]] +name = "memchr" +version = "2.7.6" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "f52b00d39961fc5b2736ea853c9cc86238e165017a493d1d5c8eac6bdc4cc273" + +[[package]] +name = "mio" +version = "1.1.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "a69bcab0ad47271a0234d9422b131806bf3968021e5dc9328caf2d4cd58557fc" +dependencies = [ + "libc", + "log", + "wasi", + "windows-sys 0.61.2", +] + +[[package]] +name = "parking_lot" +version = "0.12.5" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "93857453250e3077bd71ff98b6a65ea6621a19bb0f559a85248955ac12c45a1a" +dependencies = [ + "lock_api", + "parking_lot_core", +] + +[[package]] +name = "parking_lot_core" +version = "0.9.12" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "2621685985a2ebf1c516881c026032ac7deafcda1a2c9b7850dc81e3dfcb64c1" +dependencies = [ + "cfg-if", + "libc", + "redox_syscall", + "smallvec", + "windows-link", +] + +[[package]] +name = "paste" +version = "1.0.15" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "57c0d7b74b563b49d38dae00a0c37d4d6de9b432382b2892f0574ddcae73fd0a" + +[[package]] +name = "proc-macro2" +version = "1.0.106" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "8fd00f0bb2e90d81d1044c2b32617f68fcb9fa3bb7640c23e9c748e53fb30934" +dependencies = [ + "unicode-ident", +] + +[[package]] +name = "quote" +version = "1.0.44" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "21b2ebcf727b7760c461f091f9f0f539b77b8e87f2fd88131e7f1b433b3cece4" +dependencies = [ + "proc-macro2", +] + +[[package]] +name = "ratatui" +version = "0.29.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "eabd94c2f37801c20583fc49dd5cd6b0ba68c716787c2dd6ed18571e1e63117b" +dependencies = [ + "bitflags", + "cassowary", + "compact_str", + "crossterm", + "indoc", + "instability", + "itertools", + "lru", + "paste", + "strum", + "unicode-segmentation", + "unicode-truncate", + "unicode-width 0.2.0", +] + +[[package]] +name = "redox_syscall" +version = "0.5.18" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "ed2bf2547551a7053d6fdfafda3f938979645c44812fbfcda098faae3f1a362d" +dependencies = [ + "bitflags", +] + +[[package]] +name = "rustix" +version = "0.38.44" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "fdb5bc1ae2baa591800df16c9ca78619bf65c0488b41b96ccec5d11220d8c154" +dependencies = [ + "bitflags", + "errno", + "libc", + "linux-raw-sys", + "windows-sys 0.59.0", +] + +[[package]] +name = "rustversion" +version = "1.0.22" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "b39cdef0fa800fc44525c84ccb54a029961a8215f9619753635a9c0d2538d46d" + +[[package]] +name = "ryu" +version = "1.0.22" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "a50f4cf475b65d88e057964e0e9bb1f0aa9bbb2036dc65c64596b42932536984" + +[[package]] +name = "scopeguard" +version = "1.2.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "94143f37725109f92c262ed2cf5e59bce7498c01bcc1502d7b9afe439a4e9f49" + +[[package]] +name = "serde" +version = "1.0.228" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "9a8e94ea7f378bd32cbbd37198a4a91436180c5bb472411e48b5ec2e2124ae9e" +dependencies = [ + "serde_core", + "serde_derive", +] + +[[package]] +name = "serde_core" +version = "1.0.228" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "41d385c7d4ca58e59fc732af25c3983b67ac852c1a25000afe1175de458b67ad" +dependencies = [ + "serde_derive", +] + +[[package]] +name = "serde_derive" +version = "1.0.228" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "d540f220d3187173da220f885ab66608367b6574e925011a9353e4badda91d79" +dependencies = [ + "proc-macro2", + "quote", + "syn", +] + +[[package]] +name = "serde_json" +version = "1.0.149" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "83fc039473c5595ace860d8c4fafa220ff474b3fc6bfdb4293327f1a37e94d86" +dependencies = [ + "itoa", + "memchr", + "serde", + "serde_core", + "zmij", +] + +[[package]] +name = "signal-hook" +version = "0.3.18" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "d881a16cf4426aa584979d30bd82cb33429027e42122b169753d6ef1085ed6e2" +dependencies = [ + "libc", + "signal-hook-registry", +] + +[[package]] +name = "signal-hook-mio" +version = "0.2.5" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "b75a19a7a740b25bc7944bdee6172368f988763b744e3d4dfe753f6b4ece40cc" +dependencies = [ + "libc", + "mio", + "signal-hook", +] + +[[package]] +name = "signal-hook-registry" +version = "1.4.8" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "c4db69cba1110affc0e9f7bcd48bbf87b3f4fc7c61fc9155afd4c469eb3d6c1b" +dependencies = [ + "errno", + "libc", +] + +[[package]] +name = "smallvec" +version = "1.15.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "67b1b7a3b5fe4f1376887184045fcf45c69e92af734b7aaddc05fb777b6fbd03" + +[[package]] +name = "static_assertions" +version = "1.1.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "a2eb9349b6444b326872e140eb1cf5e7c522154d69e7a0ffb0fb81c06b37543f" + +[[package]] +name = "strsim" +version = "0.11.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "7da8b5736845d9f2fcb837ea5d9e2628564b3b043a70948a3f0b778838c5fb4f" + +[[package]] +name = "strum" +version = "0.26.3" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "8fec0f0aef304996cf250b31b5a10dee7980c85da9d759361292b8bca5a18f06" +dependencies = [ + "strum_macros", +] + +[[package]] +name = "strum_macros" +version = "0.26.4" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "4c6bee85a5a24955dc440386795aa378cd9cf82acd5f764469152d2270e581be" +dependencies = [ + "heck", + "proc-macro2", + "quote", + "rustversion", + "syn", +] + +[[package]] +name = "syn" +version = "2.0.114" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "d4d107df263a3013ef9b1879b0df87d706ff80f65a86ea879bd9c31f9b307c2a" +dependencies = [ + "proc-macro2", + "quote", + "unicode-ident", +] + +[[package]] +name = "unicode-ident" +version = "1.0.22" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "9312f7c4f6ff9069b165498234ce8be658059c6728633667c526e27dc2cf1df5" + +[[package]] +name = "unicode-segmentation" +version = "1.12.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "f6ccf251212114b54433ec949fd6a7841275f9ada20dddd2f29e9ceea4501493" + +[[package]] +name = "unicode-truncate" +version = "1.1.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "b3644627a5af5fa321c95b9b235a72fd24cd29c648c2c379431e6628655627bf" +dependencies = [ + "itertools", + "unicode-segmentation", + "unicode-width 0.1.14", +] + +[[package]] +name = "unicode-width" +version = "0.1.14" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "7dd6e30e90baa6f72411720665d41d89b9a3d039dc45b8faea1ddd07f617f6af" + +[[package]] +name = "unicode-width" +version = "0.2.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "1fc81956842c57dac11422a97c3b8195a1ff727f06e85c84ed2e8aa277c9a0fd" + +[[package]] +name = "wasi" +version = "0.11.1+wasi-snapshot-preview1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "ccf3ec651a847eb01de73ccad15eb7d99f80485de043efb2f370cd654f4ea44b" + +[[package]] +name = "winapi" +version = "0.3.9" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "5c839a674fcd7a98952e593242ea400abe93992746761e38641405d28b00f419" +dependencies = [ + "winapi-i686-pc-windows-gnu", + "winapi-x86_64-pc-windows-gnu", +] + +[[package]] +name = "winapi-i686-pc-windows-gnu" +version = "0.4.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "ac3b87c63620426dd9b991e5ce0329eff545bccbbb34f3be09ff6fb6ab51b7b6" + +[[package]] +name = "winapi-x86_64-pc-windows-gnu" +version = "0.4.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "712e227841d057c1ee1cd2fb22fa7e5a5461ae8e48fa2ca79ec42cfc1931183f" + +[[package]] +name = "windows-link" +version = "0.2.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "f0805222e57f7521d6a62e36fa9163bc891acd422f971defe97d64e70d0a4fe5" + +[[package]] +name = "windows-sys" +version = "0.59.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "1e38bc4d79ed67fd075bcc251a1c39b32a1776bbe92e5bef1f0bf1f8c531853b" +dependencies = [ + "windows-targets", +] + +[[package]] +name = "windows-sys" +version = "0.61.2" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "ae137229bcbd6cdf0f7b80a31df61766145077ddf49416a728b02cb3921ff3fc" +dependencies = [ + "windows-link", +] + +[[package]] +name = "windows-targets" +version = "0.52.6" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "9b724f72796e036ab90c1021d4780d4d3d648aca59e491e6b98e725b84e99973" +dependencies = [ + "windows_aarch64_gnullvm", + "windows_aarch64_msvc", + "windows_i686_gnu", + "windows_i686_gnullvm", + "windows_i686_msvc", + "windows_x86_64_gnu", + "windows_x86_64_gnullvm", + "windows_x86_64_msvc", +] + +[[package]] +name = "windows_aarch64_gnullvm" +version = "0.52.6" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "32a4622180e7a0ec044bb555404c800bc9fd9ec262ec147edd5989ccd0c02cd3" + +[[package]] +name = "windows_aarch64_msvc" +version = "0.52.6" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "09ec2a7bb152e2252b53fa7803150007879548bc709c039df7627cabbd05d469" + +[[package]] +name = "windows_i686_gnu" +version = "0.52.6" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "8e9b5ad5ab802e97eb8e295ac6720e509ee4c243f69d781394014ebfe8bbfa0b" + +[[package]] +name = "windows_i686_gnullvm" +version = "0.52.6" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "0eee52d38c090b3caa76c563b86c3a4bd71ef1a819287c19d586d7334ae8ed66" + +[[package]] +name = "windows_i686_msvc" +version = "0.52.6" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "240948bc05c5e7c6dabba28bf89d89ffce3e303022809e73deaefe4f6ec56c66" + +[[package]] +name = "windows_x86_64_gnu" +version = "0.52.6" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "147a5c80aabfbf0c7d901cb5895d1de30ef2907eb21fbbab29ca94c5b08b1a78" + +[[package]] +name = "windows_x86_64_gnullvm" +version = "0.52.6" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "24d5b23dc417412679681396f2b49f3de8c1473deb516bd34410872eff51ed0d" + +[[package]] +name = "windows_x86_64_msvc" +version = "0.52.6" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "589f6da84c646204747d1270a2a5661ea66ed1cced2631d546fdfb155959f9ec" + +[[package]] +name = "zeroshot-tui" +version = "0.1.0" +dependencies = [ + "crossterm", + "ratatui", + "serde", + "serde_json", +] + +[[package]] +name = "zmij" +version = "1.0.18" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "1966f8ac2c1f76987d69a74d0e0f929241c10e78136434e3be70ff7f58f64214" diff --git a/tui-rs/Cargo.toml b/tui-rs/Cargo.toml new file mode 100644 index 00000000..b63f7c4a --- /dev/null +++ b/tui-rs/Cargo.toml @@ -0,0 +1,3 @@ +[workspace] +members = ["crates/zeroshot-tui"] +resolver = "2" diff --git a/tui-rs/crates/zeroshot-tui/Cargo.toml b/tui-rs/crates/zeroshot-tui/Cargo.toml new file mode 100644 index 00000000..b1435700 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/Cargo.toml @@ -0,0 +1,10 @@ +[package] +name = "zeroshot-tui" +version = "0.1.0" +edition = "2021" + +[dependencies] +ratatui = { version = "0.29", features = ["crossterm"] } +crossterm = "0.28" +serde = { version = "1.0", features = ["derive"] } +serde_json = "1.0" diff --git a/tui-rs/crates/zeroshot-tui/src/app/agent_microscope.rs b/tui-rs/crates/zeroshot-tui/src/app/agent_microscope.rs new file mode 100644 index 00000000..3f9c0a3c --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/app/agent_microscope.rs @@ -0,0 +1,48 @@ +use crate::protocol::ClusterLogLine; +use crate::ui::shared::TimeIndexedBuffer; + +pub const MAX_LOG_LINES: usize = 1000; + +#[derive(Debug, Clone)] +pub struct State { + pub logs_time: TimeIndexedBuffer, + pub log_drop_seq: u64, + pub log_subscription: Option, + pub role: Option, + pub status: Option, +} + +impl Default for State { + fn default() -> Self { + Self { + logs_time: TimeIndexedBuffer::new(MAX_LOG_LINES), + log_drop_seq: 0, + log_subscription: None, + role: None, + status: None, + } + } +} + +impl State { + pub fn push_log_lines(&mut self, mut lines: Vec, dropped_count: Option) { + let mut to_push = Vec::new(); + if let Some(count) = dropped_count { + if count > 0 { + let line = ClusterLogLine { + id: format!("dropped-{}", self.log_drop_seq), + timestamp: lines.first().map(|line| line.timestamp).unwrap_or(0), + text: format!("[dropped {} log lines]", count), + agent: None, + role: None, + sender: None, + }; + self.log_drop_seq = self.log_drop_seq.saturating_add(1); + to_push.push(line); + } + } + + to_push.append(&mut lines); + self.logs_time.push_many(to_push); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/app/animation.rs b/tui-rs/crates/zeroshot-tui/src/app/animation.rs new file mode 100644 index 00000000..65fd7c46 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/app/animation.rs @@ -0,0 +1,102 @@ +use std::f32::consts::TAU; + +pub const DEFAULT_TICK_MS: i64 = 250; +pub const MIN_TICK_MS: i64 = 16; +pub const MAX_TICK_MS: i64 = 1000; +pub const PHASE_TICKS: u64 = 24; + +#[derive(Debug, Clone, Copy, PartialEq)] +pub struct AnimClock { + pub now_ms: i64, + pub tick: u64, + pub phase: f32, +} + +impl Default for AnimClock { + fn default() -> Self { + Self { + now_ms: 0, + tick: 0, + phase: 0.0, + } + } +} + +impl AnimClock { + pub fn advance(&mut self, now_ms: i64) { + self.now_ms = now_ms; + self.tick = self.tick.saturating_add(1); + self.phase = ((self.tick % PHASE_TICKS) as f32) / (PHASE_TICKS as f32); + } +} + +pub fn pulse_factor(phase: f32) -> f32 { + 0.5 + 0.5 * (phase * TAU).sin() +} + +pub fn smooth_factor(dt_ms: i64, base: f64) -> f64 { + let dt_scale = (dt_ms as f64 / DEFAULT_TICK_MS as f64).clamp(0.2, 2.0); + (base * dt_scale).clamp(0.0, 1.0) +} + +pub fn lerp_f64(current: f64, target: f64, t: f64) -> f64 { + current + (target - current) * t +} + +pub fn smooth_toward_f64(current: f64, target: f64, dt_ms: i64, rate: f64) -> f64 { + let t = smooth_factor(dt_ms, rate); + lerp_f64(current, target, t) +} + +pub fn step_spring_f32( + position: (f32, f32), + velocity: (f32, f32), + target: (f32, f32), + dt_ms: i64, + accel: f32, + friction: f32, +) -> ((f32, f32), (f32, f32)) { + let dt_scale = (dt_ms as f32 / DEFAULT_TICK_MS as f32).clamp(0.2, 2.0); + let accel_step = accel * dt_scale; + let friction_step = friction.powf(dt_scale); + + let mut vx = velocity.0 + (target.0 - position.0) * accel_step; + let mut vy = velocity.1 + (target.1 - position.1) * accel_step; + vx *= friction_step; + vy *= friction_step; + + let px = position.0 + vx * dt_scale; + let py = position.1 + vy * dt_scale; + + ((px, py), (vx, vy)) +} + +pub fn step_spring_f64( + position: (f64, f64), + velocity: (f64, f64), + target: (f64, f64), + dt_ms: i64, + accel: f64, + friction: f64, +) -> ((f64, f64), (f64, f64)) { + let dt_scale = (dt_ms as f64 / DEFAULT_TICK_MS as f64).clamp(0.2, 2.0); + let accel_step = accel * dt_scale; + let friction_step = friction.powf(dt_scale); + + let mut vx = velocity.0 + (target.0 - position.0) * accel_step; + let mut vy = velocity.1 + (target.1 - position.1) * accel_step; + vx *= friction_step; + vy *= friction_step; + + let px = position.0 + vx * dt_scale; + let py = position.1 + vy * dt_scale; + + ((px, py), (vx, vy)) +} + +pub fn clamp_tick_delta(last_tick_ms: Option, now_ms: i64) -> i64 { + let raw = last_tick_ms + .map(|last| now_ms.saturating_sub(last)) + .unwrap_or(DEFAULT_TICK_MS); + raw.clamp(MIN_TICK_MS, MAX_TICK_MS) +} diff --git a/tui-rs/crates/zeroshot-tui/src/app/mod.rs b/tui-rs/crates/zeroshot-tui/src/app/mod.rs new file mode 100644 index 00000000..6f68cfb7 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/app/mod.rs @@ -0,0 +1,2889 @@ +use std::collections::{HashMap, HashSet}; + +use crate::backend::{BackendExit, BackendNotification}; +use crate::protocol::ClusterMetrics; +use crate::screens::{agent, cluster, cluster_canvas, launcher, monitor, radar}; +use crate::ui::shared::InputState; + +pub mod agent_microscope; +pub mod animation; +mod spine_completion; +mod spine_hint; +use animation::{clamp_tick_delta, step_spring_f32, AnimClock}; +use spine_completion::{build_spine_completion, select_spine_completion}; +pub use spine_hint::{compute_spine_hint, SpineHint, SpineHintTone}; + +#[derive(Debug, Clone, PartialEq, Eq, Hash)] +pub struct AgentKey { + pub cluster_id: String, + pub agent_id: String, +} + +impl AgentKey { + pub fn new(cluster_id: impl Into, agent_id: impl Into) -> Self { + Self { + cluster_id: cluster_id.into(), + agent_id: agent_id.into(), + } + } +} + +#[derive(Debug, Clone, PartialEq, Eq, Hash)] +pub enum ScreenId { + Launcher, + Monitor, + IntentConsole, + FleetRadar, + Cluster { + id: String, + }, + ClusterCanvas { + id: String, + }, + Agent { + cluster_id: String, + agent_id: String, + }, + AgentMicroscope { + cluster_id: String, + agent_id: String, + }, +} + +impl ScreenId { + pub fn title(&self) -> String { + match self { + ScreenId::Launcher => "Launcher".to_string(), + ScreenId::Monitor => "Monitor".to_string(), + ScreenId::IntentConsole => "Intent Console".to_string(), + ScreenId::FleetRadar => "Fleet Radar".to_string(), + ScreenId::Cluster { id } => format!("Cluster {id}"), + ScreenId::ClusterCanvas { id } => format!("Cluster Canvas {id}"), + ScreenId::Agent { + cluster_id, + agent_id, + } => format!("Agent {agent_id} @ {cluster_id}"), + ScreenId::AgentMicroscope { + cluster_id, + agent_id, + } => format!("Agent Microscope {agent_id} @ {cluster_id}"), + } + } +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum ZoomStackContext { + Root, + FleetRadar, + Cluster { + id: String, + }, + Agent { + cluster_id: String, + agent_id: String, + }, +} + +#[derive(Debug, Clone, PartialEq, Eq, Default)] +pub enum TemporalFocus { + #[default] + None, + Cluster { + id: String, + }, + Agent { + cluster_id: String, + agent_id: String, + }, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum FocusTarget { + Cluster { + id: String, + }, + Agent { + cluster_id: String, + agent_id: String, + }, +} + +impl FocusTarget { + fn label(&self) -> String { + match self { + FocusTarget::Cluster { id } => format!("cluster {id}"), + FocusTarget::Agent { + cluster_id, + agent_id, + } => format!("agent {agent_id} @ {cluster_id}"), + } + } +} + +impl TemporalFocus { + pub fn is_active(&self) -> bool { + !matches!(self, TemporalFocus::None) + } +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum InitialScreen { + Launcher, + Monitor, +} + +impl InitialScreen { + pub fn parse(value: &str) -> Result { + match value.trim().to_lowercase().as_str() { + "launcher" => Ok(Self::Launcher), + "monitor" => Ok(Self::Monitor), + other => Err(format!( + "Unknown initial screen: {other}. Valid: launcher, monitor" + )), + } + } +} + +#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)] +pub enum UiVariant { + #[default] + Classic, + Disruptive, +} + +impl UiVariant { + pub fn parse(value: &str) -> Result { + match value.trim().to_lowercase().as_str() { + "classic" => Ok(Self::Classic), + "disruptive" => Ok(Self::Disruptive), + other => Err(format!( + "Unknown UI variant: {other}. Valid: classic, disruptive" + )), + } + } +} + +pub fn resolve_ui_variant( + cli_value: Option<&str>, + env_value: Option<&str>, +) -> Result, String> { + if let Some(raw) = cli_value { + if !raw.trim().is_empty() { + return Ok(Some(UiVariant::parse(raw)?)); + } + } + + if let Some(raw) = env_value { + if !raw.trim().is_empty() { + return Ok(Some(UiVariant::parse(raw)?)); + } + } + + Ok(None) +} + +#[derive(Debug, Clone, Copy, PartialEq)] +pub struct Camera { + pub position: (f32, f32), + pub zoom: f32, +} + +impl Default for Camera { + fn default() -> Self { + Self { + position: (0.0, 0.0), + zoom: 1.0, + } + } +} + +#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)] +pub enum TimeCursorMode { + #[default] + Live, + Scrub, +} + +const DEFAULT_TIME_WINDOW_MS: i64 = 60_000; +pub const TIME_SCRUB_STEP_MS: i64 = 1000; +pub const TIME_SCRUB_STEP_LARGE_MS: i64 = 5000; + +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub struct TimeCursor { + pub mode: TimeCursorMode, + pub t_ms: i64, + pub window_ms: i64, +} + +impl Default for TimeCursor { + fn default() -> Self { + Self { + mode: TimeCursorMode::Live, + t_ms: 0, + window_ms: DEFAULT_TIME_WINDOW_MS, + } + } +} + +#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)] +pub enum SpineMode { + #[default] + Intent, + Command, + WhisperCluster, + WhisperAgent, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct SpineCompletion { + pub candidates: Vec, + pub selected: usize, + pub ghost: String, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct SpineState { + pub mode: SpineMode, + pub input: InputState, + pub hint: SpineHint, + pub completion: Option, +} + +impl Default for SpineState { + fn default() -> Self { + Self { + mode: SpineMode::Intent, + input: InputState::default(), + hint: SpineHint::default(), + completion: None, + } + } +} + +#[derive(Debug, Clone, Default, PartialEq, Eq)] +pub struct StartupOptions { + pub initial_screen: Option, + pub provider_override: Option, + pub ui_variant: Option, +} + +#[derive(Debug, Clone)] +pub enum BackendStatus { + Disconnected, + Connected, + Error(String), + Exited(BackendExit), +} + +#[derive(Debug, Clone, Default)] +pub struct CommandBarState { + pub active: bool, + inner: InputState, +} + +impl CommandBarState { + /// Read-only access to input text. + pub fn input(&self) -> &str { + &self.inner.input + } + + /// Read-only access to cursor position. + pub fn cursor(&self) -> usize { + self.inner.cursor + } + + pub fn open_with(&mut self, prefill: String) { + self.active = true; + self.inner.input = prefill; + self.inner.move_end(); + } + + pub fn close(&mut self) { + self.active = false; + self.inner.clear(); + } + + pub fn insert_char(&mut self, ch: char) { + self.inner.insert_char(ch); + } + + pub fn backspace(&mut self) { + self.inner.backspace(); + } + + pub fn delete(&mut self) { + self.inner.delete(); + } + + pub fn move_left(&mut self) { + self.inner.move_left(); + } + + pub fn move_right(&mut self) { + self.inner.move_right(); + } + + pub fn move_home(&mut self) { + self.inner.move_home(); + } + + pub fn move_end(&mut self) { + self.inner.move_end(); + } +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum ToastLevel { + Info, + Success, + Error, +} + +#[derive(Debug, Clone)] +pub struct ToastState { + pub message: String, + pub level: ToastLevel, + pub expires_at_ms: i64, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct CommandContext { + pub provider_override: Option, + pub active_screen: ScreenId, + pub ui_variant: UiVariant, +} + +#[derive(Debug, Clone)] +pub struct AppState { + pub screen_stack: Vec, + pub launcher: launcher::State, + pub monitor: monitor::State, + pub fleet_radar: radar::FleetRadarState, + pub metrics: HashMap, + pub last_metrics_poll_at: Option, + pub clusters: HashMap, + pub cluster_canvases: HashMap, + pub agents: HashMap, + pub agent_microscopes: HashMap, + pub last_size: Option<(u16, u16)>, + pub tick_count: u64, + pub now_ms: i64, + pub anim_clock: AnimClock, + pub last_tick_ms: Option, + pub should_quit: bool, + pub backend_status: BackendStatus, + pub last_error: Option, + pub provider_override: Option, + pub ui_variant: UiVariant, + pub camera: Camera, + pub camera_target: (f32, f32), + pub camera_velocity: (f32, f32), + pub time_cursor: TimeCursor, + pub temporal_focus: TemporalFocus, + pub pinned_target: Option, + pub spine: SpineState, + pub command_bar: CommandBarState, + pub toast: Option, +} + +impl Default for AppState { + fn default() -> Self { + Self { + screen_stack: vec![ScreenId::Launcher], + launcher: launcher::State::default(), + monitor: monitor::State::default(), + fleet_radar: radar::FleetRadarState::default(), + metrics: HashMap::new(), + last_metrics_poll_at: None, + clusters: HashMap::new(), + cluster_canvases: HashMap::new(), + agents: HashMap::new(), + agent_microscopes: HashMap::new(), + last_size: None, + tick_count: 0, + now_ms: 0, + anim_clock: AnimClock::default(), + last_tick_ms: None, + should_quit: false, + backend_status: BackendStatus::Disconnected, + last_error: None, + provider_override: None, + ui_variant: UiVariant::Classic, + camera: Camera::default(), + camera_target: (0.0, 0.0), + camera_velocity: (0.0, 0.0), + time_cursor: TimeCursor::default(), + temporal_focus: TemporalFocus::default(), + pinned_target: None, + spine: SpineState::default(), + command_bar: CommandBarState::default(), + toast: None, + } + } +} + +impl AppState { + pub fn new() -> Self { + Self::default() + } + + pub fn apply_startup_options(&mut self, options: StartupOptions) { + if let Some(provider) = options.provider_override { + self.provider_override = Some(provider); + } + + if let Some(ui_variant) = options.ui_variant { + self.ui_variant = ui_variant; + } + + let initial_screen = options.initial_screen; + if matches!(self.ui_variant, UiVariant::Disruptive) { + let mut stack = vec![ScreenId::IntentConsole]; + stack.push(ScreenId::FleetRadar); + self.screen_stack = stack; + } else if let Some(initial_screen) = initial_screen { + self.screen_stack = vec![ScreenId::Launcher]; + match initial_screen { + InitialScreen::Launcher => {} + InitialScreen::Monitor => { + self.screen_stack.push(ScreenId::Monitor); + } + } + } + } + + pub fn active_screen(&self) -> &ScreenId { + self.screen_stack.last().unwrap_or(&ScreenId::Launcher) + } + + pub fn temporal_focus_scope(&self) -> Option { + match self.active_screen() { + ScreenId::ClusterCanvas { id } => Some(TemporalFocus::Cluster { id: id.clone() }), + ScreenId::AgentMicroscope { + cluster_id, + agent_id, + } => Some(TemporalFocus::Agent { + cluster_id: cluster_id.clone(), + agent_id: agent_id.clone(), + }), + _ => None, + } + } + + pub fn zoom_stack_context(&self) -> ZoomStackContext { + for screen in self.screen_stack.iter().rev() { + match screen { + ScreenId::Agent { + cluster_id, + agent_id, + } + | ScreenId::AgentMicroscope { + cluster_id, + agent_id, + } => { + return ZoomStackContext::Agent { + cluster_id: cluster_id.clone(), + agent_id: agent_id.clone(), + }; + } + ScreenId::Cluster { id } | ScreenId::ClusterCanvas { id } => { + return ZoomStackContext::Cluster { id: id.clone() }; + } + ScreenId::Monitor | ScreenId::FleetRadar => { + return ZoomStackContext::FleetRadar; + } + ScreenId::Launcher | ScreenId::IntentConsole => {} + } + } + ZoomStackContext::Root + } + + pub fn command_context(&self) -> CommandContext { + CommandContext { + provider_override: self.provider_override.clone(), + active_screen: self.active_screen().clone(), + ui_variant: self.ui_variant, + } + } + + fn metrics_poll_due(&self, now_ms: i64) -> bool { + match self.last_metrics_poll_at { + None => true, + Some(last) => now_ms.saturating_sub(last) >= METRICS_POLL_INTERVAL_MS, + } + } + + fn mark_metrics_polled(&mut self, now_ms: i64) { + self.last_metrics_poll_at = Some(now_ms); + } + + fn ensure_screen_state(&mut self, screen: &ScreenId) { + match screen { + ScreenId::Launcher + | ScreenId::Monitor + | ScreenId::IntentConsole + | ScreenId::FleetRadar => {} + ScreenId::Cluster { id } => { + self.clusters.entry(id.clone()).or_default(); + } + ScreenId::ClusterCanvas { id } => { + self.clusters.entry(id.clone()).or_default(); + self.cluster_canvases.entry(id.clone()).or_default(); + } + ScreenId::Agent { + cluster_id, + agent_id, + } => { + let key = AgentKey::new(cluster_id.clone(), agent_id.clone()); + self.agents.entry(key).or_default(); + } + ScreenId::AgentMicroscope { + cluster_id, + agent_id, + } => { + let key = AgentKey::new(cluster_id.clone(), agent_id.clone()); + self.agents.entry(key.clone()).or_default(); + self.agent_microscopes.entry(key).or_default(); + } + } + } +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum NavigationAction { + Push(ScreenId), + Pop, + ReplaceTop(ScreenId), +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum ScreenAction { + Launcher(launcher::Action), + Monitor(monitor::Action), + FleetRadar(radar::Action), + Cluster { + id: String, + action: cluster::Action, + }, + ClusterCanvas { + id: String, + action: cluster_canvas::Action, + }, + Agent { + cluster_id: String, + agent_id: String, + action: agent::Action, + }, +} + +#[derive(Debug, Clone)] +pub enum BackendAction { + Connected, + ConnectionFailed(String), + BackendExited(BackendExit), + Notification(BackendNotification), + ClustersListed(Vec), + ClusterMetricsListed { + metrics: Vec, + }, + ClusterSummary { + summary: crate::protocol::ClusterSummary, + }, + ClusterTopology { + cluster_id: String, + topology: crate::protocol::ClusterTopology, + }, + ClusterTopologyError { + cluster_id: String, + message: String, + }, + SubscribedClusterLogs { + cluster_id: String, + agent_id: Option, + subscription_id: String, + }, + SubscribedClusterTimeline { + cluster_id: String, + subscription_id: String, + }, + GuidanceToAgentResult { + cluster_id: String, + agent_id: String, + result: crate::protocol::GuidanceDeliveryResult, + }, + GuidanceToAgentError { + cluster_id: String, + agent_id: String, + message: String, + }, + StartClusterResult { + cluster_id: String, + }, + Error(String), +} + +#[derive(Debug, Clone)] +pub enum Action { + Tick { now_ms: i64 }, + Resize { width: u16, height: u16 }, + Quit, + Navigate(NavigationAction), + Screen(ScreenAction), + Backend(BackendAction), + CommandBar(CommandBarAction), + Spine(SpineAction), + TimeCursor(TimeCursorAction), + Command(CommandAction), +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum Effect { + Backend(BackendRequest), + Command(CommandRequest), +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum BackendRequest { + ListClusters, + ListClusterMetrics { + cluster_ids: Option>, + }, + GetClusterSummary { + cluster_id: String, + }, + GetClusterTopology { + cluster_id: String, + }, + SubscribeClusterLogs { + cluster_id: String, + agent_id: Option, + }, + SubscribeClusterTimeline { + cluster_id: String, + }, + StartClusterFromText { + text: String, + provider_override: Option, + }, + StartClusterFromIssue { + reference: String, + provider_override: Option, + }, + SendGuidanceToCluster { + cluster_id: String, + message: String, + }, + SendGuidanceToAgent { + cluster_id: String, + agent_id: String, + message: String, + }, + Unsubscribe { + subscription_id: String, + }, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum CommandRequest { + SubmitRaw { + raw: String, + context: CommandContext, + }, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum CommandBarAction { + Open { prefill: String }, + Close, + InsertChar(char), + Backspace, + Delete, + MoveCursorLeft, + MoveCursorRight, + MoveCursorHome, + MoveCursorEnd, + Submit, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum SpineAction { + SetMode(SpineMode), + SetHint(SpineHint), + SetCompletion(Option), + EnterMode { mode: SpineMode, prefill: String }, + Cancel, + Submit, + AcceptCompletion, + CycleCompletion, + InsertChar(char), + Backspace, + Delete, + MoveCursorLeft, + MoveCursorRight, + MoveCursorHome, + MoveCursorEnd, + Clear, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum TimeCursorAction { + Step { delta_ms: i64 }, + JumpToLive, + ToggleFollow, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum CommandAction { + ShowToast { + level: ToastLevel, + message: String, + }, + SetProviderOverride { + provider: Option, + }, + StartClusterFromIssue { + reference: String, + provider_override: Option, + }, + SendGuidance { + message: String, + prefix: Option, + }, + TogglePin, +} + +const TOAST_DURATION_MS: i64 = 5000; +const METRICS_POLL_INTERVAL_MS: i64 = 2000; +const CAMERA_ACCEL: f32 = 0.16; +const CAMERA_FRICTION: f32 = 0.82; +const CAMERA_SNAP_EPSILON: f32 = 0.08; + +pub fn update(mut state: AppState, action: Action) -> (AppState, Vec) { + let mut effects = Vec::new(); + match action { + Action::Tick { now_ms } => { + let dt_ms = clamp_tick_delta(state.last_tick_ms, now_ms); + state.last_tick_ms = Some(now_ms); + state.tick_count = state.tick_count.saturating_add(1); + state.now_ms = now_ms; + state.anim_clock.advance(now_ms); + if let Some(toast) = &state.toast { + if toast.expires_at_ms <= state.now_ms { + state.toast = None; + } + } + state.fleet_radar.tick_orb_smoothing(state.now_ms, dt_ms); + update_radar_camera_smoothing(&mut state, dt_ms); + for canvas_state in state.cluster_canvases.values_mut() { + canvas_state.tick_camera(dt_ms); + } + let should_poll = if matches!(state.ui_variant, UiVariant::Disruptive) { + state.fleet_radar.poll_due(now_ms) + } else { + match state.active_screen() { + ScreenId::Monitor => state.monitor.poll_due(now_ms), + ScreenId::FleetRadar => state.fleet_radar.poll_due(now_ms), + _ => false, + } + }; + if should_poll { + if matches!(state.ui_variant, UiVariant::Disruptive) { + state.fleet_radar.mark_polled(now_ms); + } else { + match state.active_screen() { + ScreenId::Monitor => state.monitor.mark_polled(now_ms), + ScreenId::FleetRadar => state.fleet_radar.mark_polled(now_ms), + _ => {} + } + } + effects.push(Effect::Backend(BackendRequest::ListClusters)); + } + let should_poll_metrics = matches!( + state.active_screen(), + ScreenId::Monitor + | ScreenId::FleetRadar + | ScreenId::Cluster { .. } + | ScreenId::ClusterCanvas { .. } + ) && state.metrics_poll_due(now_ms); + if should_poll_metrics { + if let Some(request) = metrics_request_for_screen(&state) { + state.mark_metrics_polled(now_ms); + effects.push(Effect::Backend(request)); + } + } + } + Action::Resize { width, height } => { + state.last_size = Some((width, height)); + } + Action::Quit => { + state.should_quit = true; + } + Action::Navigate(nav) => { + apply_navigation(&mut state, nav, &mut effects); + } + Action::Screen(screen_action) => { + handle_screen_action(&mut state, screen_action, &mut effects); + } + Action::Backend(backend_action) => { + handle_backend_action(&mut state, backend_action, &mut effects); + } + Action::CommandBar(command_action) => { + handle_command_bar_action(&mut state, command_action, &mut effects); + } + Action::Spine(spine_action) => { + handle_spine_action(&mut state, spine_action, &mut effects); + } + Action::TimeCursor(time_action) => { + handle_time_cursor_action(&mut state, time_action); + } + Action::Command(command_action) => { + handle_command_action(&mut state, command_action, &mut effects); + } + } + + (state, effects) +} + +fn apply_navigation(state: &mut AppState, nav: NavigationAction, effects: &mut Vec) { + match nav { + NavigationAction::Push(screen) => { + cleanup_active_screen(state, effects); + seed_agent_role_for_navigation(state, &screen); + state.ensure_screen_state(&screen); + if let ScreenId::ClusterCanvas { id } = &screen { + ensure_cluster_canvas_focus(state, id); + } + if matches!(screen, ScreenId::Monitor) { + state.monitor.mark_polled(state.now_ms); + } else if matches!(screen, ScreenId::FleetRadar) { + state.fleet_radar.mark_polled(state.now_ms); + } + state.screen_stack.push(screen.clone()); + queue_navigation_effects(&screen, effects); + } + NavigationAction::Pop => { + if state.screen_stack.len() > 1 { + cleanup_active_screen(state, effects); + state.screen_stack.pop(); + if let Some(active) = state.screen_stack.last() { + if matches!(active, ScreenId::Monitor) { + state.monitor.mark_polled(state.now_ms); + } else if matches!(active, ScreenId::FleetRadar) { + state.fleet_radar.mark_polled(state.now_ms); + } + queue_navigation_effects(active, effects); + } + } + } + NavigationAction::ReplaceTop(screen) => { + cleanup_active_screen(state, effects); + seed_agent_role_for_navigation(state, &screen); + state.ensure_screen_state(&screen); + if let ScreenId::ClusterCanvas { id } = &screen { + ensure_cluster_canvas_focus(state, id); + } + if matches!(screen, ScreenId::Monitor) { + state.monitor.mark_polled(state.now_ms); + } else if matches!(screen, ScreenId::FleetRadar) { + state.fleet_radar.mark_polled(state.now_ms); + } + if state.screen_stack.is_empty() { + state.screen_stack.push(screen.clone()); + } else { + let top = state.screen_stack.len() - 1; + state.screen_stack[top] = screen.clone(); + } + queue_navigation_effects(&screen, effects); + } + } + + apply_spine_defaults_for_screen(state); + refresh_spine_hint(state); + refresh_spine_completion(state); + sync_temporal_focus(state); +} + +fn ensure_cluster_canvas_focus(state: &mut AppState, id: &str) { + let Some(cluster_state) = state.clusters.get(id) else { + return; + }; + let Some(topology) = cluster_state.topology.as_ref() else { + return; + }; + if let Some(canvas_state) = state.cluster_canvases.get_mut(id) { + canvas_state.update_layout(topology); + } +} + +fn seed_agent_role_for_navigation(state: &mut AppState, screen: &ScreenId) { + let (cluster_id, agent_id) = match screen { + ScreenId::Agent { + cluster_id, + agent_id, + } + | ScreenId::AgentMicroscope { + cluster_id, + agent_id, + } => (cluster_id, agent_id), + _ => return, + }; + + let role = state.clusters.get(cluster_id).and_then(|cluster_state| { + cluster_state + .agents + .iter() + .find(|agent| agent.id == *agent_id) + .and_then(|agent| agent.role.clone()) + }); + seed_agent_role(state, cluster_id, agent_id, role); +} + +fn queue_navigation_effects(screen: &ScreenId, effects: &mut Vec) { + match screen { + ScreenId::Monitor | ScreenId::FleetRadar => { + effects.push(Effect::Backend(BackendRequest::ListClusters)); + } + ScreenId::Cluster { id } | ScreenId::ClusterCanvas { id } => { + effects.push(Effect::Backend(BackendRequest::GetClusterSummary { + cluster_id: id.clone(), + })); + effects.push(Effect::Backend(BackendRequest::GetClusterTopology { + cluster_id: id.clone(), + })); + effects.push(Effect::Backend(BackendRequest::SubscribeClusterLogs { + cluster_id: id.clone(), + agent_id: None, + })); + effects.push(Effect::Backend(BackendRequest::SubscribeClusterTimeline { + cluster_id: id.clone(), + })); + } + ScreenId::Agent { + cluster_id, + agent_id, + } => { + effects.push(Effect::Backend(BackendRequest::SubscribeClusterLogs { + cluster_id: cluster_id.clone(), + agent_id: Some(agent_id.clone()), + })); + } + ScreenId::AgentMicroscope { + cluster_id, + agent_id, + } => { + effects.push(Effect::Backend(BackendRequest::SubscribeClusterLogs { + cluster_id: cluster_id.clone(), + agent_id: Some(agent_id.clone()), + })); + effects.push(Effect::Backend(BackendRequest::SubscribeClusterTimeline { + cluster_id: cluster_id.clone(), + })); + } + ScreenId::Launcher | ScreenId::IntentConsole => {} + } +} + +fn handle_screen_action(state: &mut AppState, action: ScreenAction, effects: &mut Vec) { + match action { + ScreenAction::Launcher(action) => handle_launcher_action(state, action, effects), + ScreenAction::Monitor(action) => handle_monitor_action(state, action, effects), + ScreenAction::FleetRadar(action) => handle_radar_action(state, action, effects), + ScreenAction::Cluster { id, action } => handle_cluster_action(state, id, action, effects), + ScreenAction::ClusterCanvas { id, action } => { + handle_cluster_canvas_action(state, id, action, effects) + } + ScreenAction::Agent { + cluster_id, + agent_id, + action, + } => handle_agent_action(state, cluster_id, agent_id, action, effects), + } +} + +fn handle_launcher_action( + state: &mut AppState, + action: launcher::Action, + effects: &mut Vec, +) { + match action { + launcher::Action::Submit => { + let trimmed = state.launcher.input.trim(); + if trimmed.is_empty() { + state.last_error = Some("Enter text to start a cluster.".to_string()); + return; + } + + state.last_error = None; + if trimmed.starts_with('/') { + effects.push(Effect::Command(CommandRequest::SubmitRaw { + raw: trimmed.to_string(), + context: state.command_context(), + })); + } else { + effects.push(Effect::Backend(BackendRequest::StartClusterFromText { + text: trimmed.to_string(), + provider_override: state.provider_override.clone(), + })); + } + } + launcher::Action::InsertChar(ch) => { + state.launcher.insert_char(ch); + state.last_error = None; + } + launcher::Action::Backspace => { + state.launcher.backspace(); + state.last_error = None; + } + launcher::Action::Delete => { + state.launcher.delete(); + state.last_error = None; + } + launcher::Action::MoveCursorLeft => { + state.launcher.move_left(); + } + launcher::Action::MoveCursorRight => { + state.launcher.move_right(); + } + launcher::Action::MoveCursorHome => { + state.launcher.move_home(); + } + launcher::Action::MoveCursorEnd => { + state.launcher.move_end(); + } + } +} + +fn handle_monitor_action(state: &mut AppState, action: monitor::Action, effects: &mut Vec) { + match action { + monitor::Action::MoveSelection(delta) => { + state.monitor.move_selection(delta); + } + monitor::Action::OpenSelected => { + if let Some(cluster_id) = state.monitor.selected_cluster_id() { + apply_navigation( + state, + NavigationAction::Push(ScreenId::Cluster { id: cluster_id }), + effects, + ); + } + } + } +} + +fn handle_radar_action(state: &mut AppState, action: radar::Action, _effects: &mut Vec) { + match action { + radar::Action::MoveSelection { direction, speed } => { + if state + .fleet_radar + .move_selection_direction(state.now_ms, direction, speed) + { + sync_camera_to_selection(state); + } + } + radar::Action::CenterOnSelection => { + sync_camera_to_selection(state); + } + radar::Action::ResetView => { + state.camera = Camera::default(); + state.camera_target = state.camera.position; + state.camera_velocity = (0.0, 0.0); + } + } +} + +fn handle_cluster_action( + state: &mut AppState, + id: String, + action: cluster::Action, + effects: &mut Vec, +) { + match action { + cluster::Action::CycleFocus(direction) => { + let entry = state.clusters.entry(id).or_default(); + entry.cycle_focus(direction); + } + cluster::Action::MoveFocused(delta) => { + let entry = state.clusters.entry(id).or_default(); + entry.move_focused(delta); + } + cluster::Action::ActivateFocused => { + let (agent_id, role) = { + let entry = state.clusters.entry(id.clone()).or_default(); + let agent_id = entry.activate_focused(); + let role = agent_id.as_ref().and_then(|selected| { + entry + .agents + .iter() + .find(|agent| agent.id == *selected) + .and_then(|agent| agent.role.clone()) + }); + (agent_id, role) + }; + if let Some(agent_id) = agent_id { + seed_agent_role(state, &id, &agent_id, role); + apply_navigation( + state, + NavigationAction::Push(ScreenId::Agent { + cluster_id: id, + agent_id, + }), + effects, + ); + } + } + cluster::Action::OpenAgent(agent_id) => { + let role = { + let entry = state.clusters.entry(id.clone()).or_default(); + entry + .agents + .iter() + .find(|agent| agent.id == agent_id) + .and_then(|agent| agent.role.clone()) + }; + seed_agent_role(state, &id, &agent_id, role); + apply_navigation( + state, + NavigationAction::Push(ScreenId::Agent { + cluster_id: id, + agent_id, + }), + effects, + ); + } + } +} + +fn handle_cluster_canvas_action( + state: &mut AppState, + id: String, + action: cluster_canvas::Action, + effects: &mut Vec, +) { + match action { + cluster_canvas::Action::MoveFocus { direction, speed } => { + let entry = state.cluster_canvases.entry(id).or_default(); + entry.move_focus(direction, speed); + } + cluster_canvas::Action::ZoomIn => { + let agent_id = state + .cluster_canvases + .get(&id) + .and_then(|entry| entry.focused_agent_id()); + if let Some(agent_id) = agent_id { + let role = state.clusters.get(&id).and_then(|entry| { + entry + .agents + .iter() + .find(|agent| agent.id == agent_id) + .and_then(|agent| agent.role.clone()) + }); + seed_agent_role(state, &id, &agent_id, role); + apply_navigation( + state, + NavigationAction::Push(ScreenId::AgentMicroscope { + cluster_id: id, + agent_id, + }), + effects, + ); + } + } + } +} + +fn seed_agent_role(state: &mut AppState, cluster_id: &str, agent_id: &str, role: Option) { + if let Some(role) = role { + let key = AgentKey::new(cluster_id.to_string(), agent_id.to_string()); + let entry = state.agents.entry(key.clone()).or_default(); + if entry.role.is_none() { + entry.role = Some(role.clone()); + } + let microscope_entry = state.agent_microscopes.entry(key).or_default(); + if microscope_entry.role.is_none() { + microscope_entry.role = Some(role); + } + } +} + +fn handle_agent_action( + state: &mut AppState, + cluster_id: String, + agent_id: String, + action: agent::Action, + effects: &mut Vec, +) { + let key = AgentKey::new(cluster_id.clone(), agent_id.clone()); + let entry = state.agents.entry(key).or_default(); + match action { + agent::Action::SubmitGuidance => { + let trimmed = entry.guidance_input.input.trim(); + if trimmed.is_empty() { + entry.apply_guidance_error("Enter guidance text.".to_string()); + return; + } + entry.guidance_pending = true; + entry.last_guidance_error = None; + effects.push(Effect::Backend(BackendRequest::SendGuidanceToAgent { + cluster_id, + agent_id, + message: trimmed.to_string(), + })); + } + agent::Action::InsertChar(ch) => { + entry.guidance_input.insert_char(ch); + } + agent::Action::Backspace => { + entry.guidance_input.backspace(); + } + agent::Action::Delete => { + entry.guidance_input.delete(); + } + agent::Action::MoveCursorLeft => { + entry.guidance_input.move_left(); + } + agent::Action::MoveCursorRight => { + entry.guidance_input.move_right(); + } + agent::Action::MoveCursorHome => { + entry.guidance_input.move_home(); + } + agent::Action::MoveCursorEnd => { + entry.guidance_input.move_end(); + } + agent::Action::ScrollLogs(delta) => { + entry.move_log_scroll(delta); + } + } +} + +fn handle_backend_action(state: &mut AppState, action: BackendAction, effects: &mut Vec) { + match action { + BackendAction::Connected => handle_backend_connected(state), + BackendAction::ConnectionFailed(message) => { + handle_backend_connection_failed(state, message) + } + BackendAction::BackendExited(exit) => handle_backend_exited(state, exit), + BackendAction::Notification(notification) => { + handle_backend_notification(state, notification) + } + BackendAction::ClustersListed(clusters) => handle_clusters_listed(state, clusters), + BackendAction::ClusterMetricsListed { metrics } => { + handle_cluster_metrics_listed(state, metrics) + } + BackendAction::ClusterSummary { summary } => handle_cluster_summary(state, summary), + BackendAction::ClusterTopology { + cluster_id, + topology, + } => handle_cluster_topology(state, cluster_id, topology), + BackendAction::ClusterTopologyError { + cluster_id, + message, + } => handle_cluster_topology_error(state, cluster_id, message), + BackendAction::SubscribedClusterLogs { + cluster_id, + agent_id, + subscription_id, + } => handle_log_subscription(state, cluster_id, agent_id, subscription_id), + BackendAction::SubscribedClusterTimeline { + cluster_id, + subscription_id, + } => handle_cluster_timeline_subscription(state, cluster_id, subscription_id), + BackendAction::GuidanceToAgentResult { + cluster_id, + agent_id, + result, + } => handle_guidance_result(state, cluster_id, agent_id, result), + BackendAction::GuidanceToAgentError { + cluster_id, + agent_id, + message, + } => handle_guidance_error(state, cluster_id, agent_id, message), + BackendAction::StartClusterResult { cluster_id } => { + handle_start_cluster_result(state, cluster_id, effects) + } + BackendAction::Error(message) => handle_backend_error(state, message), + } +} + +fn handle_backend_connected(state: &mut AppState) { + state.backend_status = BackendStatus::Connected; +} + +fn handle_backend_connection_failed(state: &mut AppState, message: String) { + state.backend_status = BackendStatus::Error(message.clone()); + state.last_error = Some(message); +} + +fn handle_backend_exited(state: &mut AppState, exit: BackendExit) { + state.backend_status = BackendStatus::Exited(exit.clone()); + state.last_error = Some(exit.message); +} + +fn handle_backend_notification(state: &mut AppState, notification: BackendNotification) { + match notification { + BackendNotification::ClusterLogLines(params) => { + let latest_ts = params.lines.iter().map(|line| line.timestamp).max(); + let lines = params.lines; + let dropped_count = params.dropped_count; + let role_from_lines = lines.iter().find_map(|line| line.role.clone()); + if let Some((key, entry)) = state.agent_microscopes.iter_mut().find(|(_, entry)| { + entry.log_subscription.as_deref() == Some(params.subscription_id.as_str()) + }) { + if entry.role.is_none() { + if let Some(role) = role_from_lines.clone() { + entry.role = Some(role); + } + } + entry.push_log_lines(lines, dropped_count); + if let Some(role) = role_from_lines { + let entry = state.agents.entry(key.clone()).or_default(); + if entry.role.is_none() { + entry.role = Some(role); + } + } + advance_time_cursor_if_live(state, latest_ts); + return; + } + + if let Some(entry) = state.agents.values_mut().find(|agent| { + agent.log_subscription.as_deref() == Some(params.subscription_id.as_str()) + }) { + if entry.role.is_none() { + if let Some(role) = role_from_lines.clone() { + entry.role = Some(role); + } + } + entry.push_log_lines(lines, dropped_count); + advance_time_cursor_if_live(state, latest_ts); + return; + } + + if let Some(entry) = state.clusters.get_mut(¶ms.cluster_id) { + if entry.log_subscription.as_deref() == Some(params.subscription_id.as_str()) { + entry.push_log_lines(lines, dropped_count); + advance_time_cursor_if_live(state, latest_ts); + } + } + } + BackendNotification::ClusterTimelineEvents(params) => { + let latest_ts = params.events.iter().map(|event| event.timestamp).max(); + let entry = state.clusters.entry(params.cluster_id).or_default(); + entry.push_timeline_events(params.events); + advance_time_cursor_if_live(state, latest_ts); + } + BackendNotification::Unknown { method, .. } => { + state.last_error = Some(format!("Unhandled backend notification: {method}")); + } + } +} + +fn advance_time_cursor_if_live(state: &mut AppState, latest_ts: Option) { + if state.time_cursor.mode != TimeCursorMode::Live { + return; + } + let Some(latest_ts) = latest_ts else { + return; + }; + if latest_ts > state.time_cursor.t_ms { + state.time_cursor.t_ms = latest_ts; + } +} + +fn handle_clusters_listed(state: &mut AppState, clusters: Vec) { + let radar_clusters = clusters.clone(); + state.monitor.set_clusters(clusters, state.now_ms); + state.fleet_radar.set_clusters(radar_clusters, state.now_ms); + sync_camera_to_selection(state); + let ids: HashSet = state + .monitor + .clusters + .iter() + .map(|cluster| cluster.id.clone()) + .collect(); + state.metrics.retain(|id, _| ids.contains(id)); +} + +fn sync_camera_to_selection(state: &mut AppState) { + if let Some(layout) = state.fleet_radar.selected_layout(state.now_ms) { + state.camera_target = (layout.x as f32, layout.y as f32); + state.camera_velocity = (0.0, 0.0); + } +} + +fn update_radar_camera_smoothing(state: &mut AppState, dt_ms: i64) { + let (position, velocity) = step_spring_f32( + state.camera.position, + state.camera_velocity, + state.camera_target, + dt_ms, + CAMERA_ACCEL, + CAMERA_FRICTION, + ); + state.camera.position = position; + state.camera_velocity = velocity; + + let dx = state.camera.position.0 - state.camera_target.0; + let dy = state.camera.position.1 - state.camera_target.1; + if dx.abs() <= CAMERA_SNAP_EPSILON && dy.abs() <= CAMERA_SNAP_EPSILON { + state.camera.position = state.camera_target; + state.camera_velocity = (0.0, 0.0); + } +} + +fn handle_cluster_metrics_listed(state: &mut AppState, metrics: Vec) { + for metric in metrics { + state.metrics.insert(metric.id.clone(), metric); + } +} + +fn handle_cluster_summary(state: &mut AppState, summary: crate::protocol::ClusterSummary) { + let entry = state.clusters.entry(summary.id.clone()).or_default(); + entry.summary = Some(summary); +} + +fn handle_cluster_topology( + state: &mut AppState, + cluster_id: String, + topology: crate::protocol::ClusterTopology, +) { + let canvas_entry = state + .cluster_canvases + .entry(cluster_id.clone()) + .or_default(); + canvas_entry.update_layout(&topology); + + let entry = state.clusters.entry(cluster_id).or_default(); + entry.topology = Some(topology); + entry.topology_error = None; +} + +fn handle_cluster_topology_error(state: &mut AppState, cluster_id: String, message: String) { + let entry = state.clusters.entry(cluster_id.clone()).or_default(); + entry.topology = None; + entry.topology_error = Some(message); + + if let Some(canvas_entry) = state.cluster_canvases.get_mut(&cluster_id) { + canvas_entry.clear_layout(); + } +} + +fn handle_log_subscription( + state: &mut AppState, + cluster_id: String, + agent_id: Option, + subscription_id: String, +) { + match agent_id { + Some(agent_id) => { + let key = AgentKey::new(cluster_id, agent_id); + let active_is_microscope = matches!( + state.active_screen(), + ScreenId::AgentMicroscope { cluster_id, agent_id } + if cluster_id == &key.cluster_id && agent_id == &key.agent_id + ); + if active_is_microscope { + let entry = state.agent_microscopes.entry(key).or_default(); + entry.log_subscription = Some(subscription_id); + } else { + let entry = state.agents.entry(key).or_default(); + entry.log_subscription = Some(subscription_id); + } + } + None => { + let entry = state.clusters.entry(cluster_id.clone()).or_default(); + entry.log_subscription = Some(subscription_id.clone()); + if let Some(canvas_entry) = state.cluster_canvases.get_mut(&cluster_id) { + canvas_entry.log_subscription = Some(subscription_id); + } + } + } +} + +fn handle_cluster_timeline_subscription( + state: &mut AppState, + cluster_id: String, + subscription_id: String, +) { + let entry = state.clusters.entry(cluster_id.clone()).or_default(); + entry.timeline_subscription = Some(subscription_id.clone()); + if let Some(canvas_entry) = state.cluster_canvases.get_mut(&cluster_id) { + canvas_entry.timeline_subscription = Some(subscription_id); + } +} + +fn handle_guidance_result( + state: &mut AppState, + cluster_id: String, + agent_id: String, + result: crate::protocol::GuidanceDeliveryResult, +) { + let key = AgentKey::new(cluster_id, agent_id); + let entry = state.agents.entry(key).or_default(); + entry.apply_guidance_result(result); +} + +fn handle_guidance_error( + state: &mut AppState, + cluster_id: String, + agent_id: String, + message: String, +) { + let key = AgentKey::new(cluster_id, agent_id); + let entry = state.agents.entry(key).or_default(); + entry.apply_guidance_error(message.clone()); + state.last_error = Some(message); +} + +fn handle_start_cluster_result( + state: &mut AppState, + cluster_id: String, + effects: &mut Vec, +) { + state.launcher.clear(); + let screen = if matches!(state.ui_variant, UiVariant::Disruptive) { + ScreenId::ClusterCanvas { id: cluster_id } + } else { + ScreenId::Cluster { id: cluster_id } + }; + apply_navigation(state, NavigationAction::Push(screen), effects); +} + +fn handle_backend_error(state: &mut AppState, message: String) { + state.last_error = Some(message); +} + +fn handle_command_bar_action( + state: &mut AppState, + action: CommandBarAction, + effects: &mut Vec, +) { + match action { + CommandBarAction::Open { prefill } => { + state.command_bar.open_with(prefill); + } + CommandBarAction::Close => { + state.command_bar.close(); + } + CommandBarAction::InsertChar(ch) => { + if state.command_bar.active { + state.command_bar.insert_char(ch); + } + } + CommandBarAction::Backspace => { + if state.command_bar.active { + state.command_bar.backspace(); + } + } + CommandBarAction::Delete => { + if state.command_bar.active { + state.command_bar.delete(); + } + } + CommandBarAction::MoveCursorLeft => { + if state.command_bar.active { + state.command_bar.move_left(); + } + } + CommandBarAction::MoveCursorRight => { + if state.command_bar.active { + state.command_bar.move_right(); + } + } + CommandBarAction::MoveCursorHome => { + if state.command_bar.active { + state.command_bar.move_home(); + } + } + CommandBarAction::MoveCursorEnd => { + if state.command_bar.active { + state.command_bar.move_end(); + } + } + CommandBarAction::Submit => { + let raw = state.command_bar.input().to_string(); + let context = state.command_context(); + state.command_bar.close(); + effects.push(Effect::Command(CommandRequest::SubmitRaw { raw, context })); + } + } +} + +fn set_spine_input(state: &mut AppState, value: String) { + state.spine.input.input = value; + state.spine.input.cursor = state.spine.input.input.chars().count(); +} + +fn reset_spine_state(state: &mut AppState) { + state.spine.mode = SpineMode::Intent; + state.spine.input.clear(); + state.spine.completion = None; + state.spine.hint = SpineHint::default(); + apply_spine_defaults_for_screen(state); +} + +fn set_disruptive_spine_toast(state: &mut AppState, level: ToastLevel, message: String) { + if !matches!(state.ui_variant, UiVariant::Disruptive) { + return; + } + state.toast = Some(ToastState { + message, + level, + expires_at_ms: state.now_ms.saturating_add(TOAST_DURATION_MS), + }); +} + +fn spine_idle(state: &SpineState) -> bool { + matches!(state.mode, SpineMode::Intent) + && state.input.input.is_empty() + && state.completion.is_none() +} + +fn apply_spine_defaults_for_screen(state: &mut AppState) { + if matches!(state.active_screen(), ScreenId::AgentMicroscope { .. }) && spine_idle(&state.spine) + { + state.spine.mode = SpineMode::WhisperAgent; + } +} + +fn refresh_spine_hint(state: &mut AppState) { + state.spine.hint = compute_spine_hint(state); +} + +fn refresh_spine_completion(state: &mut AppState) { + state.spine.completion = build_spine_completion( + state.spine.mode, + state.spine.input.input.as_str(), + state.spine.input.cursor, + ); +} + +fn sync_temporal_focus(state: &mut AppState) { + if !state.temporal_focus.is_active() { + return; + } + state.temporal_focus = state.temporal_focus_scope().unwrap_or(TemporalFocus::None); +} + +fn detect_issue_reference(input: &str) -> Option { + let trimmed = input.trim(); + if trimmed.is_empty() { + return None; + } + if trimmed.chars().all(|ch| ch.is_ascii_digit()) { + return Some(trimmed.to_string()); + } + if let Some(reference) = parse_owner_repo_issue(trimmed) { + return Some(reference); + } + parse_github_issue_url(trimmed) +} + +fn parse_owner_repo_issue(input: &str) -> Option { + let mut parts = input.split('#'); + let repo_ref = parts.next()?; + let number = parts.next()?; + if parts.next().is_some() { + return None; + } + if number.is_empty() || !number.chars().all(|ch| ch.is_ascii_digit()) { + return None; + } + let mut repo_parts = repo_ref.split('/'); + let owner = repo_parts.next()?; + let repo = repo_parts.next()?; + if owner.is_empty() || repo.is_empty() { + return None; + } + if repo_parts.next().is_some() { + return None; + } + Some(format!("{owner}/{repo}#{number}")) +} + +fn parse_github_issue_url(input: &str) -> Option { + let trimmed = input.trim(); + let without_scheme = trimmed + .strip_prefix("https://") + .or_else(|| trimmed.strip_prefix("http://")) + .unwrap_or(trimmed); + let without_host = without_scheme.strip_prefix("github.com/")?; + let mut parts = without_host.split('/'); + let owner = parts.next()?; + let repo = parts.next()?; + let issues = parts.next()?; + if owner.is_empty() || repo.is_empty() || issues != "issues" { + return None; + } + let number_segment = parts.next()?; + let number = number_segment.split(['?', '#']).next().unwrap_or(""); + if number.is_empty() || !number.chars().all(|ch| ch.is_ascii_digit()) { + return None; + } + Some(format!("{owner}/{repo}#{number}")) +} + +fn resolve_spine_cluster_target(state: &AppState) -> Option { + match state.zoom_stack_context() { + ZoomStackContext::Agent { cluster_id, .. } => Some(cluster_id), + ZoomStackContext::Cluster { id } => Some(id), + ZoomStackContext::FleetRadar => match state.active_screen() { + ScreenId::Monitor => state.monitor.selected_cluster_id(), + _ => state.fleet_radar.selected_cluster_id(), + }, + ZoomStackContext::Root => None, + } +} + +fn resolve_spine_agent_target(state: &AppState) -> Option<(String, String)> { + match state.zoom_stack_context() { + ZoomStackContext::Agent { + cluster_id, + agent_id, + } => Some((cluster_id, agent_id)), + ZoomStackContext::Cluster { id } => { + let cluster_state = state.clusters.get(&id)?; + let agent = cluster_state.agents.get(cluster_state.selected_agent)?; + Some((id, agent.id.clone())) + } + ZoomStackContext::FleetRadar | ZoomStackContext::Root => None, + } +} + +fn handle_spine_action(state: &mut AppState, action: SpineAction, effects: &mut Vec) { + let mut should_refresh_hint = false; + let mut should_refresh_completion = false; + match action { + SpineAction::SetMode(mode) => { + state.spine.mode = mode; + should_refresh_hint = true; + should_refresh_completion = true; + } + SpineAction::SetHint(hint) => { + state.spine.hint = hint; + } + SpineAction::SetCompletion(completion) => { + state.spine.completion = completion; + } + SpineAction::EnterMode { mode, prefill } => { + state.spine.mode = mode; + set_spine_input(state, prefill); + state.spine.completion = None; + should_refresh_hint = true; + should_refresh_completion = true; + } + SpineAction::Cancel => { + reset_spine_state(state); + should_refresh_hint = true; + should_refresh_completion = true; + } + SpineAction::Submit => { + let mode = state.spine.mode; + let raw_input = state.spine.input.input.clone(); + let trimmed = raw_input.trim(); + let mut toast_message: Option = None; + match mode { + SpineMode::Command => { + let raw = if raw_input.starts_with('/') { + raw_input + } else { + format!("/{}", raw_input) + }; + let context = state.command_context(); + effects.push(Effect::Command(CommandRequest::SubmitRaw { raw, context })); + } + SpineMode::Intent => { + if !trimmed.is_empty() { + if let Some(reference) = detect_issue_reference(trimmed) { + effects.push(Effect::Backend(BackendRequest::StartClusterFromIssue { + reference: reference.clone(), + provider_override: state.provider_override.clone(), + })); + toast_message = + Some(format!("Starting cluster from issue {reference}...")); + } else { + effects.push(Effect::Backend(BackendRequest::StartClusterFromText { + text: trimmed.to_string(), + provider_override: state.provider_override.clone(), + })); + toast_message = Some("Starting cluster...".to_string()); + } + } + } + SpineMode::WhisperCluster => { + if !trimmed.is_empty() { + if let Some(cluster_id) = resolve_spine_cluster_target(state) { + effects.push(Effect::Backend(BackendRequest::SendGuidanceToCluster { + cluster_id, + message: trimmed.to_string(), + })); + toast_message = Some("Whisper sent to cluster.".to_string()); + } + } + } + SpineMode::WhisperAgent => { + if !trimmed.is_empty() { + if let Some((cluster_id, agent_id)) = resolve_spine_agent_target(state) { + effects.push(Effect::Backend(BackendRequest::SendGuidanceToAgent { + cluster_id, + agent_id, + message: trimmed.to_string(), + })); + toast_message = Some("Whisper sent to agent.".to_string()); + } + } + } + } + if let Some(message) = toast_message { + set_disruptive_spine_toast(state, ToastLevel::Success, message); + } + reset_spine_state(state); + should_refresh_hint = true; + should_refresh_completion = true; + } + SpineAction::AcceptCompletion => { + if let Some(completion) = state.spine.completion.take() { + if !completion.ghost.is_empty() { + state.spine.input.input.push_str(completion.ghost.as_str()); + state.spine.input.cursor = state.spine.input.input.chars().count(); + } + } + should_refresh_hint = true; + should_refresh_completion = true; + } + SpineAction::CycleCompletion => { + if let Some(completion) = state.spine.completion.as_ref() { + if completion.candidates.len() > 1 { + let next = (completion.selected + 1) % completion.candidates.len(); + state.spine.completion = select_spine_completion( + state.spine.mode, + state.spine.input.input.as_str(), + state.spine.input.cursor, + next, + ); + } + } + } + SpineAction::InsertChar(ch) => { + state.spine.input.insert_char(ch); + should_refresh_hint = true; + should_refresh_completion = true; + } + SpineAction::Backspace => { + state.spine.input.backspace(); + should_refresh_hint = true; + should_refresh_completion = true; + } + SpineAction::Delete => { + state.spine.input.delete(); + should_refresh_hint = true; + should_refresh_completion = true; + } + SpineAction::MoveCursorLeft => { + state.spine.input.move_left(); + should_refresh_completion = true; + } + SpineAction::MoveCursorRight => { + state.spine.input.move_right(); + should_refresh_completion = true; + } + SpineAction::MoveCursorHome => { + state.spine.input.move_home(); + should_refresh_completion = true; + } + SpineAction::MoveCursorEnd => { + state.spine.input.move_end(); + should_refresh_completion = true; + } + SpineAction::Clear => { + state.spine.input.clear(); + state.spine.completion = None; + should_refresh_hint = true; + should_refresh_completion = true; + } + } + + if should_refresh_completion { + refresh_spine_completion(state); + } + if should_refresh_hint { + refresh_spine_hint(state); + } +} + +fn handle_time_cursor_action(state: &mut AppState, action: TimeCursorAction) { + if !state.temporal_focus.is_active() { + match action { + TimeCursorAction::ToggleFollow => { + let Some(scope) = state.temporal_focus_scope() else { + return; + }; + state.temporal_focus = scope; + } + _ => return, + } + } + + let bounds = time_bounds_for_focus(state); + match action { + TimeCursorAction::Step { delta_ms } => { + let Some((min, max)) = bounds else { + return; + }; + let next = state.time_cursor.t_ms.saturating_add(delta_ms); + state.time_cursor.t_ms = next.clamp(min, max); + state.time_cursor.mode = TimeCursorMode::Scrub; + } + TimeCursorAction::JumpToLive => { + state.time_cursor.mode = TimeCursorMode::Live; + if let Some((_, max)) = bounds { + state.time_cursor.t_ms = max; + } + } + TimeCursorAction::ToggleFollow => { + if matches!(state.time_cursor.mode, TimeCursorMode::Live) { + state.time_cursor.mode = TimeCursorMode::Scrub; + } else { + state.time_cursor.mode = TimeCursorMode::Live; + if let Some((_, max)) = bounds { + state.time_cursor.t_ms = max; + } + } + } + } + + if let Some((min, max)) = bounds { + state.time_cursor.t_ms = state.time_cursor.t_ms.clamp(min, max); + } + + if matches!(state.time_cursor.mode, TimeCursorMode::Live) { + state.temporal_focus = TemporalFocus::None; + } +} + +fn time_bounds_for_focus(state: &AppState) -> Option<(i64, i64)> { + match &state.temporal_focus { + TemporalFocus::None => None, + TemporalFocus::Cluster { id } => time_bounds_for_cluster(state, id, None), + TemporalFocus::Agent { + cluster_id, + agent_id, + } => time_bounds_for_agent_microscope(state, cluster_id, agent_id) + .or_else(|| time_bounds_for_cluster(state, cluster_id, Some(agent_id.as_str()))), + } +} + +fn time_bounds_for_agent_microscope( + state: &AppState, + cluster_id: &str, + agent_id: &str, +) -> Option<(i64, i64)> { + let key = AgentKey::new(cluster_id.to_string(), agent_id.to_string()); + let entry = state.agent_microscopes.get(&key)?; + let mut min: Option = None; + let mut max: Option = None; + for line in entry.logs_time.iter() { + min = Some(min.map_or(line.timestamp, |value| value.min(line.timestamp))); + max = Some(max.map_or(line.timestamp, |value| value.max(line.timestamp))); + } + match (min, max) { + (Some(min), Some(max)) => Some((min, max)), + _ => None, + } +} + +fn time_bounds_for_cluster( + state: &AppState, + cluster_id: &str, + agent_id: Option<&str>, +) -> Option<(i64, i64)> { + let cluster_state = state.clusters.get(cluster_id)?; + let mut min: Option = None; + let mut max: Option = None; + let update = |ts: i64, min: &mut Option, max: &mut Option| { + *min = Some(min.map_or(ts, |value| value.min(ts))); + *max = Some(max.map_or(ts, |value| value.max(ts))); + }; + + for line in cluster_state.logs_time.iter() { + if let Some(agent_id) = agent_id { + let matches_agent = + line.agent.as_deref() == Some(agent_id) || line.sender.as_deref() == Some(agent_id); + if !matches_agent { + continue; + } + } + update(line.timestamp, &mut min, &mut max); + } + + if agent_id.is_none() { + for event in cluster_state.timeline_time.iter() { + update(event.timestamp, &mut min, &mut max); + } + } + + match (min, max) { + (Some(min), Some(max)) => Some((min, max)), + _ => None, + } +} + +fn build_guidance_message(prefix: Option<&str>, message: &str) -> String { + let trimmed = message.trim(); + match prefix { + Some(prefix) if trimmed.is_empty() => prefix.to_string(), + Some(prefix) => format!("{prefix} {trimmed}"), + None => trimmed.to_string(), + } +} + +fn resolve_canvas_focused_agent(state: &AppState, cluster_id: &str) -> Option { + let canvas_state = state.cluster_canvases.get(cluster_id)?; + let focused_id = canvas_state.focused_id.as_deref()?; + if let Some(layout) = canvas_state.layout.as_ref() { + if let Some(node) = layout.nodes.get(focused_id) { + if matches!(node.kind, cluster_canvas::NodeKind::Agent) { + return Some(node.id.clone()); + } + } + } + let cluster_state = state.clusters.get(cluster_id)?; + let topology = cluster_state.topology.as_ref()?; + if topology.agents.iter().any(|agent| agent.id == focused_id) { + Some(focused_id.to_string()) + } else { + None + } +} + +fn resolve_focus_target(state: &AppState) -> Option { + match state.active_screen() { + ScreenId::Agent { + cluster_id, + agent_id, + } + | ScreenId::AgentMicroscope { + cluster_id, + agent_id, + } => Some(FocusTarget::Agent { + cluster_id: cluster_id.clone(), + agent_id: agent_id.clone(), + }), + ScreenId::ClusterCanvas { id } => { + if let Some(agent_id) = resolve_canvas_focused_agent(state, id) { + return Some(FocusTarget::Agent { + cluster_id: id.clone(), + agent_id, + }); + } + Some(FocusTarget::Cluster { id: id.clone() }) + } + ScreenId::Cluster { id } => Some(FocusTarget::Cluster { id: id.clone() }), + ScreenId::Monitor => state + .monitor + .selected_cluster_id() + .map(|id| FocusTarget::Cluster { id }), + ScreenId::FleetRadar => state + .fleet_radar + .selected_cluster_id() + .map(|id| FocusTarget::Cluster { id }), + ScreenId::Launcher | ScreenId::IntentConsole => None, + } +} + +fn handle_command_action(state: &mut AppState, action: CommandAction, effects: &mut Vec) { + match action { + CommandAction::ShowToast { level, message } => { + state.toast = Some(ToastState { + message, + level, + expires_at_ms: state.now_ms.saturating_add(TOAST_DURATION_MS), + }); + } + CommandAction::SetProviderOverride { provider } => { + state.provider_override = provider; + refresh_spine_hint(state); + } + CommandAction::StartClusterFromIssue { + reference, + provider_override, + } => { + effects.push(Effect::Backend(BackendRequest::StartClusterFromIssue { + reference, + provider_override, + })); + } + CommandAction::SendGuidance { message, prefix } => { + let Some(target) = resolve_focus_target(state) else { + state.toast = Some(ToastState { + message: "No focused cluster or agent.".to_string(), + level: ToastLevel::Error, + expires_at_ms: state.now_ms.saturating_add(TOAST_DURATION_MS), + }); + return; + }; + let built = build_guidance_message(prefix.as_deref(), &message); + if built.trim().is_empty() { + state.toast = Some(ToastState { + message: "Guidance text is required.".to_string(), + level: ToastLevel::Error, + expires_at_ms: state.now_ms.saturating_add(TOAST_DURATION_MS), + }); + return; + } + let toast_message = match &target { + FocusTarget::Cluster { id } => { + effects.push(Effect::Backend(BackendRequest::SendGuidanceToCluster { + cluster_id: id.clone(), + message: built.clone(), + })); + format!("Guidance sent to cluster {id}.") + } + FocusTarget::Agent { + cluster_id, + agent_id, + } => { + effects.push(Effect::Backend(BackendRequest::SendGuidanceToAgent { + cluster_id: cluster_id.clone(), + agent_id: agent_id.clone(), + message: built.clone(), + })); + format!("Guidance sent to agent {agent_id} @ {cluster_id}.") + } + }; + state.toast = Some(ToastState { + message: toast_message, + level: ToastLevel::Success, + expires_at_ms: state.now_ms.saturating_add(TOAST_DURATION_MS), + }); + } + CommandAction::TogglePin => { + if !matches!(state.ui_variant, UiVariant::Disruptive) { + state.toast = Some(ToastState { + message: "Pinning is only available in Disruptive UI.".to_string(), + level: ToastLevel::Error, + expires_at_ms: state.now_ms.saturating_add(TOAST_DURATION_MS), + }); + return; + } + let Some(target) = resolve_focus_target(state) else { + state.toast = Some(ToastState { + message: "No focus target to pin.".to_string(), + level: ToastLevel::Error, + expires_at_ms: state.now_ms.saturating_add(TOAST_DURATION_MS), + }); + return; + }; + let message = if state.pinned_target.as_ref() == Some(&target) { + state.pinned_target = None; + format!("Unpinned {}.", target.label()) + } else { + state.pinned_target = Some(target.clone()); + format!("Pinned {}.", target.label()) + }; + state.toast = Some(ToastState { + message, + level: ToastLevel::Success, + expires_at_ms: state.now_ms.saturating_add(TOAST_DURATION_MS), + }); + } + } +} + +fn cleanup_active_screen(state: &mut AppState, effects: &mut Vec) { + let active = state.screen_stack.last().cloned(); + match active { + Some(ScreenId::Cluster { id }) | Some(ScreenId::ClusterCanvas { id }) => { + cleanup_cluster_subscriptions(state, &id, effects) + } + Some(ScreenId::Agent { + cluster_id, + agent_id, + }) => cleanup_agent_subscriptions(state, &cluster_id, &agent_id, effects), + Some(ScreenId::AgentMicroscope { + cluster_id, + agent_id, + }) => { + cleanup_agent_subscriptions(state, &cluster_id, &agent_id, effects); + cleanup_cluster_timeline_subscription(state, &cluster_id, effects); + } + _ => {} + } +} + +fn cleanup_cluster_subscriptions(state: &mut AppState, id: &str, effects: &mut Vec) { + let Some(entry) = state.clusters.get_mut(id) else { + return; + }; + if let Some(subscription_id) = entry.log_subscription.take() { + effects.push(Effect::Backend(BackendRequest::Unsubscribe { + subscription_id, + })); + } + if let Some(subscription_id) = entry.timeline_subscription.take() { + effects.push(Effect::Backend(BackendRequest::Unsubscribe { + subscription_id, + })); + } + if let Some(canvas_entry) = state.cluster_canvases.get_mut(id) { + canvas_entry.log_subscription = None; + canvas_entry.timeline_subscription = None; + } +} + +fn cleanup_cluster_timeline_subscription( + state: &mut AppState, + id: &str, + effects: &mut Vec, +) { + let Some(entry) = state.clusters.get_mut(id) else { + return; + }; + if let Some(subscription_id) = entry.timeline_subscription.take() { + effects.push(Effect::Backend(BackendRequest::Unsubscribe { + subscription_id, + })); + } + if let Some(canvas_entry) = state.cluster_canvases.get_mut(id) { + canvas_entry.timeline_subscription = None; + } +} + +fn cleanup_agent_subscriptions( + state: &mut AppState, + cluster_id: &str, + agent_id: &str, + effects: &mut Vec, +) { + let key = AgentKey::new(cluster_id.to_string(), agent_id.to_string()); + let mut unsubscribed: HashSet = HashSet::new(); + if let Some(entry) = state.agents.get_mut(&key) { + if let Some(subscription_id) = entry.log_subscription.take() { + unsubscribed.insert(subscription_id.clone()); + effects.push(Effect::Backend(BackendRequest::Unsubscribe { + subscription_id, + })); + } + } + if let Some(entry) = state.agent_microscopes.get_mut(&key) { + if let Some(subscription_id) = entry.log_subscription.take() { + if !unsubscribed.contains(&subscription_id) { + effects.push(Effect::Backend(BackendRequest::Unsubscribe { + subscription_id, + })); + } + } + } +} + +fn metrics_request_for_screen(state: &AppState) -> Option { + match state.active_screen() { + ScreenId::Monitor | ScreenId::FleetRadar => { + let ids: Vec = state + .monitor + .clusters + .iter() + .map(|cluster| cluster.id.clone()) + .collect(); + if ids.is_empty() { + None + } else { + Some(BackendRequest::ListClusterMetrics { + cluster_ids: Some(ids), + }) + } + } + ScreenId::Cluster { id } | ScreenId::ClusterCanvas { id } => { + Some(BackendRequest::ListClusterMetrics { + cluster_ids: Some(vec![id.clone()]), + }) + } + _ => None, + } +} + +#[cfg(test)] +mod tests { + use super::*; + use crate::commands; + use crate::protocol::{ + ClusterLogLine, ClusterLogLinesParams, ClusterSummary, ClusterTimelineEventsParams, + TimelineEvent, + }; + + fn radar_cluster(id: &str) -> ClusterSummary { + ClusterSummary { + id: id.to_string(), + state: "running".to_string(), + provider: None, + created_at: 0, + agent_count: 1, + message_count: 0, + cwd: None, + } + } + + fn apply_actions(mut state: AppState, actions: Vec) -> (AppState, Vec) { + let mut effects = Vec::new(); + for action in actions { + let (next_state, next_effects) = update(state, action); + state = next_state; + effects.extend(next_effects); + } + (state, effects) + } + + #[test] + fn provider_override_applies_to_issue_start() { + let state = AppState::default(); + let actions = commands::dispatch(CommandRequest::SubmitRaw { + raw: "/provider codex".to_string(), + context: state.command_context(), + }) + .expect("dispatch provider"); + let (state, _) = apply_actions(state, actions); + assert_eq!(state.provider_override, Some("codex".to_string())); + + let actions = commands::dispatch(CommandRequest::SubmitRaw { + raw: "/issue org/repo#123".to_string(), + context: state.command_context(), + }) + .expect("dispatch issue"); + let (_state, effects) = apply_actions(state, actions); + let mut found = false; + for effect in effects { + if let Effect::Backend(BackendRequest::StartClusterFromIssue { + reference, + provider_override, + }) = effect + { + found = true; + assert_eq!(reference, "org/repo#123"); + assert_eq!(provider_override, Some("codex".to_string())); + } + } + assert!(found, "expected StartClusterFromIssue effect"); + } + + #[test] + fn command_guidance_sends_to_agent_with_prefix() { + let mut state = AppState::default(); + state.screen_stack = vec![ScreenId::Agent { + cluster_id: "cluster-1".to_string(), + agent_id: "agent-1".to_string(), + }]; + + let (_state, effects) = update( + state, + Action::Command(CommandAction::SendGuidance { + message: "hi".to_string(), + prefix: Some("[nudge]".to_string()), + }), + ); + + assert!( + effects.contains(&Effect::Backend(BackendRequest::SendGuidanceToAgent { + cluster_id: "cluster-1".to_string(), + agent_id: "agent-1".to_string(), + message: "[nudge] hi".to_string(), + })) + ); + } + + #[test] + fn pin_toggles_pinned_target() { + let mut state = AppState::default(); + state.ui_variant = UiVariant::Disruptive; + state.screen_stack = vec![ScreenId::FleetRadar]; + state + .fleet_radar + .set_clusters(vec![radar_cluster("cluster-1")], 0); + + let (state, _) = update(state, Action::Command(CommandAction::TogglePin)); + assert_eq!( + state.pinned_target, + Some(FocusTarget::Cluster { + id: "cluster-1".to_string() + }) + ); + + let (state, _) = update(state, Action::Command(CommandAction::TogglePin)); + assert_eq!(state.pinned_target, None); + } + + #[test] + fn spine_state_editing() { + let mut state = AppState::default(); + + let (next, _) = update(state, Action::Spine(SpineAction::InsertChar('a'))); + state = next; + assert_eq!(state.spine.input.input, "a"); + assert_eq!(state.spine.input.cursor, 1); + + let (next, _) = update(state, Action::Spine(SpineAction::InsertChar('b'))); + state = next; + assert_eq!(state.spine.input.input, "ab"); + assert_eq!(state.spine.input.cursor, 2); + + let (next, _) = update(state, Action::Spine(SpineAction::InsertChar('c'))); + state = next; + assert_eq!(state.spine.input.input, "abc"); + assert_eq!(state.spine.input.cursor, 3); + + let (next, _) = update(state, Action::Spine(SpineAction::MoveCursorLeft)); + state = next; + assert_eq!(state.spine.input.cursor, 2); + + let (next, _) = update(state, Action::Spine(SpineAction::Backspace)); + state = next; + assert_eq!(state.spine.input.input, "ac"); + assert_eq!(state.spine.input.cursor, 1); + + let (next, _) = update(state, Action::Spine(SpineAction::Delete)); + state = next; + assert_eq!(state.spine.input.input, "a"); + assert_eq!(state.spine.input.cursor, 1); + + let (next, _) = update(state, Action::Spine(SpineAction::MoveCursorHome)); + state = next; + assert_eq!(state.spine.input.cursor, 0); + + let (next, _) = update(state, Action::Spine(SpineAction::InsertChar('z'))); + state = next; + assert_eq!(state.spine.input.input, "za"); + assert_eq!(state.spine.input.cursor, 1); + + let (next, _) = update(state, Action::Spine(SpineAction::MoveCursorEnd)); + state = next; + assert_eq!(state.spine.input.cursor, 2); + + let (next, _) = update(state, Action::Spine(SpineAction::MoveCursorRight)); + state = next; + assert_eq!(state.spine.input.cursor, 2); + + let (next, _) = update(state, Action::Spine(SpineAction::Clear)); + state = next; + assert_eq!(state.spine.input.input, ""); + assert_eq!(state.spine.input.cursor, 0); + } + + #[test] + fn spine_accept_completion_appends_ghost() { + let mut state = AppState::default(); + state.spine.mode = SpineMode::Command; + + let (state, _) = update(state, Action::Spine(SpineAction::InsertChar('p'))); + let completion = state.spine.completion.as_ref().expect("completion"); + assert_eq!(completion.ghost, "rovider"); + + let (state, _) = update(state, Action::Spine(SpineAction::AcceptCompletion)); + assert_eq!(state.spine.input.input, "provider"); + assert!(state.spine.completion.is_none()); + } + + #[test] + fn spine_cycle_completion_updates_ghost() { + let mut state = AppState::default(); + state.spine.mode = SpineMode::Command; + + let (state, _) = update(state, Action::Spine(SpineAction::InsertChar('i'))); + let completion = state.spine.completion.as_ref().expect("completion"); + assert_eq!(completion.ghost, "ssue"); + + let (state, _) = update(state, Action::Spine(SpineAction::CycleCompletion)); + let completion = state.spine.completion.as_ref().expect("completion"); + assert_eq!(completion.ghost, "nterrupt"); + } + + #[test] + fn spine_cancel_resets_mode_and_input() { + let mut state = AppState::default(); + state.spine.mode = SpineMode::Command; + state.spine.input.input = "help".to_string(); + state.spine.input.cursor = 4; + state.spine.completion = Some(SpineCompletion { + candidates: vec!["help".to_string()], + selected: 0, + ghost: "er".to_string(), + }); + + let (state, effects) = update(state, Action::Spine(SpineAction::Cancel)); + assert!(effects.is_empty()); + assert_eq!(state.spine.mode, SpineMode::Intent); + assert_eq!(state.spine.input.input, ""); + assert_eq!(state.spine.input.cursor, 0); + assert!(state.spine.completion.is_none()); + } + + #[test] + fn spine_submit_command_emits_command_effect() { + let mut state = AppState::default(); + state.spine.mode = SpineMode::Command; + state.spine.input.input = "help ".to_string(); + state.spine.input.cursor = 5; + + let (state, effects) = update(state, Action::Spine(SpineAction::Submit)); + assert!(effects.iter().any(|effect| { + matches!( + effect, + Effect::Command(CommandRequest::SubmitRaw { raw, .. }) if raw == "/help " + ) + })); + assert_eq!(state.spine.mode, SpineMode::Intent); + assert_eq!(state.spine.input.input, ""); + assert!(state.spine.completion.is_none()); + } + + #[test] + fn spine_submit_intent_starts_cluster() { + let mut state = AppState::default(); + state.spine.mode = SpineMode::Intent; + state.spine.input.input = "launch".to_string(); + state.spine.input.cursor = 6; + + let (state, effects) = update(state, Action::Spine(SpineAction::Submit)); + assert!( + effects.contains(&Effect::Backend(BackendRequest::StartClusterFromText { + text: "launch".to_string(), + provider_override: None, + })) + ); + assert_eq!(state.spine.mode, SpineMode::Intent); + assert_eq!(state.spine.input.input, ""); + } + + #[test] + fn spine_submit_whisper_cluster_sends_guidance() { + let mut state = AppState::default(); + state.screen_stack = vec![ScreenId::Cluster { + id: "cluster-1".to_string(), + }]; + state.spine.mode = SpineMode::WhisperCluster; + state.spine.input.input = "ping".to_string(); + state.spine.input.cursor = 4; + + let (state, effects) = update(state, Action::Spine(SpineAction::Submit)); + assert!( + effects.contains(&Effect::Backend(BackendRequest::SendGuidanceToCluster { + cluster_id: "cluster-1".to_string(), + message: "ping".to_string(), + })) + ); + assert_eq!(state.spine.mode, SpineMode::Intent); + assert_eq!(state.spine.input.input, ""); + } + + #[test] + fn spine_submit_whisper_agent_sends_guidance() { + let mut state = AppState::default(); + state.screen_stack = vec![ScreenId::Agent { + cluster_id: "cluster-1".to_string(), + agent_id: "agent-1".to_string(), + }]; + state.spine.mode = SpineMode::WhisperAgent; + state.spine.input.input = "ping".to_string(); + state.spine.input.cursor = 4; + + let (state, effects) = update(state, Action::Spine(SpineAction::Submit)); + assert!( + effects.contains(&Effect::Backend(BackendRequest::SendGuidanceToAgent { + cluster_id: "cluster-1".to_string(), + agent_id: "agent-1".to_string(), + message: "ping".to_string(), + })) + ); + assert_eq!(state.spine.mode, SpineMode::Intent); + assert_eq!(state.spine.input.input, ""); + } + + #[test] + fn navigation_to_microscope_sets_spine_mode() { + let state = AppState::default(); + let (state, _) = update( + state, + Action::Navigate(NavigationAction::Push(ScreenId::AgentMicroscope { + cluster_id: "cluster-1".to_string(), + agent_id: "agent-1".to_string(), + })), + ); + assert_eq!(state.spine.mode, SpineMode::WhisperAgent); + } + + #[test] + fn navigation_to_microscope_subscribes_timeline() { + let state = AppState::default(); + let (_, effects) = update( + state, + Action::Navigate(NavigationAction::Push(ScreenId::AgentMicroscope { + cluster_id: "cluster-1".to_string(), + agent_id: "agent-1".to_string(), + })), + ); + assert!( + effects.contains(&Effect::Backend(BackendRequest::SubscribeClusterTimeline { + cluster_id: "cluster-1".to_string(), + })) + ); + } + + #[test] + fn spine_submit_whisper_agent_sends_guidance_from_microscope() { + let mut state = AppState::default(); + state.screen_stack = vec![ScreenId::AgentMicroscope { + cluster_id: "cluster-1".to_string(), + agent_id: "agent-1".to_string(), + }]; + state.spine.mode = SpineMode::WhisperAgent; + state.spine.input.input = "ping".to_string(); + state.spine.input.cursor = 4; + + let (state, effects) = update(state, Action::Spine(SpineAction::Submit)); + assert!( + effects.contains(&Effect::Backend(BackendRequest::SendGuidanceToAgent { + cluster_id: "cluster-1".to_string(), + agent_id: "agent-1".to_string(), + message: "ping".to_string(), + })) + ); + assert_eq!(state.spine.mode, SpineMode::WhisperAgent); + assert_eq!(state.spine.input.input, ""); + } + + #[test] + fn time_cursor_live_updates() { + let mut state = AppState::default(); + state.time_cursor.mode = TimeCursorMode::Live; + state.time_cursor.t_ms = 10; + + let cluster_id = "cluster-1".to_string(); + let subscription_id = "sub-logs".to_string(); + state + .clusters + .entry(cluster_id.clone()) + .or_default() + .log_subscription = Some(subscription_id.clone()); + + handle_backend_notification( + &mut state, + BackendNotification::ClusterLogLines(ClusterLogLinesParams { + subscription_id, + cluster_id: cluster_id.clone(), + lines: vec![ + ClusterLogLine { + id: "line-1".to_string(), + timestamp: 100, + text: "hello".to_string(), + agent: None, + role: None, + sender: None, + }, + ClusterLogLine { + id: "line-2".to_string(), + timestamp: 150, + text: "world".to_string(), + agent: None, + role: None, + sender: None, + }, + ], + dropped_count: None, + }), + ); + assert_eq!(state.time_cursor.t_ms, 150); + + handle_backend_notification( + &mut state, + BackendNotification::ClusterTimelineEvents(ClusterTimelineEventsParams { + subscription_id: "sub-timeline".to_string(), + cluster_id, + events: vec![TimelineEvent { + id: "event-1".to_string(), + timestamp: 175, + topic: "ISSUE_OPENED".to_string(), + label: "opened".to_string(), + approved: None, + sender: None, + }], + }), + ); + assert_eq!(state.time_cursor.t_ms, 175); + } + + #[test] + fn time_cursor_scrub_does_not_update() { + let mut state = AppState::default(); + state.time_cursor.mode = TimeCursorMode::Scrub; + state.time_cursor.t_ms = 200; + + let cluster_id = "cluster-2".to_string(); + let subscription_id = "sub-logs".to_string(); + state + .clusters + .entry(cluster_id.clone()) + .or_default() + .log_subscription = Some(subscription_id.clone()); + + handle_backend_notification( + &mut state, + BackendNotification::ClusterLogLines(ClusterLogLinesParams { + subscription_id, + cluster_id, + lines: vec![ClusterLogLine { + id: "line-3".to_string(), + timestamp: 500, + text: "late".to_string(), + agent: None, + role: None, + sender: None, + }], + dropped_count: None, + }), + ); + assert_eq!(state.time_cursor.t_ms, 200); + } + + fn sample_log_line(id: &str, timestamp: i64) -> ClusterLogLine { + ClusterLogLine { + id: id.to_string(), + timestamp, + text: "log".to_string(), + agent: None, + role: None, + sender: None, + } + } + + #[test] + fn time_cursor_step_clamps_and_enters_scrub() { + let mut state = AppState::default(); + state.temporal_focus = TemporalFocus::Cluster { + id: "cluster-1".to_string(), + }; + state.time_cursor.mode = TimeCursorMode::Live; + state.time_cursor.t_ms = 200; + + let mut cluster_state = cluster::State::default(); + cluster_state.push_log_lines( + vec![sample_log_line("l1", 100), sample_log_line("l2", 200)], + None, + ); + state + .clusters + .insert("cluster-1".to_string(), cluster_state); + + let (state, _) = update( + state, + Action::TimeCursor(TimeCursorAction::Step { + delta_ms: -TIME_SCRUB_STEP_MS, + }), + ); + assert_eq!(state.time_cursor.mode, TimeCursorMode::Scrub); + assert_eq!(state.time_cursor.t_ms, 100); + } + + #[test] + fn time_cursor_jump_and_toggle_follow() { + let mut state = AppState::default(); + state.screen_stack = vec![ScreenId::ClusterCanvas { + id: "cluster-2".to_string(), + }]; + state.temporal_focus = TemporalFocus::Cluster { + id: "cluster-2".to_string(), + }; + state.time_cursor.mode = TimeCursorMode::Scrub; + state.time_cursor.t_ms = 120; + + let mut cluster_state = cluster::State::default(); + cluster_state.push_log_lines( + vec![sample_log_line("l1", 100), sample_log_line("l2", 250)], + None, + ); + state + .clusters + .insert("cluster-2".to_string(), cluster_state); + + let (state, _) = update(state, Action::TimeCursor(TimeCursorAction::JumpToLive)); + assert_eq!(state.time_cursor.mode, TimeCursorMode::Live); + assert_eq!(state.time_cursor.t_ms, 250); + + let (state, _) = update(state, Action::TimeCursor(TimeCursorAction::ToggleFollow)); + assert_eq!(state.time_cursor.mode, TimeCursorMode::Scrub); + + let (state, _) = update(state, Action::TimeCursor(TimeCursorAction::ToggleFollow)); + assert_eq!(state.time_cursor.mode, TimeCursorMode::Live); + assert_eq!(state.time_cursor.t_ms, 250); + } + + #[test] + fn time_cursor_large_step_clamps_to_max() { + let mut state = AppState::default(); + state.temporal_focus = TemporalFocus::Cluster { + id: "cluster-3".to_string(), + }; + state.time_cursor.mode = TimeCursorMode::Scrub; + state.time_cursor.t_ms = 150; + + let mut cluster_state = cluster::State::default(); + cluster_state.push_log_lines( + vec![sample_log_line("l1", 100), sample_log_line("l2", 220)], + None, + ); + state + .clusters + .insert("cluster-3".to_string(), cluster_state); + + let (state, _) = update( + state, + Action::TimeCursor(TimeCursorAction::Step { + delta_ms: TIME_SCRUB_STEP_LARGE_MS, + }), + ); + assert_eq!(state.time_cursor.t_ms, 220); + } + + #[test] + fn radar_camera_centering() { + let mut state = AppState::default(); + state.ui_variant = UiVariant::Disruptive; + state.now_ms = 10_000; + state.fleet_radar.set_clusters( + vec![radar_cluster("west"), radar_cluster("east")], + state.now_ms, + ); + state + .fleet_radar + .layout_angles + .insert("west".to_string(), std::f64::consts::PI); + state + .fleet_radar + .layout_angles + .insert("east".to_string(), 0.0); + state.fleet_radar.selected = 0; + + let (state, _) = update( + state, + Action::Screen(ScreenAction::FleetRadar(radar::Action::CenterOnSelection)), + ); + assert!(state.camera_target.0 < 0.0); + + let (state, _) = update( + state, + Action::Screen(ScreenAction::FleetRadar(radar::Action::MoveSelection { + direction: radar::Direction::Right, + speed: radar::MoveSpeed::Step, + })), + ); + assert_eq!( + state.fleet_radar.selected_cluster_id().as_deref(), + Some("east") + ); + assert!(state.camera_target.0 > 0.0); + + let (state, _) = update( + state, + Action::Screen(ScreenAction::FleetRadar(radar::Action::ResetView)), + ); + assert_eq!(state.camera, Camera::default()); + assert_eq!(state.camera_target, (0.0, 0.0)); + assert_eq!(state.camera_velocity, (0.0, 0.0)); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/app/spine_completion.rs b/tui-rs/crates/zeroshot-tui/src/app/spine_completion.rs new file mode 100644 index 00000000..3b430df2 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/app/spine_completion.rs @@ -0,0 +1,179 @@ +use crate::commands::VALID_PROVIDERS; + +use super::{SpineCompletion, SpineMode}; + +const COMMAND_CANDIDATES: [&str; 10] = [ + "help", + "monitor", + "issue", + "provider", + "guide", + "nudge", + "interrupt", + "pin", + "quit", + "exit", +]; + +#[derive(Debug, Clone, PartialEq, Eq)] +struct CompletionContext { + prefix: String, + candidates: Vec, +} + +pub fn build_spine_completion( + mode: SpineMode, + input: &str, + cursor: usize, +) -> Option { + if cursor != input.chars().count() { + return None; + } + + match mode { + SpineMode::Command => build_command_completion(input), + SpineMode::Intent | SpineMode::WhisperCluster | SpineMode::WhisperAgent => None, + } +} + +pub fn select_spine_completion( + mode: SpineMode, + input: &str, + cursor: usize, + selected: usize, +) -> Option { + let context = completion_context(mode, input, cursor)?; + build_completion(&context.prefix, &context.candidates, selected) +} + +fn build_command_completion(input: &str) -> Option { + let context = completion_context(SpineMode::Command, input, input.chars().count())?; + build_completion(&context.prefix, &context.candidates, 0) +} + +fn completion_context(mode: SpineMode, input: &str, cursor: usize) -> Option { + if cursor != input.chars().count() { + return None; + } + match mode { + SpineMode::Command => command_completion_context(input), + SpineMode::Intent | SpineMode::WhisperCluster | SpineMode::WhisperAgent => None, + } +} + +fn command_completion_context(input: &str) -> Option { + let trimmed = input.trim_start_matches('/'); + if trimmed.is_empty() { + return None; + } + + let ends_with_space = trimmed.ends_with(' '); + let mut parts = trimmed.split_whitespace(); + let command = parts.next().unwrap_or(""); + if command.is_empty() { + return None; + } + let args = parts.collect::>(); + + let (prefix, candidates, allow_empty): (&str, &[&str], bool) = if args.is_empty() + && !ends_with_space + { + (command, &COMMAND_CANDIDATES, false) + } else if command.eq_ignore_ascii_case("provider") && (!args.is_empty() || ends_with_space) { + let prefix = if ends_with_space { + "" + } else { + args.last().copied().unwrap_or("") + }; + (prefix, &VALID_PROVIDERS, true) + } else { + return None; + }; + + let mut candidates = prefix_matches(prefix, candidates, allow_empty); + if args.is_empty() && !ends_with_space && prefix.chars().count() == 1 { + // Keep `/p` completion unambiguous; "pin" still appears for `pi`. + candidates.retain(|candidate| candidate != "pin"); + } + if candidates.is_empty() { + return None; + } + + Some(CompletionContext { + prefix: prefix.to_string(), + candidates, + }) +} + +fn prefix_matches(prefix: &str, candidates: &[&str], allow_empty: bool) -> Vec { + if prefix.is_empty() && !allow_empty { + return Vec::new(); + } + + let prefix_lower = prefix.to_lowercase(); + candidates + .iter() + .filter(|candidate| candidate.starts_with(prefix_lower.as_str())) + .filter(|candidate| candidate.chars().count() > prefix_lower.chars().count()) + .map(|candidate| candidate.to_string()) + .collect() +} + +fn build_completion( + prefix: &str, + candidates: &[String], + selected: usize, +) -> Option { + if candidates.is_empty() || selected >= candidates.len() { + return None; + } + + let candidate = candidates.get(selected)?; + let ghost = suffix_after_prefix(prefix, candidate); + if ghost.is_empty() { + return None; + } + + Some(SpineCompletion { + candidates: candidates.to_vec(), + selected, + ghost, + }) +} + +fn suffix_after_prefix(prefix: &str, candidate: &str) -> String { + let prefix_len = prefix.chars().count(); + candidate.chars().skip(prefix_len).collect() +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn command_prefix_suggests_provider() { + let completion = build_spine_completion(SpineMode::Command, "p", 1).expect("completion"); + assert_eq!(completion.ghost, "rovider"); + } + + #[test] + fn provider_arg_suggests_known_provider() { + let completion = + build_spine_completion(SpineMode::Command, "provider c", 10).expect("completion"); + assert_eq!(completion.ghost, "laude"); + } + + #[test] + fn empty_prefix_does_not_suggest_command_names() { + let completion = build_spine_completion(SpineMode::Command, "", 0); + assert!(completion.is_none()); + } + + #[test] + fn select_completion_cycles_candidates() { + let completion = build_spine_completion(SpineMode::Command, "i", 1).expect("completion"); + let cycled = select_spine_completion(SpineMode::Command, "i", 1, 1).expect("cycle"); + assert_ne!(completion.ghost, cycled.ghost); + assert_eq!(cycled.ghost, "nterrupt"); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/app/spine_hint.rs b/tui-rs/crates/zeroshot-tui/src/app/spine_hint.rs new file mode 100644 index 00000000..ee374951 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/app/spine_hint.rs @@ -0,0 +1,346 @@ +use crate::commands::{parse, VALID_PROVIDERS}; +use crate::protocol::GuidanceDeliveryResult; + +use super::{ + detect_issue_reference, resolve_focus_target, resolve_spine_agent_target, + resolve_spine_cluster_target, AgentKey, AppState, SpineMode, ToastLevel, UiVariant, + ZoomStackContext, +}; + +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub enum SpineHintTone { + Muted, + Info, + Success, + Error, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct SpineHint { + pub text: String, + pub tone: SpineHintTone, +} + +impl SpineHint { + pub fn new(text: impl Into, tone: SpineHintTone) -> Self { + Self { + text: text.into(), + tone, + } + } + + pub fn empty() -> Self { + Self::new("", SpineHintTone::Muted) + } + + pub fn is_empty(&self) -> bool { + self.text.is_empty() + } + + pub fn from_toast(text: String, level: ToastLevel) -> Self { + let tone = match level { + ToastLevel::Info => SpineHintTone::Info, + ToastLevel::Success => SpineHintTone::Success, + ToastLevel::Error => SpineHintTone::Error, + }; + Self::new(text, tone) + } +} + +impl Default for SpineHint { + fn default() -> Self { + Self::empty() + } +} + +pub fn compute_spine_hint(state: &AppState) -> SpineHint { + match state.spine.mode { + SpineMode::Command => command_hint(state), + SpineMode::Intent => intent_hint(state), + SpineMode::WhisperCluster => whisper_cluster_hint(state), + SpineMode::WhisperAgent => whisper_agent_hint(state), + } +} + +fn command_hint(state: &AppState) -> SpineHint { + let trimmed = state.spine.input.input.trim(); + if trimmed.is_empty() { + return SpineHint::new(command_help_line(), SpineHintTone::Muted); + } + + let raw = if trimmed.starts_with('/') { + trimmed.to_string() + } else { + format!("/{trimmed}") + }; + + match parse(&raw) { + Ok(parsed) => match parsed.name.as_str() { + "help" => SpineHint::new(command_help_line(), SpineHintTone::Info), + "monitor" => { + let label = if matches!(state.ui_variant, UiVariant::Disruptive) { + "Fleet Radar" + } else { + "Monitor" + }; + SpineHint::new(format!("Open {label}"), SpineHintTone::Info) + } + "issue" => { + let Some(reference) = parsed.args.first() else { + return SpineHint::new(issue_usage(), SpineHintTone::Error); + }; + SpineHint::new( + format!("Start cluster from issue {reference}"), + SpineHintTone::Info, + ) + } + "guide" => guidance_command_hint(state, &parsed.args, "Guide", true), + "nudge" => guidance_command_hint(state, &parsed.args, "Nudge", true), + "interrupt" => guidance_command_hint(state, &parsed.args, "Interrupt", false), + "pin" => pin_command_hint(state), + "provider" => provider_hint(&parsed.args), + "quit" | "exit" => SpineHint::new("Quit TUI", SpineHintTone::Info), + other => SpineHint::new( + format!("Unknown command: {other}. Try /help."), + SpineHintTone::Error, + ), + }, + Err(err) => SpineHint::new(err.to_string(), SpineHintTone::Error), + } +} + +fn provider_hint(args: &[String]) -> SpineHint { + let Some(name) = args.first() else { + return SpineHint::new(provider_usage(), SpineHintTone::Error); + }; + let normalized = name.to_lowercase(); + if !VALID_PROVIDERS.contains(&normalized.as_str()) { + return SpineHint::new( + format!( + "Unknown provider '{name}'. Use one of: {}", + VALID_PROVIDERS.join(", ") + ), + SpineHintTone::Error, + ); + } + SpineHint::new( + format!("Set provider override to {normalized}"), + SpineHintTone::Info, + ) +} + +fn command_help_line() -> String { + "Commands: /help /monitor /issue /provider /guide /nudge /interrupt [text] /pin /quit /exit".to_string() +} + +fn provider_usage() -> String { + format!("Usage: /provider <{}>", VALID_PROVIDERS.join("|")) +} + +fn issue_usage() -> &'static str { + "Usage: /issue " +} + +fn guidance_command_hint( + state: &AppState, + args: &[String], + verb: &str, + require_text: bool, +) -> SpineHint { + if require_text && args.is_empty() { + return SpineHint::new( + format!("Usage: /{} ", verb.to_lowercase()), + SpineHintTone::Error, + ); + } + let Some(target) = resolve_focus_target(state) else { + return SpineHint::new( + "Select a cluster or agent to guide.".to_string(), + SpineHintTone::Error, + ); + }; + let label = target.label(); + SpineHint::new(format!("{verb} {label}"), SpineHintTone::Info) +} + +fn pin_command_hint(state: &AppState) -> SpineHint { + if !matches!(state.ui_variant, UiVariant::Disruptive) { + return SpineHint::new( + "Pinning is only available in Disruptive UI.".to_string(), + SpineHintTone::Error, + ); + } + let Some(target) = resolve_focus_target(state) else { + return SpineHint::new( + "Select a cluster or agent to pin.".to_string(), + SpineHintTone::Error, + ); + }; + let action = if state.pinned_target.as_ref() == Some(&target) { + "Unpin" + } else { + "Pin" + }; + SpineHint::new(format!("{action} {}", target.label()), SpineHintTone::Info) +} + +fn intent_hint(state: &AppState) -> SpineHint { + if !matches!(state.zoom_stack_context(), ZoomStackContext::Root) { + return SpineHint::empty(); + } + + let trimmed = state.spine.input.input.trim(); + if trimmed.is_empty() { + return SpineHint::empty(); + } + + let mut hint = if detect_issue_reference(trimmed).is_some() { + SpineHint::new("Start cluster from issue", SpineHintTone::Info) + } else { + SpineHint::new("Start cluster from text", SpineHintTone::Info) + }; + + if let Some(provider) = state.provider_override.as_deref() { + hint.text = format!("{} (provider: {provider})", hint.text); + } + + hint +} + +fn whisper_cluster_hint(state: &AppState) -> SpineHint { + let Some(cluster_id) = resolve_spine_cluster_target(state) else { + return SpineHint::empty(); + }; + SpineHint::new( + format!("Whisper to cluster {cluster_id}"), + SpineHintTone::Info, + ) +} + +fn whisper_agent_hint(state: &AppState) -> SpineHint { + let Some((cluster_id, agent_id)) = resolve_spine_agent_target(state) else { + return SpineHint::empty(); + }; + let mut hint = SpineHint::new( + format!("Whisper to agent {agent_id} @ {cluster_id}"), + SpineHintTone::Info, + ); + if let Some(status) = guidance_status_hint(state, &cluster_id, &agent_id) { + hint.text = format!("{} ({status})", hint.text); + } + hint +} + +fn guidance_status_hint( + state: &AppState, + cluster_id: &str, + agent_id: &str, +) -> Option<&'static str> { + let key = AgentKey::new(cluster_id, agent_id); + let agent_state = state.agents.get(&key)?; + let result = agent_state.last_guidance.as_ref()?; + guidance_status(result) +} + +fn guidance_status(result: &GuidanceDeliveryResult) -> Option<&'static str> { + match result.status.to_lowercase().as_str() { + "injected" => Some("likely injected"), + "queued" => Some("likely queued"), + _ => None, + } +} + +#[cfg(test)] +mod tests { + use super::*; + use crate::app::{ScreenId, SpineAction}; + use crate::protocol::GuidanceDeliveryResult; + use crate::screens::agent; + + #[test] + fn command_provider_missing_arg_shows_usage() { + let mut state = AppState::default(); + state.spine.mode = SpineMode::Command; + state.spine.input.input = "provider".to_string(); + + let hint = compute_spine_hint(&state); + + assert_eq!(hint.tone, SpineHintTone::Error); + assert!(hint.text.contains("Usage: /provider")); + } + + #[test] + fn command_unknown_shows_error() { + let mut state = AppState::default(); + state.spine.mode = SpineMode::Command; + state.spine.input.input = "nope".to_string(); + + let hint = compute_spine_hint(&state); + + assert_eq!(hint.tone, SpineHintTone::Error); + assert!(hint.text.contains("Unknown command")); + } + + #[test] + fn intent_issue_prediction() { + let mut state = AppState::default(); + state.spine.mode = SpineMode::Intent; + state.spine.input.input = "123".to_string(); + + let hint = compute_spine_hint(&state); + + assert!(hint.text.contains("Start cluster from issue")); + } + + #[test] + fn intent_text_prediction() { + let mut state = AppState::default(); + state.spine.mode = SpineMode::Intent; + state.spine.input.input = "Implement X".to_string(); + + let hint = compute_spine_hint(&state); + + assert!(hint.text.contains("Start cluster from text")); + } + + #[test] + fn whisper_agent_includes_delivery_hint() { + let mut state = AppState::default(); + state.spine.mode = SpineMode::WhisperAgent; + state.screen_stack = vec![ScreenId::AgentMicroscope { + cluster_id: "cluster-1".to_string(), + agent_id: "agent-1".to_string(), + }]; + let mut agent_state = agent::State::default(); + agent_state.last_guidance = Some(GuidanceDeliveryResult { + status: "queued".to_string(), + reason: None, + method: Some("pty".to_string()), + task_id: None, + }); + state.agents.insert( + AgentKey::new("cluster-1".to_string(), "agent-1".to_string()), + agent_state, + ); + + let hint = compute_spine_hint(&state); + + assert!(hint.text.contains("agent-1")); + assert!(hint.text.contains("cluster-1")); + assert!(hint.text.contains("queued")); + } + + #[test] + fn spine_action_insert_char_updates_hint_without_backend_effects() { + let mut state = AppState::default(); + state.spine.mode = SpineMode::Intent; + + let (next, effects) = crate::app::update( + state, + crate::app::Action::Spine(SpineAction::InsertChar('1')), + ); + + assert!(effects.is_empty()); + assert!(next.spine.hint.text.contains("Start cluster from issue")); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/backend/framing.rs b/tui-rs/crates/zeroshot-tui/src/backend/framing.rs new file mode 100644 index 00000000..b8dc9204 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/backend/framing.rs @@ -0,0 +1,130 @@ +const HEADER_DELIMITER: &[u8] = b"\r\n\r\n"; +const MAX_HEADER_BYTES: usize = 8 * 1024; + +pub const MAX_FRAME_SIZE: usize = 10 * 1024 * 1024; + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum FrameError { + MissingContentLength, + InvalidContentLength(String), + InvalidHeader(String), + FrameTooLarge(usize), + HeaderTooLarge(usize), +} + +impl std::fmt::Display for FrameError { + fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { + match self { + FrameError::MissingContentLength => write!(f, "Missing Content-Length header"), + FrameError::InvalidContentLength(value) => { + write!(f, "Invalid Content-Length: {value}") + } + FrameError::InvalidHeader(value) => write!(f, "Invalid header: {value}"), + FrameError::FrameTooLarge(size) => write!(f, "Frame too large: {size} bytes"), + FrameError::HeaderTooLarge(size) => write!(f, "Header too large: {size} bytes"), + } + } +} + +impl std::error::Error for FrameError {} + +pub struct FrameEncoder; + +impl FrameEncoder { + pub fn encode(payload: &[u8]) -> Result, FrameError> { + if payload.len() > MAX_FRAME_SIZE { + return Err(FrameError::FrameTooLarge(payload.len())); + } + let mut framed = Vec::with_capacity(payload.len() + 64); + let header = format!("Content-Length: {}\r\n\r\n", payload.len()); + framed.extend_from_slice(header.as_bytes()); + framed.extend_from_slice(payload); + Ok(framed) + } +} + +#[derive(Debug, Default)] +pub struct FrameDecoder { + buffer: Vec, +} + +impl FrameDecoder { + pub fn new() -> Self { + Self { buffer: Vec::new() } + } + + pub fn push(&mut self, chunk: &[u8]) -> Result>, FrameError> { + self.buffer.extend_from_slice(chunk); + if self.buffer.len() > MAX_FRAME_SIZE + MAX_HEADER_BYTES { + return Err(FrameError::FrameTooLarge(self.buffer.len())); + } + let mut frames = Vec::new(); + loop { + let header_end = match find_header_end(&self.buffer) { + Some(index) => index, + None => break, + }; + if header_end > MAX_HEADER_BYTES { + return Err(FrameError::HeaderTooLarge(header_end)); + } + let header_bytes = &self.buffer[..header_end]; + let header_str = std::str::from_utf8(header_bytes) + .map_err(|err| FrameError::InvalidHeader(err.to_string()))?; + let content_length = parse_content_length(header_str)?; + if content_length > MAX_FRAME_SIZE { + return Err(FrameError::FrameTooLarge(content_length)); + } + let payload_start = header_end + HEADER_DELIMITER.len(); + let payload_end = payload_start + content_length; + if self.buffer.len() < payload_end { + break; + } + let payload = self.buffer[payload_start..payload_end].to_vec(); + self.buffer.drain(0..payload_end); + frames.push(payload); + } + Ok(frames) + } +} + +fn find_header_end(buffer: &[u8]) -> Option { + buffer + .windows(HEADER_DELIMITER.len()) + .position(|window| window == HEADER_DELIMITER) +} + +fn parse_content_length(header: &str) -> Result { + let mut content_length: Option = None; + for line in header.split("\r\n") { + let trimmed = line.trim(); + if trimmed.is_empty() { + continue; + } + let mut parts = trimmed.splitn(2, ':'); + let name = parts.next().unwrap_or("").trim(); + let value = parts.next().unwrap_or("").trim(); + if name.eq_ignore_ascii_case("content-length") { + if value.is_empty() { + return Err(FrameError::InvalidContentLength(value.to_string())); + } + let parsed = value + .parse::() + .map_err(|_| FrameError::InvalidContentLength(value.to_string()))?; + content_length = Some(parsed); + } + } + content_length.ok_or(FrameError::MissingContentLength) +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn parse_content_length_handles_case() { + let header = "Content-Length: 10\r\nX-Other: abc"; + assert_eq!(parse_content_length(header).unwrap(), 10); + let header_lower = "content-length: 5"; + assert_eq!(parse_content_length(header_lower).unwrap(), 5); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/backend/mod.rs b/tui-rs/crates/zeroshot-tui/src/backend/mod.rs new file mode 100644 index 00000000..2f1094d1 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/backend/mod.rs @@ -0,0 +1,126 @@ +use std::path::PathBuf; +use std::time::Duration; + +use crate::protocol::{ + ClientCapabilities, ClientInfo, ClusterLogLinesParams, ClusterTimelineEventsParams, RpcError, + ServerCapabilities, +}; + +pub mod framing; +pub mod stdio; + +pub const DEFAULT_PROTOCOL_VERSION: i64 = 1; +pub const BACKEND_PATH_ENV: &str = "ZEROSHOT_TUI_BACKEND_PATH"; +pub const DEFAULT_BACKEND_RELATIVE_PATH: &str = "lib/tui-backend/server.js"; + +#[derive(Debug, Clone)] +pub struct BackendConfig { + pub backend_path: Option, + pub protocol_version: i64, + pub client: ClientInfo, + pub capabilities: Option, + pub request_timeout: Option, +} + +impl Default for BackendConfig { + fn default() -> Self { + let client = ClientInfo { + name: "zeroshot-tui".to_string(), + version: env!("CARGO_PKG_VERSION").to_string(), + pid: Some(std::process::id() as i64), + }; + let backend_path = std::env::var(BACKEND_PATH_ENV) + .ok() + .filter(|value| !value.trim().is_empty()) + .map(PathBuf::from); + Self { + backend_path, + protocol_version: DEFAULT_PROTOCOL_VERSION, + client, + capabilities: None, + request_timeout: Some(Duration::from_secs(30)), + } + } +} + +impl BackendConfig { + pub fn with_backend_path(path: impl Into) -> Self { + let mut config = Self::default(); + config.backend_path = Some(path.into()); + config + } +} + +#[derive(Debug, Clone)] +pub struct BackendExit { + pub code: Option, + pub message: String, +} + +#[derive(Debug, Clone)] +pub enum BackendNotification { + ClusterLogLines(ClusterLogLinesParams), + ClusterTimelineEvents(ClusterTimelineEventsParams), + Unknown { + method: String, + params: Option, + }, +} + +#[derive(Debug, Clone)] +pub enum BackendEvent { + Notification(BackendNotification), + BackendExited(BackendExit), +} + +#[derive(Debug)] +pub enum BackendError { + Io(std::io::Error), + Json(serde_json::Error), + Frame(framing::FrameError), + Rpc(RpcError), + Protocol(String), + Disconnected(String), + Timeout(String), +} + +impl std::fmt::Display for BackendError { + fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { + match self { + BackendError::Io(err) => write!(f, "IO error: {err}"), + BackendError::Json(err) => write!(f, "JSON error: {err}"), + BackendError::Frame(err) => write!(f, "Frame error: {err}"), + BackendError::Rpc(err) => write!(f, "RPC error {}: {}", err.code, err.message), + BackendError::Protocol(message) => write!(f, "Protocol error: {message}"), + BackendError::Disconnected(message) => write!(f, "Backend disconnected: {message}"), + BackendError::Timeout(message) => write!(f, "Request timeout: {message}"), + } + } +} + +impl std::error::Error for BackendError {} + +impl From for BackendError { + fn from(err: std::io::Error) -> Self { + BackendError::Io(err) + } +} + +impl From for BackendError { + fn from(err: serde_json::Error) -> Self { + BackendError::Json(err) + } +} + +impl From for BackendError { + fn from(err: framing::FrameError) -> Self { + BackendError::Frame(err) + } +} + +pub trait BackendClient { + fn take_event_receiver(&mut self) -> Option>; + fn server_capabilities(&self) -> Option; + fn protocol_version(&self) -> i64; + fn shutdown(&mut self) -> Result<(), BackendError>; +} diff --git a/tui-rs/crates/zeroshot-tui/src/backend/stdio.rs b/tui-rs/crates/zeroshot-tui/src/backend/stdio.rs new file mode 100644 index 00000000..c3a910d5 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/backend/stdio.rs @@ -0,0 +1,638 @@ +use std::collections::HashMap; +use std::env; +use std::io::{Read, Write}; +use std::path::{Path, PathBuf}; +use std::process::{Child, Command, Stdio}; +use std::sync::atomic::{AtomicI64, Ordering}; +use std::sync::{mpsc, Arc, Mutex}; +use std::thread; + +use serde::de::DeserializeOwned; +use serde::Serialize; +use serde_json::Value; + +use crate::backend::framing::{FrameDecoder, FrameEncoder}; +use crate::backend::{ + BackendClient, BackendConfig, BackendError, BackendEvent, BackendExit, BackendNotification, + BACKEND_PATH_ENV, DEFAULT_BACKEND_RELATIVE_PATH, +}; +use crate::protocol::{ + ClientCapabilities, ClientInfo, ClusterLogLinesParams, ClusterTimelineEventsParams, + GetClusterSummaryParams, GetClusterSummaryResult, GetClusterTopologyParams, + GetClusterTopologyResult, InitializeParams, InitializeResult, JsonRpcId, JsonRpcRequest, + ListClusterMetricsParams, ListClusterMetricsResult, ListClustersParams, ListClustersResult, + SendGuidanceToAgentParams, SendGuidanceToAgentResult, SendGuidanceToClusterParams, + SendGuidanceToClusterResult, ServerCapabilities, StartClusterFromIssueParams, + StartClusterFromTextParams, StartClusterResult, SubscribeClusterLogsParams, + SubscribeClusterTimelineParams, SubscribeResult, UnsubscribeParams, UnsubscribeResult, +}; + +const JSONRPC_VERSION: &str = "2.0"; + +#[derive(Debug, Clone, PartialEq, Eq, Hash)] +enum RequestKey { + Number(i64), + String(String), +} + +impl RequestKey { + fn from_value(value: &Value) -> Option { + match value { + Value::Number(number) => number.as_i64().map(RequestKey::Number), + Value::String(value) => Some(RequestKey::String(value.clone())), + _ => None, + } + } +} + +enum WriterCommand { + Frame(Vec), + Shutdown, +} + +pub struct StdioBackendClient { + config: BackendConfig, + writer: Option>, + events: Option>, + pending: Arc>>>>, + next_id: AtomicI64, + read_handle: Option>, + write_handle: Option>, + child: Arc>>, + protocol_version: i64, + server_capabilities: Option, +} + +impl StdioBackendClient { + pub fn connect(config: BackendConfig) -> Result { + let backend_path = resolve_backend_path(&config)?; + let mut command = Command::new("node"); + command.arg(backend_path); + command.stdin(Stdio::piped()); + command.stdout(Stdio::piped()); + command.stderr(Stdio::inherit()); + let mut child = command.spawn()?; + let stdout = child + .stdout + .take() + .ok_or_else(|| BackendError::Protocol("Failed to capture backend stdout".into()))?; + let stdin = child + .stdin + .take() + .ok_or_else(|| BackendError::Protocol("Failed to capture backend stdin".into()))?; + + let (event_tx, event_rx) = mpsc::channel(); + let (writer_tx, writer_rx) = mpsc::channel(); + let pending: Arc>>>> = + Arc::new(Mutex::new(HashMap::new())); + let child = Arc::new(Mutex::new(Some(child))); + + let write_handle = spawn_writer_thread(stdin, writer_rx, pending.clone(), event_tx.clone()); + let read_handle = spawn_reader_thread(stdout, pending.clone(), event_tx.clone()); + + let mut client = Self { + config, + writer: Some(writer_tx), + events: Some(event_rx), + pending, + next_id: AtomicI64::new(1), + read_handle: Some(read_handle), + write_handle: Some(write_handle), + child, + protocol_version: 0, + server_capabilities: None, + }; + + let initialize = client.initialize()?; + client.validate_initialize(&initialize)?; + client.protocol_version = initialize.protocol_version; + client.server_capabilities = Some(initialize.capabilities.clone()); + + Ok(client) + } + + pub fn list_clusters(&self) -> Result { + self.send_request("listClusters", ListClustersParams {}) + } + + pub fn list_cluster_metrics( + &self, + params: ListClusterMetricsParams, + ) -> Result { + self.send_request("listClusterMetrics", params) + } + + pub fn get_cluster_summary( + &self, + params: GetClusterSummaryParams, + ) -> Result { + self.send_request("getClusterSummary", params) + } + + pub fn get_cluster_topology( + &self, + params: GetClusterTopologyParams, + ) -> Result { + self.send_request("getClusterTopology", params) + } + + pub fn subscribe_cluster_logs( + &self, + params: SubscribeClusterLogsParams, + ) -> Result { + self.send_request("subscribeClusterLogs", params) + } + + pub fn subscribe_cluster_timeline( + &self, + params: SubscribeClusterTimelineParams, + ) -> Result { + self.send_request("subscribeClusterTimeline", params) + } + + pub fn unsubscribe( + &self, + params: UnsubscribeParams, + ) -> Result { + self.send_request("unsubscribe", params) + } + + pub fn start_cluster_from_text( + &self, + params: StartClusterFromTextParams, + ) -> Result { + self.send_request("startClusterFromText", params) + } + + pub fn start_cluster_from_issue( + &self, + params: StartClusterFromIssueParams, + ) -> Result { + self.send_request("startClusterFromIssue", params) + } + + pub fn send_guidance_to_agent( + &self, + params: SendGuidanceToAgentParams, + ) -> Result { + self.send_request("sendGuidanceToAgent", params) + } + + pub fn send_guidance_to_cluster( + &self, + params: SendGuidanceToClusterParams, + ) -> Result { + self.send_request("sendGuidanceToCluster", params) + } + + pub fn client_info(&self) -> &ClientInfo { + &self.config.client + } + + pub fn client_capabilities(&self) -> Option<&ClientCapabilities> { + self.config.capabilities.as_ref() + } + + fn initialize(&self) -> Result { + let params = InitializeParams { + protocol_version: self.config.protocol_version, + client: self.config.client.clone(), + capabilities: self.config.capabilities.clone(), + }; + self.send_request("initialize", params) + } + + fn validate_initialize(&self, initialize: &InitializeResult) -> Result<(), BackendError> { + if initialize.protocol_version != self.config.protocol_version { + return Err(BackendError::Protocol(format!( + "Protocol version mismatch: expected {}, got {}", + self.config.protocol_version, initialize.protocol_version + ))); + } + Ok(()) + } + + fn send_request( + &self, + method: &str, + params: P, + ) -> Result { + let (key, frame, rx) = self.prepare_request(method, params)?; + self.send_frame(&key, frame)?; + let response = self.await_response(method, &key, rx)?; + Ok(serde_json::from_value(response)?) + } + + fn prepare_request( + &self, + method: &str, + params: P, + ) -> Result< + ( + RequestKey, + Vec, + mpsc::Receiver>, + ), + BackendError, + > { + let id = self.next_id.fetch_add(1, Ordering::SeqCst); + let request = JsonRpcRequest { + jsonrpc: JSONRPC_VERSION.to_string(), + id: JsonRpcId::Number(id), + method: method.to_string(), + params: Some(params), + }; + let payload = serde_json::to_vec(&request)?; + let frame = FrameEncoder::encode(&payload)?; + let (tx, rx) = mpsc::channel(); + let key = RequestKey::Number(id); + let mut pending = self + .pending + .lock() + .map_err(|_| BackendError::Protocol("Pending request lock poisoned".into()))?; + pending.insert(key.clone(), tx); + Ok((key, frame, rx)) + } + + fn send_frame(&self, key: &RequestKey, frame: Vec) -> Result<(), BackendError> { + let writer = self + .writer + .as_ref() + .ok_or_else(|| BackendError::Disconnected("Backend writer closed".into()))?; + if writer.send(WriterCommand::Frame(frame)).is_err() { + self.remove_pending(key)?; + return Err(BackendError::Disconnected( + "Backend writer channel closed".into(), + )); + } + Ok(()) + } + + fn await_response( + &self, + method: &str, + key: &RequestKey, + rx: mpsc::Receiver>, + ) -> Result { + let response = match self.config.request_timeout { + Some(timeout) => match rx.recv_timeout(timeout) { + Ok(value) => value, + Err(err) => { + self.remove_pending(key)?; + return Err(BackendError::Timeout(format!("{method} failed: {err}"))); + } + }, + None => match rx.recv() { + Ok(value) => value, + Err(_) => { + self.remove_pending(key)?; + return Err(BackendError::Disconnected("Backend disconnected".into())); + } + }, + }?; + Ok(response) + } + + fn remove_pending(&self, key: &RequestKey) -> Result<(), BackendError> { + let mut pending = self + .pending + .lock() + .map_err(|_| BackendError::Protocol("Pending request lock poisoned".into()))?; + pending.remove(key); + Ok(()) + } +} + +impl BackendClient for StdioBackendClient { + fn take_event_receiver(&mut self) -> Option> { + self.events.take() + } + + fn server_capabilities(&self) -> Option { + self.server_capabilities.clone() + } + + fn protocol_version(&self) -> i64 { + self.protocol_version + } + + fn shutdown(&mut self) -> Result<(), BackendError> { + if let Some(writer) = self.writer.take() { + let _ = writer.send(WriterCommand::Shutdown); + } + + if let Some(mut child) = self + .child + .lock() + .map_err(|_| BackendError::Protocol("Child lock poisoned".into()))? + .take() + { + let _ = child.kill(); + let _ = child.wait(); + } + + if let Some(handle) = self.write_handle.take() { + let _ = handle.join(); + } + if let Some(handle) = self.read_handle.take() { + let _ = handle.join(); + } + Ok(()) + } +} + +impl Drop for StdioBackendClient { + fn drop(&mut self) { + let _ = self.shutdown(); + } +} + +fn spawn_writer_thread( + mut stdin: impl Write + Send + 'static, + receiver: mpsc::Receiver, + pending: Arc>>>>, + event_tx: mpsc::Sender, +) -> thread::JoinHandle<()> { + thread::spawn(move || writer_loop(&mut stdin, receiver, pending, event_tx)) +} + +fn writer_loop( + stdin: &mut (impl Write + Send + 'static), + receiver: mpsc::Receiver, + pending: Arc>>>>, + event_tx: mpsc::Sender, +) { + for command in receiver { + match command { + WriterCommand::Frame(frame) => { + if let Err(err) = stdin.write_all(&frame) { + let error = BackendError::Io(err); + let message = error.to_string(); + drain_pending(&pending, error); + let _ = event_tx.send(BackendEvent::BackendExited(BackendExit { + code: None, + message, + })); + break; + } + let _ = stdin.flush(); + } + WriterCommand::Shutdown => break, + } + } +} + +fn spawn_reader_thread( + mut stdout: impl Read + Send + 'static, + pending: Arc>>>>, + event_tx: mpsc::Sender, +) -> thread::JoinHandle<()> { + thread::spawn(move || reader_loop(&mut stdout, pending, event_tx)) +} + +fn reader_loop( + stdout: &mut (impl Read + Send + 'static), + pending: Arc>>>>, + event_tx: mpsc::Sender, +) { + let mut decoder = FrameDecoder::new(); + let mut buffer = [0u8; 8192]; + loop { + match stdout.read(&mut buffer) { + Ok(0) => { + handle_reader_disconnect( + &pending, + &event_tx, + BackendError::Disconnected("Backend closed stdout".into()), + "Backend closed stdout", + ); + break; + } + Ok(bytes) => { + if let Err(err) = + handle_reader_bytes(&buffer[..bytes], &mut decoder, &pending, &event_tx) + { + let message = err.to_string(); + drain_pending(&pending, err); + let _ = event_tx.send(BackendEvent::BackendExited(BackendExit { + code: None, + message, + })); + return; + } + } + Err(err) => { + handle_reader_disconnect( + &pending, + &event_tx, + BackendError::Io(err), + "Backend stdout read failed", + ); + break; + } + } + } +} + +fn handle_reader_bytes( + bytes: &[u8], + decoder: &mut FrameDecoder, + pending: &Arc>>>>, + event_tx: &mpsc::Sender, +) -> Result<(), BackendError> { + let frames = decoder.push(bytes)?; + for frame in frames { + handle_frame(&frame, pending, event_tx)?; + } + Ok(()) +} + +fn handle_reader_disconnect( + pending: &Arc>>>>, + event_tx: &mpsc::Sender, + error: BackendError, + message: &str, +) { + let event_message = if matches!(error, BackendError::Disconnected(_)) { + message.to_string() + } else { + error.to_string() + }; + drain_pending(pending, error); + let _ = event_tx.send(BackendEvent::BackendExited(BackendExit { + code: None, + message: event_message, + })); +} + +fn handle_frame( + frame: &[u8], + pending: &Arc>>>>, + event_tx: &mpsc::Sender, +) -> Result<(), BackendError> { + let value = parse_frame_json(frame)?; + ensure_jsonrpc_version(&value)?; + if is_notification(&value) { + return handle_notification(value, event_tx); + } + let key = parse_response_id(&value)?; + dispatch_response(value, key, pending); + Ok(()) +} + +fn parse_frame_json(frame: &[u8]) -> Result { + let value: Value = serde_json::from_slice(frame)?; + if !value.is_object() { + return Err(BackendError::Protocol("Non-object JSON-RPC message".into())); + } + Ok(value) +} + +fn ensure_jsonrpc_version(value: &Value) -> Result<(), BackendError> { + let jsonrpc = value + .get("jsonrpc") + .and_then(|value| value.as_str()) + .ok_or_else(|| BackendError::Protocol("Missing jsonrpc version".into()))?; + if jsonrpc != JSONRPC_VERSION { + return Err(BackendError::Protocol(format!( + "Unsupported jsonrpc version: {jsonrpc}" + ))); + } + Ok(()) +} + +fn is_notification(value: &Value) -> bool { + value.get("id").is_none() +} + +fn parse_response_id(value: &Value) -> Result { + let id_value = value + .get("id") + .ok_or_else(|| BackendError::Protocol("Missing id".into()))?; + RequestKey::from_value(id_value).ok_or_else(|| BackendError::Protocol("Invalid id type".into())) +} + +fn dispatch_response( + value: Value, + key: RequestKey, + pending: &Arc>>>>, +) { + let sender = { + let mut pending = match pending.lock() { + Ok(lock) => lock, + Err(poisoned) => poisoned.into_inner(), + }; + pending.remove(&key) + }; + + if let Some(sender) = sender { + if let Some(error_value) = value.get("error") { + let error = parse_rpc_error(error_value); + let _ = sender.send(Err(BackendError::Rpc(error))); + return; + } + if let Some(result) = value.get("result") { + let _ = sender.send(Ok(result.clone())); + return; + } + let _ = sender.send(Err(BackendError::Protocol( + "Response missing result or error".into(), + ))); + } +} + +fn parse_rpc_error(error_value: &Value) -> crate::protocol::RpcError { + serde_json::from_value(error_value.clone()).unwrap_or_else(|err| crate::protocol::RpcError { + code: -32603, + message: format!("Failed to parse RPC error: {err}"), + data: None, + }) +} + +fn handle_notification( + value: Value, + event_tx: &mpsc::Sender, +) -> Result<(), BackendError> { + let method = value + .get("method") + .and_then(|value| value.as_str()) + .ok_or_else(|| BackendError::Protocol("Notification missing method".into()))?; + let params = value.get("params").cloned(); + + let notification = match method { + "clusterLogLines" => { + let params_value = params + .ok_or_else(|| BackendError::Protocol("clusterLogLines missing params".into()))?; + let parsed: ClusterLogLinesParams = serde_json::from_value(params_value)?; + BackendNotification::ClusterLogLines(parsed) + } + "clusterTimelineEvents" => { + let params_value = params.ok_or_else(|| { + BackendError::Protocol("clusterTimelineEvents missing params".into()) + })?; + let parsed: ClusterTimelineEventsParams = serde_json::from_value(params_value)?; + BackendNotification::ClusterTimelineEvents(parsed) + } + _ => BackendNotification::Unknown { + method: method.to_string(), + params, + }, + }; + + let _ = event_tx.send(BackendEvent::Notification(notification)); + Ok(()) +} + +fn drain_pending( + pending: &Arc>>>>, + error: BackendError, +) { + let message = error.to_string(); + let senders = { + let mut pending = match pending.lock() { + Ok(lock) => lock, + Err(poisoned) => poisoned.into_inner(), + }; + let mut items = Vec::with_capacity(pending.len()); + for (_, sender) in pending.drain() { + items.push(sender); + } + items + }; + + for sender in senders { + let _ = sender.send(Err(BackendError::Disconnected(message.clone()))); + } +} + +pub fn resolve_backend_path(config: &BackendConfig) -> Result { + if let Some(path) = config.backend_path.clone() { + return Ok(path); + } + + if let Ok(value) = env::var(BACKEND_PATH_ENV) { + let trimmed = value.trim(); + if !trimmed.is_empty() { + return Ok(PathBuf::from(trimmed)); + } + } + + let cwd = env::current_dir()?; + if let Some(found) = find_in_ancestors(&cwd, DEFAULT_BACKEND_RELATIVE_PATH) { + return Ok(found); + } + + Err(BackendError::Protocol(format!( + "Backend path not found (set {BACKEND_PATH_ENV} or build {DEFAULT_BACKEND_RELATIVE_PATH})" + ))) +} + +fn find_in_ancestors(start: &Path, relative: &str) -> Option { + for ancestor in start.ancestors() { + let candidate = ancestor.join(relative); + if candidate.is_file() { + return Some(candidate); + } + } + None +} diff --git a/tui-rs/crates/zeroshot-tui/src/commands/dispatcher.rs b/tui-rs/crates/zeroshot-tui/src/commands/dispatcher.rs new file mode 100644 index 00000000..4bcc08e0 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/commands/dispatcher.rs @@ -0,0 +1,272 @@ +use crate::app::{ + Action, CommandAction, CommandContext, NavigationAction, ScreenId, ToastLevel, UiVariant, +}; +use crate::commands::types::{ParsedCommand, VALID_PROVIDERS}; + +pub fn dispatch(parsed: ParsedCommand, context: CommandContext) -> Vec { + match parsed.name.as_str() { + "help" => vec![Action::Command(CommandAction::ShowToast { + level: ToastLevel::Info, + message: help_message(), + })], + "monitor" => handle_monitor(context), + "guide" => handle_guidance(parsed, None, true), + "nudge" => handle_guidance(parsed, Some("[nudge]"), true), + "interrupt" => handle_guidance(parsed, Some("[interrupt]"), false), + "pin" => vec![Action::Command(CommandAction::TogglePin)], + "issue" => handle_issue(parsed, context), + "provider" => handle_provider(parsed), + "quit" | "exit" => vec![Action::Quit], + other => vec![Action::Command(CommandAction::ShowToast { + level: ToastLevel::Error, + message: format!("Unknown command: {other}. Try /help."), + })], + } +} + +fn handle_monitor(context: CommandContext) -> Vec { + let (target, label) = if matches!(context.ui_variant, UiVariant::Disruptive) { + (ScreenId::FleetRadar, "Fleet Radar") + } else { + (ScreenId::Monitor, "Monitor") + }; + + if context.active_screen == target { + return vec![Action::Command(CommandAction::ShowToast { + level: ToastLevel::Info, + message: format!("Already on {label}."), + })]; + } + + vec![ + Action::Navigate(NavigationAction::Push(target)), + Action::Command(CommandAction::ShowToast { + level: ToastLevel::Success, + message: format!("Opened {label}."), + }), + ] +} + +fn handle_issue(parsed: ParsedCommand, context: CommandContext) -> Vec { + let Some(reference) = parsed.args.first() else { + return vec![Action::Command(CommandAction::ShowToast { + level: ToastLevel::Error, + message: "Usage: /issue ".to_string(), + })]; + }; + + vec![ + Action::Command(CommandAction::StartClusterFromIssue { + reference: reference.to_string(), + provider_override: context.provider_override, + }), + Action::Command(CommandAction::ShowToast { + level: ToastLevel::Info, + message: format!("Starting cluster from issue {reference}..."), + }), + ] +} + +fn handle_guidance( + parsed: ParsedCommand, + prefix: Option<&'static str>, + require_text: bool, +) -> Vec { + let message = parsed.args.join(" "); + if require_text && message.trim().is_empty() { + return vec![Action::Command(CommandAction::ShowToast { + level: ToastLevel::Error, + message: format!("Usage: /{} ", parsed.name), + })]; + } + vec![Action::Command(CommandAction::SendGuidance { + message, + prefix: prefix.map(|value| value.to_string()), + })] +} + +fn handle_provider(parsed: ParsedCommand) -> Vec { + let Some(name) = parsed.args.first() else { + return vec![Action::Command(CommandAction::ShowToast { + level: ToastLevel::Error, + message: format!("Usage: /provider <{}>", VALID_PROVIDERS.join("|")), + })]; + }; + + let normalized = name.to_lowercase(); + if !VALID_PROVIDERS.contains(&normalized.as_str()) { + return vec![Action::Command(CommandAction::ShowToast { + level: ToastLevel::Error, + message: format!( + "Unknown provider '{name}'. Use one of: {}", + VALID_PROVIDERS.join(", ") + ), + })]; + } + + vec![ + Action::Command(CommandAction::SetProviderOverride { + provider: Some(normalized.clone()), + }), + Action::Command(CommandAction::ShowToast { + level: ToastLevel::Success, + message: format!("Provider override set to {normalized}."), + }), + ] +} + +fn help_message() -> String { + let lines = [ + "Commands: /help /monitor /issue /provider /guide /nudge /interrupt [text] /pin /quit /exit", + "Keys: / command bar, ? help, Esc back, q quit (not in Launcher), Ctrl+C quit, j/k or arrows move, PgUp/PgDn fast, Tab/Shift+Tab or h/l switch panes", + ]; + lines.join("\n") +} + +#[cfg(test)] +mod tests { + use super::dispatch; + use crate::app::{Action, CommandContext, NavigationAction, ScreenId, ToastLevel, UiVariant}; + use crate::commands::types::ParsedCommand; + + fn context() -> CommandContext { + CommandContext { + provider_override: None, + active_screen: ScreenId::Launcher, + ui_variant: UiVariant::Classic, + } + } + + #[test] + fn unknown_command_returns_error_toast() { + let parsed = ParsedCommand { + raw: "/nope".to_string(), + name: "nope".to_string(), + args: vec![], + }; + let actions = dispatch(parsed, context()); + let toast = actions + .iter() + .find_map(|action| match action { + crate::app::Action::Command(crate::app::CommandAction::ShowToast { + level, + message, + }) => Some((level, message)), + _ => None, + }) + .expect("expected toast action"); + assert_eq!(toast.0, &ToastLevel::Error); + assert!(toast.1.contains("Unknown command")); + } + + #[test] + fn invalid_provider_is_rejected() { + let parsed = ParsedCommand { + raw: "/provider nope".to_string(), + name: "provider".to_string(), + args: vec!["nope".to_string()], + }; + let actions = dispatch(parsed, context()); + let toast = actions + .iter() + .find_map(|action| match action { + crate::app::Action::Command(crate::app::CommandAction::ShowToast { + level, + message, + }) => Some((level, message)), + _ => None, + }) + .expect("expected toast action"); + assert_eq!(toast.0, &ToastLevel::Error); + assert!(toast.1.contains("Unknown provider")); + } + + #[test] + fn monitor_command_targets_fleet_radar_in_disruptive() { + let parsed = ParsedCommand { + raw: "/monitor".to_string(), + name: "monitor".to_string(), + args: vec![], + }; + let mut context = context(); + context.ui_variant = UiVariant::Disruptive; + let actions = dispatch(parsed, context); + assert!(actions.iter().any(|action| matches!( + action, + Action::Navigate(NavigationAction::Push(ScreenId::FleetRadar)) + ))); + } + + #[test] + fn monitor_command_targets_monitor_in_classic() { + let parsed = ParsedCommand { + raw: "/monitor".to_string(), + name: "monitor".to_string(), + args: vec![], + }; + let actions = dispatch(parsed, context()); + assert!(actions.iter().any(|action| matches!( + action, + Action::Navigate(NavigationAction::Push(ScreenId::Monitor)) + ))); + } + + #[test] + fn guide_command_dispatches_guidance_without_prefix() { + let parsed = ParsedCommand { + raw: "/guide hi there".to_string(), + name: "guide".to_string(), + args: vec!["hi".to_string(), "there".to_string()], + }; + let actions = dispatch(parsed, context()); + assert!(actions.iter().any(|action| matches!( + action, + Action::Command(crate::app::CommandAction::SendGuidance { message, prefix }) + if message == "hi there" && prefix.is_none() + ))); + } + + #[test] + fn nudge_command_dispatches_guidance_with_prefix() { + let parsed = ParsedCommand { + raw: "/nudge hi".to_string(), + name: "nudge".to_string(), + args: vec!["hi".to_string()], + }; + let actions = dispatch(parsed, context()); + assert!(actions.iter().any(|action| matches!( + action, + Action::Command(crate::app::CommandAction::SendGuidance { message, prefix }) + if message == "hi" && prefix.as_deref() == Some("[nudge]") + ))); + } + + #[test] + fn interrupt_command_allows_empty_text() { + let parsed = ParsedCommand { + raw: "/interrupt".to_string(), + name: "interrupt".to_string(), + args: vec![], + }; + let actions = dispatch(parsed, context()); + assert!(actions.iter().any(|action| matches!( + action, + Action::Command(crate::app::CommandAction::SendGuidance { message, prefix }) + if message.is_empty() && prefix.as_deref() == Some("[interrupt]") + ))); + } + + #[test] + fn pin_command_dispatches_toggle() { + let parsed = ParsedCommand { + raw: "/pin".to_string(), + name: "pin".to_string(), + args: vec![], + }; + let actions = dispatch(parsed, context()); + assert!(actions.iter().any(|action| matches!( + action, + Action::Command(crate::app::CommandAction::TogglePin) + ))); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/commands/mod.rs b/tui-rs/crates/zeroshot-tui/src/commands/mod.rs new file mode 100644 index 00000000..eca3748f --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/commands/mod.rs @@ -0,0 +1,20 @@ +use crate::app::{Action, CommandAction, CommandRequest, ToastLevel}; + +mod dispatcher; +mod parser; +mod types; + +pub use parser::parse; +pub use types::{CommandError, ParsedCommand, VALID_PROVIDERS}; + +pub fn dispatch(request: CommandRequest) -> Result, CommandError> { + match request { + CommandRequest::SubmitRaw { raw, context } => match parser::parse(&raw) { + Ok(parsed) => Ok(dispatcher::dispatch(parsed, context)), + Err(err) => Ok(vec![Action::Command(CommandAction::ShowToast { + level: ToastLevel::Error, + message: err.to_string(), + })]), + }, + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/commands/parser.rs b/tui-rs/crates/zeroshot-tui/src/commands/parser.rs new file mode 100644 index 00000000..7a6b8d01 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/commands/parser.rs @@ -0,0 +1,66 @@ +use crate::commands::types::{CommandError, ParsedCommand}; + +pub fn parse(raw: &str) -> Result { + let trimmed = raw.trim(); + if trimmed.is_empty() { + return Err(CommandError::new("Enter a command.")); + } + if !trimmed.starts_with('/') { + return Err(CommandError::new("Commands must start with '/'.")); + } + + let body = trimmed.trim_start_matches('/').trim(); + if body.is_empty() { + return Err(CommandError::new("Enter a command after '/'.")); + } + + let mut parts = body.split_whitespace(); + let Some(name) = parts.next() else { + return Err(CommandError::new("Enter a command after '/'.")); + }; + + let args = parts.map(|part| part.to_string()).collect::>(); + Ok(ParsedCommand { + raw: trimmed.to_string(), + name: name.to_lowercase(), + args, + }) +} + +#[cfg(test)] +mod tests { + use super::parse; + + #[test] + fn parse_empty() { + let err = parse("").expect_err("expected error"); + assert_eq!(err.to_string(), "Enter a command."); + } + + #[test] + fn parse_slash_only() { + let err = parse("/").expect_err("expected error"); + assert_eq!(err.to_string(), "Enter a command after '/'."); + } + + #[test] + fn parse_whitespace_tolerant() { + let parsed = parse(" /provider codex ").expect("expected command"); + assert_eq!(parsed.name, "provider"); + assert_eq!(parsed.args, vec!["codex".to_string()]); + } + + #[test] + fn parse_provider_command() { + let parsed = parse("/provider codex").expect("expected command"); + assert_eq!(parsed.name, "provider"); + assert_eq!(parsed.args, vec!["codex".to_string()]); + } + + #[test] + fn parse_issue_command() { + let parsed = parse("/issue org/repo#123").expect("expected command"); + assert_eq!(parsed.name, "issue"); + assert_eq!(parsed.args, vec!["org/repo#123".to_string()]); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/commands/types.rs b/tui-rs/crates/zeroshot-tui/src/commands/types.rs new file mode 100644 index 00000000..cdfcc726 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/commands/types.rs @@ -0,0 +1,41 @@ +use std::fmt; + +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct ParsedCommand { + pub raw: String, + pub name: String, + pub args: Vec, +} + +impl ParsedCommand { + pub fn name(&self) -> &str { + self.name.as_str() + } + + pub fn args(&self) -> &[String] { + self.args.as_slice() + } +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct CommandError { + message: String, +} + +impl CommandError { + pub fn new(message: impl Into) -> Self { + Self { + message: message.into(), + } + } +} + +impl fmt::Display for CommandError { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + write!(f, "{}", self.message) + } +} + +impl std::error::Error for CommandError {} + +pub const VALID_PROVIDERS: [&str; 4] = ["claude", "codex", "gemini", "opencode"]; diff --git a/tui-rs/crates/zeroshot-tui/src/input.rs b/tui-rs/crates/zeroshot-tui/src/input.rs new file mode 100644 index 00000000..6013fb07 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/input.rs @@ -0,0 +1,489 @@ +use crossterm::event::{KeyCode, KeyEvent, KeyModifiers}; + +use crate::app::{ + Action, AppState, CommandBarAction, NavigationAction, ScreenAction, ScreenId, SpineAction, + SpineMode, TimeCursorAction, UiVariant, ZoomStackContext, TIME_SCRUB_STEP_LARGE_MS, + TIME_SCRUB_STEP_MS, +}; +use crate::screens::{agent, cluster, cluster_canvas, launcher, monitor, radar}; + +pub fn route_key(state: &AppState, key: KeyEvent) -> Option { + if matches!(state.ui_variant, UiVariant::Disruptive) { + return route_disruptive(state, key); + } + + if state.command_bar.active { + return route_command_bar(key); + } + + let screen = state.active_screen(); + if let Some(action) = route_global(screen, key) { + return Some(action); + } + + if !matches!(screen, ScreenId::Launcher | ScreenId::IntentConsole) { + match key.code { + KeyCode::Char('/') + if !key.modifiers.contains(KeyModifiers::CONTROL) + && !key.modifiers.contains(KeyModifiers::ALT) => + { + return Some(Action::CommandBar(CommandBarAction::Open { + prefill: "/".to_string(), + })); + } + KeyCode::Char('?') + if !key.modifiers.contains(KeyModifiers::CONTROL) + && !key.modifiers.contains(KeyModifiers::ALT) => + { + return Some(Action::CommandBar(CommandBarAction::Open { + prefill: "/help ".to_string(), + })); + } + _ => {} + } + } + + match screen { + ScreenId::Launcher => route_launcher(key), + ScreenId::Monitor => route_monitor(key), + ScreenId::Cluster { id } => route_cluster(id, key), + ScreenId::Agent { + cluster_id, + agent_id, + } => route_agent(cluster_id, agent_id, key), + ScreenId::IntentConsole + | ScreenId::FleetRadar + | ScreenId::ClusterCanvas { .. } + | ScreenId::AgentMicroscope { .. } => None, + } +} + +fn route_disruptive(state: &AppState, key: KeyEvent) -> Option { + let ctrl = key.modifiers.contains(KeyModifiers::CONTROL); + let alt = key.modifiers.contains(KeyModifiers::ALT); + + if ctrl && matches!(key.code, KeyCode::Char('c')) { + return Some(Action::Quit); + } + + if !spine_active(state) { + if let Some(action) = route_time_scrub(state, key, ctrl, alt) { + return Some(action); + } + if let Some(action) = route_disruptive_radar(state, key, ctrl, alt) { + return Some(action); + } + if let Some(action) = route_disruptive_cluster_canvas(state, key, ctrl, alt) { + return Some(action); + } + } + + match key.code { + KeyCode::Esc => { + if spine_active(state) { + Some(Action::Spine(SpineAction::Cancel)) + } else { + Some(Action::Navigate(NavigationAction::Pop)) + } + } + KeyCode::Enter => { + if spine_active(state) { + Some(Action::Spine(SpineAction::Submit)) + } else if let Some(action) = cluster_canvas_zoom_action(state) { + Some(action) + } else { + zoom_in_action(state) + } + } + KeyCode::Char('?') if !ctrl && !alt => Some(Action::Spine(SpineAction::EnterMode { + mode: SpineMode::Command, + prefill: "help ".to_string(), + })), + KeyCode::Char('/') if !ctrl && !alt => Some(Action::Spine(SpineAction::EnterMode { + mode: SpineMode::Command, + prefill: String::new(), + })), + KeyCode::Char('i') if !ctrl && !alt => Some(Action::Spine(SpineAction::EnterMode { + mode: intent_mode_for_context(state), + prefill: String::new(), + })), + KeyCode::Char('u') if ctrl => Some(Action::Spine(SpineAction::Clear)), + KeyCode::Tab => match state.spine.completion.as_ref() { + Some(completion) if completion.candidates.len() > 1 => { + Some(Action::Spine(SpineAction::CycleCompletion)) + } + Some(completion) if !completion.ghost.is_empty() => { + Some(Action::Spine(SpineAction::AcceptCompletion)) + } + _ => None, + }, + KeyCode::Backspace => Some(Action::Spine(SpineAction::Backspace)), + KeyCode::Delete => Some(Action::Spine(SpineAction::Delete)), + KeyCode::Left => Some(Action::Spine(SpineAction::MoveCursorLeft)), + KeyCode::Right => Some(Action::Spine(SpineAction::MoveCursorRight)), + KeyCode::Home => Some(Action::Spine(SpineAction::MoveCursorHome)), + KeyCode::End => Some(Action::Spine(SpineAction::MoveCursorEnd)), + KeyCode::Char(ch) if !ctrl && !alt => Some(Action::Spine(SpineAction::InsertChar(ch))), + _ => None, + } +} + +fn route_time_scrub(state: &AppState, key: KeyEvent, ctrl: bool, alt: bool) -> Option { + if ctrl || alt { + return None; + } + let scope_available = state.temporal_focus_scope().is_some(); + let focus_active = state.temporal_focus.is_active(); + + let large = key.modifiers.contains(KeyModifiers::SHIFT); + let step = if large { + TIME_SCRUB_STEP_LARGE_MS + } else { + TIME_SCRUB_STEP_MS + }; + + match key.code { + KeyCode::Left if focus_active => Some(Action::TimeCursor(TimeCursorAction::Step { + delta_ms: -step, + })), + KeyCode::Right if focus_active => Some(Action::TimeCursor(TimeCursorAction::Step { + delta_ms: step, + })), + KeyCode::End if focus_active => Some(Action::TimeCursor(TimeCursorAction::JumpToLive)), + KeyCode::Char(' ') if focus_active || scope_available => { + Some(Action::TimeCursor(TimeCursorAction::ToggleFollow)) + } + _ => None, + } +} + +fn route_disruptive_radar( + state: &AppState, + key: KeyEvent, + ctrl: bool, + alt: bool, +) -> Option { + if ctrl || alt { + return None; + } + if !matches!(state.zoom_stack_context(), ZoomStackContext::FleetRadar) { + return None; + } + + let speed = if key.modifiers.contains(KeyModifiers::SHIFT) { + radar::MoveSpeed::Fast + } else { + radar::MoveSpeed::Step + }; + + let action = match key.code { + KeyCode::Left | KeyCode::Char('h') | KeyCode::Char('H') => radar::Action::MoveSelection { + direction: radar::Direction::Left, + speed, + }, + KeyCode::Right | KeyCode::Char('l') | KeyCode::Char('L') => radar::Action::MoveSelection { + direction: radar::Direction::Right, + speed, + }, + KeyCode::Up | KeyCode::Char('k') | KeyCode::Char('K') => radar::Action::MoveSelection { + direction: radar::Direction::Up, + speed, + }, + KeyCode::Down | KeyCode::Char('j') | KeyCode::Char('J') => radar::Action::MoveSelection { + direction: radar::Direction::Down, + speed, + }, + KeyCode::Char('g') => radar::Action::CenterOnSelection, + KeyCode::Char('G') => radar::Action::ResetView, + _ => return None, + }; + + Some(Action::Screen(ScreenAction::FleetRadar(action))) +} + +fn route_disruptive_cluster_canvas( + state: &AppState, + key: KeyEvent, + ctrl: bool, + alt: bool, +) -> Option { + if ctrl || alt { + return None; + } + + let ScreenId::ClusterCanvas { id } = state.active_screen() else { + return None; + }; + + let speed = if key.modifiers.contains(KeyModifiers::SHIFT) { + cluster_canvas::MoveSpeed::Fast + } else { + cluster_canvas::MoveSpeed::Step + }; + + let action = match key.code { + KeyCode::Left | KeyCode::Char('h') | KeyCode::Char('H') => { + cluster_canvas::Action::MoveFocus { + direction: cluster_canvas::Direction::Left, + speed, + } + } + KeyCode::Right | KeyCode::Char('l') | KeyCode::Char('L') => { + cluster_canvas::Action::MoveFocus { + direction: cluster_canvas::Direction::Right, + speed, + } + } + KeyCode::Up | KeyCode::Char('k') | KeyCode::Char('K') => { + cluster_canvas::Action::MoveFocus { + direction: cluster_canvas::Direction::Up, + speed, + } + } + KeyCode::Down | KeyCode::Char('j') | KeyCode::Char('J') => { + cluster_canvas::Action::MoveFocus { + direction: cluster_canvas::Direction::Down, + speed, + } + } + _ => return None, + }; + + Some(Action::Screen(ScreenAction::ClusterCanvas { + id: id.clone(), + action, + })) +} + +fn spine_active(state: &AppState) -> bool { + if matches!(state.active_screen(), ScreenId::AgentMicroscope { .. }) { + return !state.spine.input.input.is_empty() || state.spine.completion.is_some(); + } + !state.spine.input.input.is_empty() + || !matches!(state.spine.mode, SpineMode::Intent) + || state.spine.completion.is_some() +} + +fn intent_mode_for_context(state: &AppState) -> SpineMode { + match state.zoom_stack_context() { + ZoomStackContext::Agent { .. } => SpineMode::WhisperAgent, + ZoomStackContext::Cluster { .. } => SpineMode::WhisperCluster, + ZoomStackContext::FleetRadar => { + if selected_cluster_id_for_zoom(state).is_some() { + SpineMode::WhisperCluster + } else { + SpineMode::Intent + } + } + ZoomStackContext::Root => SpineMode::Intent, + } +} + +fn zoom_in_action(state: &AppState) -> Option { + match state.zoom_stack_context() { + ZoomStackContext::FleetRadar => selected_cluster_id_for_zoom(state).map(|cluster_id| { + Action::Navigate(NavigationAction::Push(ScreenId::ClusterCanvas { + id: cluster_id, + })) + }), + ZoomStackContext::Cluster { id } => selected_agent_id(state, &id).map(|agent_id| { + Action::Navigate(NavigationAction::Push(ScreenId::AgentMicroscope { + cluster_id: id, + agent_id, + })) + }), + ZoomStackContext::Agent { .. } | ZoomStackContext::Root => None, + } +} + +fn cluster_canvas_zoom_action(state: &AppState) -> Option { + match state.active_screen() { + ScreenId::ClusterCanvas { id } => Some(Action::Screen(ScreenAction::ClusterCanvas { + id: id.clone(), + action: cluster_canvas::Action::ZoomIn, + })), + _ => None, + } +} + +fn selected_cluster_id_for_zoom(state: &AppState) -> Option { + match state.active_screen() { + ScreenId::Monitor => state.monitor.selected_cluster_id(), + _ => state.fleet_radar.selected_cluster_id(), + } +} + +fn selected_agent_id(state: &AppState, cluster_id: &str) -> Option { + let cluster_state = state.clusters.get(cluster_id)?; + let agent = cluster_state.agents.get(cluster_state.selected_agent)?; + Some(agent.id.clone()) +} + +fn route_command_bar(key: KeyEvent) -> Option { + match key.code { + KeyCode::Esc => Some(Action::CommandBar(CommandBarAction::Close)), + KeyCode::Enter => Some(Action::CommandBar(CommandBarAction::Submit)), + KeyCode::Backspace => Some(Action::CommandBar(CommandBarAction::Backspace)), + KeyCode::Delete => Some(Action::CommandBar(CommandBarAction::Delete)), + KeyCode::Left => Some(Action::CommandBar(CommandBarAction::MoveCursorLeft)), + KeyCode::Right => Some(Action::CommandBar(CommandBarAction::MoveCursorRight)), + KeyCode::Home => Some(Action::CommandBar(CommandBarAction::MoveCursorHome)), + KeyCode::End => Some(Action::CommandBar(CommandBarAction::MoveCursorEnd)), + KeyCode::Char(ch) + if !key.modifiers.contains(KeyModifiers::CONTROL) + && !key.modifiers.contains(KeyModifiers::ALT) => + { + Some(Action::CommandBar(CommandBarAction::InsertChar(ch))) + } + _ => None, + } +} + +fn route_global(screen: &ScreenId, key: KeyEvent) -> Option { + match key.code { + KeyCode::Esc => Some(Action::Navigate(NavigationAction::Pop)), + KeyCode::Char('c') if key.modifiers.contains(KeyModifiers::CONTROL) => Some(Action::Quit), + KeyCode::Char('q') => match screen { + ScreenId::Launcher | ScreenId::IntentConsole => None, + _ => Some(Action::Quit), + }, + _ => None, + } +} + +fn route_launcher(key: KeyEvent) -> Option { + match key.code { + KeyCode::Enter => Some(Action::Screen(ScreenAction::Launcher( + launcher::Action::Submit, + ))), + KeyCode::Backspace => Some(Action::Screen(ScreenAction::Launcher( + launcher::Action::Backspace, + ))), + KeyCode::Delete => Some(Action::Screen(ScreenAction::Launcher( + launcher::Action::Delete, + ))), + KeyCode::Left => Some(Action::Screen(ScreenAction::Launcher( + launcher::Action::MoveCursorLeft, + ))), + KeyCode::Right => Some(Action::Screen(ScreenAction::Launcher( + launcher::Action::MoveCursorRight, + ))), + KeyCode::Home => Some(Action::Screen(ScreenAction::Launcher( + launcher::Action::MoveCursorHome, + ))), + KeyCode::End => Some(Action::Screen(ScreenAction::Launcher( + launcher::Action::MoveCursorEnd, + ))), + KeyCode::Char(ch) + if !key.modifiers.contains(KeyModifiers::CONTROL) + && !key.modifiers.contains(KeyModifiers::ALT) => + { + Some(Action::Screen(ScreenAction::Launcher( + launcher::Action::InsertChar(ch), + ))) + } + _ => None, + } +} + +fn route_monitor(key: KeyEvent) -> Option { + match key.code { + KeyCode::Up | KeyCode::Char('k') => Some(Action::Screen(ScreenAction::Monitor( + monitor::Action::MoveSelection(-1), + ))), + KeyCode::Down | KeyCode::Char('j') => Some(Action::Screen(ScreenAction::Monitor( + monitor::Action::MoveSelection(1), + ))), + KeyCode::PageUp => Some(Action::Screen(ScreenAction::Monitor( + monitor::Action::MoveSelection(-5), + ))), + KeyCode::PageDown => Some(Action::Screen(ScreenAction::Monitor( + monitor::Action::MoveSelection(5), + ))), + KeyCode::Enter => Some(Action::Screen(ScreenAction::Monitor( + monitor::Action::OpenSelected, + ))), + _ => None, + } +} + +fn route_cluster(id: &str, key: KeyEvent) -> Option { + let action = match key.code { + KeyCode::Tab | KeyCode::Right | KeyCode::Char('l') => { + cluster::Action::CycleFocus(cluster::FocusDirection::Next) + } + KeyCode::BackTab | KeyCode::Left | KeyCode::Char('h') => { + cluster::Action::CycleFocus(cluster::FocusDirection::Prev) + } + KeyCode::Up | KeyCode::Char('k') => cluster::Action::MoveFocused(-1), + KeyCode::Down | KeyCode::Char('j') => cluster::Action::MoveFocused(1), + KeyCode::PageUp => cluster::Action::MoveFocused(-5), + KeyCode::PageDown => cluster::Action::MoveFocused(5), + KeyCode::Enter => cluster::Action::ActivateFocused, + _ => return None, + }; + + Some(Action::Screen(ScreenAction::Cluster { + id: id.to_string(), + action, + })) +} + +fn route_agent(cluster_id: &str, agent_id: &str, key: KeyEvent) -> Option { + let action = match key.code { + KeyCode::Enter => agent::Action::SubmitGuidance, + KeyCode::Backspace => agent::Action::Backspace, + KeyCode::Delete => agent::Action::Delete, + KeyCode::Left => agent::Action::MoveCursorLeft, + KeyCode::Right => agent::Action::MoveCursorRight, + KeyCode::Home => agent::Action::MoveCursorHome, + KeyCode::End => agent::Action::MoveCursorEnd, + KeyCode::Up | KeyCode::Char('k') => agent::Action::ScrollLogs(-1), + KeyCode::Down | KeyCode::Char('j') => agent::Action::ScrollLogs(1), + KeyCode::PageUp => agent::Action::ScrollLogs(-5), + KeyCode::PageDown => agent::Action::ScrollLogs(5), + KeyCode::Char(ch) + if !key.modifiers.contains(KeyModifiers::CONTROL) + && !key.modifiers.contains(KeyModifiers::ALT) => + { + agent::Action::InsertChar(ch) + } + _ => return None, + }; + + Some(Action::Screen(ScreenAction::Agent { + cluster_id: cluster_id.to_string(), + agent_id: agent_id.to_string(), + action, + })) +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn disruptive_esc_pops_microscope() { + let mut state = AppState::default(); + state.ui_variant = UiVariant::Disruptive; + state.screen_stack = vec![ + ScreenId::ClusterCanvas { + id: "cluster-1".to_string(), + }, + ScreenId::AgentMicroscope { + cluster_id: "cluster-1".to_string(), + agent_id: "agent-1".to_string(), + }, + ]; + state.spine.mode = SpineMode::WhisperAgent; + state.spine.input.input.clear(); + state.spine.input.cursor = 0; + state.spine.completion = None; + + let action = route_key(&state, KeyEvent::new(KeyCode::Esc, KeyModifiers::NONE)); + assert!(matches!( + action, + Some(Action::Navigate(NavigationAction::Pop)) + )); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/lib.rs b/tui-rs/crates/zeroshot-tui/src/lib.rs new file mode 100644 index 00000000..0e761b83 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/lib.rs @@ -0,0 +1,15 @@ +// Pre-existing clippy lints in backend/terminal modules ā€” fix in a dedicated cleanup PR. +#![allow(clippy::type_complexity)] +#![allow(clippy::field_reassign_with_default)] +#![allow(clippy::question_mark)] +#![allow(clippy::while_let_loop)] +#![allow(clippy::needless_return)] + +pub mod app; +pub mod backend; +pub mod commands; +pub mod input; +pub mod protocol; +pub mod screens; +pub mod terminal; +pub mod ui; diff --git a/tui-rs/crates/zeroshot-tui/src/main.rs b/tui-rs/crates/zeroshot-tui/src/main.rs new file mode 100644 index 00000000..11412087 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/main.rs @@ -0,0 +1,557 @@ +#![allow(clippy::needless_return)] +#![allow(clippy::io_other_error)] + +use std::env; +use std::io::{self, stdout}; +use std::sync::mpsc; +use std::thread; +use std::time::{Duration, Instant, SystemTime, UNIX_EPOCH}; + +use crossterm::event::{self, Event, KeyEventKind}; +use ratatui::backend::CrosstermBackend; +use ratatui::Terminal; + +use zeroshot_tui::app::{ + resolve_ui_variant, Action, AppState, BackendAction, BackendRequest, Effect, InitialScreen, + StartupOptions, +}; +use zeroshot_tui::backend::stdio::StdioBackendClient; +use zeroshot_tui::backend::{BackendClient, BackendConfig, BackendError, BackendEvent}; +use zeroshot_tui::commands; +use zeroshot_tui::input; +use zeroshot_tui::terminal::TerminalGuard; +use zeroshot_tui::ui; + +type ActionSender = mpsc::Sender; +type ActionReceiver = mpsc::Receiver; +type TuiTerminal = Terminal>; + +const INITIAL_SCREEN_ENV: &str = "ZEROSHOT_TUI_INITIAL_SCREEN"; +const PROVIDER_OVERRIDE_ENV: &str = "ZEROSHOT_TUI_PROVIDER_OVERRIDE"; +const UI_VARIANT_ENV: &str = "ZEROSHOT_TUI_UI"; + +fn main() -> io::Result<()> { + if handle_cli_flags()? { + return Ok(()); + } + run_app() +} + +fn run_app() -> io::Result<()> { + let guard = init_terminal_guard()?; + let mut terminal = setup_terminal()?; + maybe_force_panic(); + let startup_options = parse_startup_options()?; + + let (action_tx, action_rx) = mpsc::channel::(); + let mut backend = connect_backend(&action_tx)?; + let mut state = AppState::new(); + state.apply_startup_options(startup_options); + let tick_rate = Duration::from_millis(250); + let mut last_tick = Instant::now(); + + app_loop( + &mut terminal, + &mut state, + &action_tx, + &action_rx, + &mut backend, + tick_rate, + &mut last_tick, + )?; + + shutdown_backend(backend)?; + drop(terminal); + guard.restore()?; + Ok(()) +} + +fn handle_cli_flags() -> io::Result { + let args: Vec = env::args().skip(1).collect(); + if args.iter().any(|arg| arg == "--version") { + println!("{}", env!("CARGO_PKG_VERSION")); + return Ok(true); + } + if args.iter().any(|arg| arg == "--smoke-test") { + println!("ok"); + return Ok(true); + } + Ok(false) +} + +fn init_terminal_guard() -> io::Result { + let guard = TerminalGuard::new()?; + guard.install_panic_hook(); + Ok(guard) +} + +fn setup_terminal() -> io::Result { + Terminal::new(CrosstermBackend::new(stdout())) +} + +fn maybe_force_panic() { + if env::var("ZEROSHOT_TUI_PANIC").ok().as_deref() == Some("1") { + panic!("ZEROSHOT_TUI_PANIC=1 requested"); + } +} + +fn parse_startup_options() -> io::Result { + let mut options = StartupOptions::default(); + let mut args = env::args().skip(1); + let mut ui_arg: Option = None; + + while let Some(arg) = args.next() { + match arg.as_str() { + "--initial-screen" => { + let value = args.next().ok_or_else(|| { + io::Error::new( + io::ErrorKind::InvalidInput, + "--initial-screen requires a value", + ) + })?; + options.initial_screen = Some(parse_initial_screen(&value)?); + } + "--provider-override" => { + let value = args.next().ok_or_else(|| { + io::Error::new( + io::ErrorKind::InvalidInput, + "--provider-override requires a value", + ) + })?; + if !value.trim().is_empty() { + options.provider_override = Some(value.trim().to_string()); + } + } + "--ui" => { + let value = args.next().ok_or_else(|| { + io::Error::new(io::ErrorKind::InvalidInput, "--ui requires a value") + })?; + ui_arg = Some(value); + } + _ => {} + } + } + + if options.initial_screen.is_none() { + if let Ok(value) = env::var(INITIAL_SCREEN_ENV) { + if !value.trim().is_empty() { + options.initial_screen = Some(parse_initial_screen(&value)?); + } + } + } + + if options.provider_override.is_none() { + if let Ok(value) = env::var(PROVIDER_OVERRIDE_ENV) { + if !value.trim().is_empty() { + options.provider_override = Some(value.trim().to_string()); + } + } + } + + let env_ui = env::var(UI_VARIANT_ENV).ok(); + options.ui_variant = resolve_ui_variant(ui_arg.as_deref(), env_ui.as_deref()) + .map_err(|err| io::Error::new(io::ErrorKind::InvalidInput, err))?; + + Ok(options) +} + +fn parse_initial_screen(value: &str) -> io::Result { + InitialScreen::parse(value).map_err(|err| io::Error::new(io::ErrorKind::InvalidInput, err)) +} + +fn app_loop( + terminal: &mut TuiTerminal, + state: &mut AppState, + action_tx: &ActionSender, + action_rx: &ActionReceiver, + backend: &mut Option, + tick_rate: Duration, + last_tick: &mut Instant, +) -> io::Result<()> { + loop { + handle_terminal_events(state, action_tx, tick_rate, last_tick)?; + drain_actions(state, action_rx, backend, action_tx)?; + terminal.draw(|frame| ui::render(frame, state))?; + + if state.should_quit { + break; + } + } + + Ok(()) +} + +fn handle_terminal_events( + state: &AppState, + action_tx: &ActionSender, + tick_rate: Duration, + last_tick: &mut Instant, +) -> io::Result<()> { + let timeout = tick_rate.saturating_sub(last_tick.elapsed()); + if event::poll(timeout)? { + match event::read()? { + Event::Key(key) if key.kind == KeyEventKind::Press => { + if let Some(action) = input::route_key(state, key) { + send_action(action_tx, action)?; + } + } + Event::Resize(width, height) => { + send_action(action_tx, Action::Resize { width, height })?; + } + _ => {} + } + } + + if last_tick.elapsed() >= tick_rate { + send_action(action_tx, Action::Tick { now_ms: now_ms() })?; + *last_tick = Instant::now(); + } + + Ok(()) +} + +fn now_ms() -> i64 { + SystemTime::now() + .duration_since(UNIX_EPOCH) + .map(|duration| duration.as_millis() as i64) + .unwrap_or(0) +} + +fn drain_actions( + state: &mut AppState, + action_rx: &ActionReceiver, + backend: &mut Option, + action_tx: &ActionSender, +) -> io::Result<()> { + loop { + match action_rx.try_recv() { + Ok(action) => { + let (next_state, effects) = + zeroshot_tui::app::update(std::mem::take(state), action); + *state = next_state; + execute_effects(effects, backend, action_tx)?; + } + Err(mpsc::TryRecvError::Empty) => break, + Err(mpsc::TryRecvError::Disconnected) => { + return Err(io::Error::new( + io::ErrorKind::BrokenPipe, + "action channel disconnected", + )); + } + } + } + + Ok(()) +} + +fn connect_backend(action_tx: &ActionSender) -> io::Result> { + match StdioBackendClient::connect(BackendConfig::default()) { + Ok(mut client) => { + if let Some(events) = client.take_event_receiver() { + let tx = action_tx.clone(); + thread::spawn(move || handle_backend_events(events, tx)); + } + send_action(action_tx, Action::Backend(BackendAction::Connected))?; + Ok(Some(client)) + } + Err(err) => { + send_action( + action_tx, + Action::Backend(BackendAction::ConnectionFailed(err.to_string())), + )?; + Ok(None) + } + } +} + +fn handle_backend_events(events: mpsc::Receiver, action_tx: ActionSender) { + for event in events { + let action = match event { + BackendEvent::Notification(notification) => { + Action::Backend(BackendAction::Notification(notification)) + } + BackendEvent::BackendExited(exit) => { + Action::Backend(BackendAction::BackendExited(exit)) + } + }; + + if !send_action_thread(&action_tx, action) { + break; + } + } +} + +fn send_action(action_tx: &ActionSender, action: Action) -> io::Result<()> { + action_tx.send(action).map_err(|err| { + io::Error::new( + io::ErrorKind::BrokenPipe, + format!("action channel closed: {err}"), + ) + }) +} + +fn send_action_thread(action_tx: &ActionSender, action: Action) -> bool { + if let Err(err) = action_tx.send(action) { + eprintln!("backend event send failed: {err}"); + return false; + } + true +} + +fn execute_effects( + effects: Vec, + backend: &mut Option, + action_tx: &ActionSender, +) -> io::Result<()> { + for effect in effects { + match effect { + Effect::Backend(request) => { + if let Some(client) = backend.as_ref() { + match execute_backend_request(client, request) { + Ok(Some(action)) => { + send_action(action_tx, Action::Backend(action))?; + } + Ok(None) => {} + Err(err) => { + send_action( + action_tx, + Action::Backend(BackendAction::Error(err.to_string())), + )?; + } + } + } else { + send_action( + action_tx, + Action::Backend(BackendAction::ConnectionFailed( + "Backend unavailable".to_string(), + )), + )?; + } + } + Effect::Command(request) => match commands::dispatch(request) { + Ok(actions) => { + for action in actions { + send_action(action_tx, action)?; + } + } + Err(err) => { + send_action( + action_tx, + Action::Backend(BackendAction::Error(err.to_string())), + )?; + } + }, + } + } + + Ok(()) +} + +fn execute_backend_request( + client: &StdioBackendClient, + request: BackendRequest, +) -> Result, BackendError> { + match request { + BackendRequest::ListClusters => list_clusters(client), + BackendRequest::ListClusterMetrics { cluster_ids } => { + list_cluster_metrics(client, cluster_ids) + } + BackendRequest::GetClusterSummary { cluster_id } => get_cluster_summary(client, cluster_id), + BackendRequest::GetClusterTopology { cluster_id } => { + get_cluster_topology(client, cluster_id) + } + BackendRequest::SubscribeClusterLogs { + cluster_id, + agent_id, + } => subscribe_cluster_logs(client, cluster_id, agent_id), + BackendRequest::SubscribeClusterTimeline { cluster_id } => { + subscribe_cluster_timeline(client, cluster_id) + } + BackendRequest::StartClusterFromText { + text, + provider_override, + } => start_cluster_from_text(client, text, provider_override), + BackendRequest::StartClusterFromIssue { + reference, + provider_override, + } => start_cluster_from_issue(client, reference, provider_override), + BackendRequest::SendGuidanceToCluster { + cluster_id, + message, + } => send_guidance_to_cluster(client, cluster_id, message), + BackendRequest::SendGuidanceToAgent { + cluster_id, + agent_id, + message, + } => send_guidance_to_agent(client, cluster_id, agent_id, message), + BackendRequest::Unsubscribe { subscription_id } => unsubscribe(client, subscription_id), + } +} + +fn list_clusters(client: &StdioBackendClient) -> Result, BackendError> { + let result = client.list_clusters()?; + Ok(Some(BackendAction::ClustersListed(result.clusters))) +} + +fn list_cluster_metrics( + client: &StdioBackendClient, + cluster_ids: Option>, +) -> Result, BackendError> { + let result = client + .list_cluster_metrics(zeroshot_tui::protocol::ListClusterMetricsParams { cluster_ids })?; + Ok(Some(BackendAction::ClusterMetricsListed { + metrics: result.metrics, + })) +} + +fn get_cluster_summary( + client: &StdioBackendClient, + cluster_id: String, +) -> Result, BackendError> { + let result = client + .get_cluster_summary(zeroshot_tui::protocol::GetClusterSummaryParams { cluster_id })?; + Ok(Some(BackendAction::ClusterSummary { + summary: result.summary, + })) +} + +fn get_cluster_topology( + client: &StdioBackendClient, + cluster_id: String, +) -> Result, BackendError> { + match client.get_cluster_topology(zeroshot_tui::protocol::GetClusterTopologyParams { + cluster_id: cluster_id.clone(), + }) { + Ok(result) => Ok(Some(BackendAction::ClusterTopology { + cluster_id, + topology: result.topology, + })), + Err(err) => Ok(Some(BackendAction::ClusterTopologyError { + cluster_id, + message: err.to_string(), + })), + } +} + +fn subscribe_cluster_logs( + client: &StdioBackendClient, + cluster_id: String, + agent_id: Option, +) -> Result, BackendError> { + let result = + client.subscribe_cluster_logs(zeroshot_tui::protocol::SubscribeClusterLogsParams { + cluster_id: cluster_id.clone(), + agent_id: agent_id.clone(), + })?; + Ok(Some(BackendAction::SubscribedClusterLogs { + cluster_id, + agent_id, + subscription_id: result.subscription_id, + })) +} + +fn subscribe_cluster_timeline( + client: &StdioBackendClient, + cluster_id: String, +) -> Result, BackendError> { + let result = client.subscribe_cluster_timeline( + zeroshot_tui::protocol::SubscribeClusterTimelineParams { + cluster_id: cluster_id.clone(), + }, + )?; + Ok(Some(BackendAction::SubscribedClusterTimeline { + cluster_id, + subscription_id: result.subscription_id, + })) +} + +fn start_cluster_from_text( + client: &StdioBackendClient, + text: String, + provider_override: Option, +) -> Result, BackendError> { + let result = + client.start_cluster_from_text(zeroshot_tui::protocol::StartClusterFromTextParams { + text, + provider_override, + cluster_id: None, + })?; + Ok(Some(BackendAction::StartClusterResult { + cluster_id: result.cluster_id, + })) +} + +fn start_cluster_from_issue( + client: &StdioBackendClient, + reference: String, + provider_override: Option, +) -> Result, BackendError> { + let result = + client.start_cluster_from_issue(zeroshot_tui::protocol::StartClusterFromIssueParams { + r#ref: reference, + provider_override, + cluster_id: None, + })?; + Ok(Some(BackendAction::StartClusterResult { + cluster_id: result.cluster_id, + })) +} + +fn send_guidance_to_cluster( + client: &StdioBackendClient, + cluster_id: String, + message: String, +) -> Result, BackendError> { + client.send_guidance_to_cluster(zeroshot_tui::protocol::SendGuidanceToClusterParams { + cluster_id, + text: message, + timeout_ms: None, + })?; + Ok(None) +} + +fn send_guidance_to_agent( + client: &StdioBackendClient, + cluster_id: String, + agent_id: String, + message: String, +) -> Result, BackendError> { + match client.send_guidance_to_agent(zeroshot_tui::protocol::SendGuidanceToAgentParams { + cluster_id: cluster_id.clone(), + agent_id: agent_id.clone(), + text: message, + timeout_ms: None, + }) { + Ok(result) => Ok(Some(BackendAction::GuidanceToAgentResult { + cluster_id, + agent_id, + result: result.result, + })), + Err(err) => Ok(Some(BackendAction::GuidanceToAgentError { + cluster_id, + agent_id, + message: err.to_string(), + })), + } +} + +fn unsubscribe( + client: &StdioBackendClient, + subscription_id: String, +) -> Result, BackendError> { + client.unsubscribe(zeroshot_tui::protocol::UnsubscribeParams { subscription_id })?; + Ok(None) +} + +fn shutdown_backend(mut backend: Option) -> io::Result<()> { + if let Some(mut backend) = backend.take() { + backend.shutdown().map_err(|err| { + io::Error::new( + io::ErrorKind::Other, + format!("backend shutdown failed: {err}"), + ) + })?; + } + + Ok(()) +} diff --git a/tui-rs/crates/zeroshot-tui/src/protocol/mod.rs b/tui-rs/crates/zeroshot-tui/src/protocol/mod.rs new file mode 100644 index 00000000..0d30a4b7 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/protocol/mod.rs @@ -0,0 +1,393 @@ +use serde::{Deserialize, Serialize}; +use std::collections::HashMap; + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(untagged)] +pub enum JsonRpcId { + String(String), + Number(i64), +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +pub struct JsonRpcRequest { + pub jsonrpc: String, + pub id: JsonRpcId, + pub method: String, + #[serde(skip_serializing_if = "Option::is_none")] + pub params: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +pub struct JsonRpcNotification { + pub jsonrpc: String, + pub method: String, + #[serde(skip_serializing_if = "Option::is_none")] + pub params: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +pub struct JsonRpcSuccessResponse { + pub jsonrpc: String, + pub id: JsonRpcId, + pub result: T, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +pub struct JsonRpcErrorResponse { + pub jsonrpc: String, + pub id: JsonRpcId, + pub error: RpcError, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct RpcErrorData { + #[serde(skip_serializing_if = "Option::is_none")] + pub detail: Option, + #[serde(skip_serializing_if = "Option::is_none")] + pub fields: Option>, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct RpcError { + pub code: i64, + pub message: String, + #[serde(skip_serializing_if = "Option::is_none")] + pub data: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct ClusterSummary { + pub id: String, + pub state: String, + pub provider: Option, + pub created_at: i64, + pub agent_count: i64, + pub message_count: i64, + pub cwd: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct ClusterMetrics { + pub id: String, + pub supported: bool, + pub cpu_percent: Option, + pub memory_mb: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct ClusterLogLine { + pub id: String, + pub timestamp: i64, + pub text: String, + pub agent: Option, + pub role: Option, + pub sender: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct TimelineEvent { + pub id: String, + pub timestamp: i64, + pub topic: String, + pub label: String, + pub approved: Option, + pub sender: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct TopologyAgent { + pub id: String, + pub role: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct TopologyEdge { + pub from: String, + pub to: String, + pub topic: String, + pub kind: TopologyEdgeKind, + #[serde(skip_serializing_if = "Option::is_none")] + pub dynamic: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "lowercase")] +pub enum TopologyEdgeKind { + Trigger, + Publish, + Source, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct ClusterTopology { + pub agents: Vec, + pub edges: Vec, + pub topics: Vec, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct GuidanceDeliveryResult { + pub status: String, + pub reason: Option, + pub method: Option, + #[serde(skip_serializing_if = "Option::is_none")] + pub task_id: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct ClusterGuidanceSummary { + pub injected: i64, + pub queued: i64, + pub total: i64, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct ClusterGuidanceDelivery { + pub summary: ClusterGuidanceSummary, + pub agents: HashMap, + pub timestamp: i64, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct ClientInfo { + pub name: String, + pub version: String, + #[serde(skip_serializing_if = "Option::is_none")] + pub pid: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct ClientCapabilities { + #[serde(skip_serializing_if = "Option::is_none")] + pub wants_metrics: Option, + #[serde(skip_serializing_if = "Option::is_none")] + pub wants_topology: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct InitializeParams { + pub protocol_version: i64, + pub client: ClientInfo, + #[serde(skip_serializing_if = "Option::is_none")] + pub capabilities: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct ServerInfo { + pub name: String, + pub version: String, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct ServerCapabilities { + pub methods: Vec, + pub notifications: Vec, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct InitializeResult { + pub protocol_version: i64, + pub server: ServerInfo, + pub capabilities: ServerCapabilities, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct ListClustersResult { + pub clusters: Vec, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct GetClusterSummaryParams { + pub cluster_id: String, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct GetClusterSummaryResult { + pub summary: ClusterSummary, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct ListClusterMetricsParams { + #[serde(skip_serializing_if = "Option::is_none")] + pub cluster_ids: Option>, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct ListClusterMetricsResult { + pub metrics: Vec, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct StartClusterFromTextParams { + pub text: String, + #[serde(skip_serializing_if = "Option::is_none")] + pub provider_override: Option, + #[serde(skip_serializing_if = "Option::is_none")] + pub cluster_id: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct StartClusterFromIssueParams { + #[serde(rename = "ref")] + pub r#ref: String, + #[serde(skip_serializing_if = "Option::is_none")] + pub provider_override: Option, + #[serde(skip_serializing_if = "Option::is_none")] + pub cluster_id: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct StartClusterResult { + pub cluster_id: String, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct SendGuidanceToAgentParams { + pub cluster_id: String, + pub agent_id: String, + pub text: String, + #[serde(skip_serializing_if = "Option::is_none")] + pub timeout_ms: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct SendGuidanceToClusterParams { + pub cluster_id: String, + pub text: String, + #[serde(skip_serializing_if = "Option::is_none")] + pub timeout_ms: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct SendGuidanceToAgentResult { + pub result: GuidanceDeliveryResult, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct SendGuidanceToClusterResult { + pub result: ClusterGuidanceDelivery, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct SubscribeClusterLogsParams { + pub cluster_id: String, + #[serde(skip_serializing_if = "Option::is_none")] + pub agent_id: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct SubscribeClusterTimelineParams { + pub cluster_id: String, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct SubscribeResult { + pub subscription_id: String, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct UnsubscribeParams { + pub subscription_id: String, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct UnsubscribeResult { + pub removed: bool, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct GetClusterTopologyParams { + pub cluster_id: String, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct GetClusterTopologyResult { + pub topology: ClusterTopology, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct ClusterLogLinesParams { + pub subscription_id: String, + pub cluster_id: String, + pub lines: Vec, + #[serde(skip_serializing_if = "Option::is_none")] + pub dropped_count: Option, +} + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct ClusterTimelineEventsParams { + pub subscription_id: String, + pub cluster_id: String, + pub events: Vec, +} + +pub type InitializeRequest = JsonRpcRequest; +pub type InitializeResponse = JsonRpcSuccessResponse; +pub type ListClustersRequest = JsonRpcRequest; +pub type ListClustersResponse = JsonRpcSuccessResponse; +pub type GetClusterSummaryRequest = JsonRpcRequest; +pub type GetClusterSummaryResponse = JsonRpcSuccessResponse; +pub type ListClusterMetricsRequest = JsonRpcRequest; +pub type ListClusterMetricsResponse = JsonRpcSuccessResponse; +pub type StartClusterFromTextRequest = JsonRpcRequest; +pub type StartClusterFromTextResponse = JsonRpcSuccessResponse; +pub type StartClusterFromIssueRequest = JsonRpcRequest; +pub type StartClusterFromIssueResponse = JsonRpcSuccessResponse; +pub type SendGuidanceToAgentRequest = JsonRpcRequest; +pub type SendGuidanceToAgentResponse = JsonRpcSuccessResponse; +pub type SendGuidanceToClusterRequest = JsonRpcRequest; +pub type SendGuidanceToClusterResponse = JsonRpcSuccessResponse; +pub type SubscribeClusterLogsRequest = JsonRpcRequest; +pub type SubscribeClusterLogsResponse = JsonRpcSuccessResponse; +pub type SubscribeClusterTimelineRequest = JsonRpcRequest; +pub type SubscribeClusterTimelineResponse = JsonRpcSuccessResponse; +pub type UnsubscribeRequest = JsonRpcRequest; +pub type UnsubscribeResponse = JsonRpcSuccessResponse; +pub type GetClusterTopologyRequest = JsonRpcRequest; +pub type GetClusterTopologyResponse = JsonRpcSuccessResponse; + +pub type ClusterLogLinesNotification = JsonRpcNotification; +pub type ClusterTimelineEventsNotification = JsonRpcNotification; + +#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct ListClustersParams {} diff --git a/tui-rs/crates/zeroshot-tui/src/screens/agent.rs b/tui-rs/crates/zeroshot-tui/src/screens/agent.rs new file mode 100644 index 00000000..d8e5970d --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/screens/agent.rs @@ -0,0 +1,304 @@ +use ratatui::layout::{Constraint, Layout, Position, Rect}; +use ratatui::style::Style; +use ratatui::text::{Line, Span}; +use ratatui::widgets::{ + Block, Borders, Paragraph, Scrollbar, ScrollbarOrientation, ScrollbarState, Wrap, +}; +use ratatui::Frame; + +use crate::protocol::{ClusterLogLine, GuidanceDeliveryResult}; +use crate::ui::shared::{pane_block, InputState, ScrollableBuffer}; +use crate::ui::theme; +use crate::ui::widgets::stream; + +pub const MAX_LOG_LINES: usize = 500; + +#[derive(Debug, Clone)] +pub struct State { + pub logs: ScrollableBuffer, + pub log_drop_seq: u64, + pub log_subscription: Option, + pub guidance_input: InputState, + pub guidance_pending: bool, + pub last_guidance: Option, + pub last_guidance_error: Option, + pub role: Option, + pub status: Option, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum Action { + SubmitGuidance, + InsertChar(char), + Backspace, + Delete, + MoveCursorLeft, + MoveCursorRight, + MoveCursorHome, + MoveCursorEnd, + ScrollLogs(i32), +} + +impl State { + pub fn push_log_lines(&mut self, mut lines: Vec, dropped_count: Option) { + let mut to_push = Vec::new(); + if let Some(count) = dropped_count { + if count > 0 { + let line = ClusterLogLine { + id: format!("dropped-{}", self.log_drop_seq), + timestamp: lines.first().map(|line| line.timestamp).unwrap_or(0), + text: format!("[dropped {} log lines]", count), + agent: None, + role: None, + sender: None, + }; + self.log_drop_seq = self.log_drop_seq.saturating_add(1); + to_push.push(line); + } + } + + to_push.append(&mut lines); + self.logs.push_many(to_push); + } + + pub fn move_log_scroll(&mut self, delta: i32) { + self.logs.move_scroll(delta); + } + + pub fn apply_guidance_result(&mut self, result: GuidanceDeliveryResult) { + self.last_guidance = Some(result); + self.last_guidance_error = None; + self.guidance_pending = false; + self.guidance_input.clear(); + } + + pub fn apply_guidance_error(&mut self, message: String) { + self.last_guidance_error = Some(message); + self.guidance_pending = false; + } + + pub fn guidance_status_line(&self) -> String { + if self.guidance_pending { + return "Sending guidance...".to_string(); + } + if let Some(error) = &self.last_guidance_error { + return format!("Last send failed: {error}"); + } + if let Some(result) = &self.last_guidance { + return format!("Last send: {}", format_guidance_result(result)); + } + "Guidance ready. Enter to send.".to_string() + } +} + +impl Default for State { + fn default() -> Self { + Self { + logs: ScrollableBuffer::new(MAX_LOG_LINES), + log_drop_seq: 0, + log_subscription: None, + guidance_input: InputState::default(), + guidance_pending: false, + last_guidance: None, + last_guidance_error: None, + role: None, + status: None, + } + } +} + +pub fn render(frame: &mut Frame<'_>, area: Rect, state: &State, cluster_id: &str, agent_id: &str) { + let [header_area, logs_area, guidance_area] = Layout::vertical([ + Constraint::Length(3), + Constraint::Min(4), + Constraint::Length(4), + ]) + .areas(area); + + render_header(frame, header_area, state, cluster_id, agent_id); + render_logs(frame, logs_area, state); + render_guidance(frame, guidance_area, state); +} + +fn render_header( + frame: &mut Frame<'_>, + area: Rect, + state: &State, + cluster_id: &str, + agent_id: &str, +) { + let role = state.role.as_deref().unwrap_or("unknown"); + let status = state.status.as_deref().unwrap_or("unknown"); + let (status_dot, status_style) = match status { + "executing" | "running" | "active" => ("\u{25cf}", theme::status_style("running")), + "waiting" | "idle" => ("\u{25cf}", theme::status_style("pending")), + "error" | "failed" => ("\u{25cf}", theme::status_style("error")), + "done" | "completed" => ("\u{25cf}", theme::status_style("done")), + _ => ("\u{25cb}", theme::dim_style()), + }; + let agent_color = theme::agent_color(agent_id); + + let lines = vec![ + Line::from(vec![ + Span::styled(" Agent: ", theme::dim_style()), + Span::styled(agent_id, Style::default().fg(agent_color)), + Span::styled(" Role: ", theme::dim_style()), + Span::styled(role, theme::dim_style()), + Span::styled(" Status: ", theme::dim_style()), + Span::styled(status_dot, status_style), + Span::raw(" "), + Span::styled(status, status_style), + ]), + Line::from(vec![ + Span::styled(" Cluster: ", theme::dim_style()), + Span::styled(cluster_id, theme::muted_style()), + ]), + ]; + let widget = Paragraph::new(lines); + frame.render_widget(widget, area); +} + +fn render_logs(frame: &mut Frame<'_>, area: Rect, state: &State) { + let title = if state.logs.scroll_offset > 0 { + format!("Logs (up {})", state.logs.scroll_offset) + } else { + "Logs".to_string() + }; + let block = pane_block(title, true); + let inner = block.inner(area); + let height = inner.height as usize; + let lines = if state.logs.is_empty() || height == 0 { + stream::log_placeholder_lines(stream::LogPlaceholderContext::Agent) + } else { + let total = state.logs.len(); + let max_start = total.saturating_sub(height); + let start = max_start.saturating_sub(state.logs.scroll_offset.min(max_start)); + state + .logs + .items + .iter() + .skip(start) + .take(height) + .map(stream::format_log_line_styled) + .collect() + }; + let widget = Paragraph::new(lines) + .block(block) + .wrap(Wrap { trim: false }); + frame.render_widget(widget, area); + + // Scrollbar + if !state.logs.is_empty() && height > 0 { + let total = state.logs.len(); + let position = total + .saturating_sub(height) + .saturating_sub(state.logs.scroll_offset); + let mut scrollbar_state = + ScrollbarState::new(total.saturating_sub(height)).position(position); + frame.render_stateful_widget( + Scrollbar::new(ScrollbarOrientation::VerticalRight), + inner, + &mut scrollbar_state, + ); + } +} + +fn render_guidance(frame: &mut Frame<'_>, area: Rect, state: &State) { + let status_line = Line::from(state.guidance_status_line()); + let input_line = if state.guidance_input.input.is_empty() { + Line::from(Span::styled("Type guidance...", theme::muted_style())) + } else { + Line::from(state.guidance_input.input.as_str()) + }; + let lines = vec![status_line, input_line]; + let block = Block::default() + .title("Guidance (Enter to send)") + .borders(Borders::ALL) + .border_style(theme::focus_border_style()); + let input = Paragraph::new(lines).block(block); + frame.render_widget(input, area); + + if area.height > 3 && area.width > 2 { + let max_x = area.x + area.width.saturating_sub(2); + let cursor_x = area.x + 1 + state.guidance_input.cursor as u16; + let cursor_x = cursor_x.min(max_x); + let cursor_y = area.y + 2; + frame.set_cursor_position(Position::new(cursor_x, cursor_y)); + } +} + +pub fn format_guidance_result(result: &GuidanceDeliveryResult) -> String { + let mut parts = vec![result.status.clone()]; + if let Some(method) = &result.method { + parts.push(format!("via {method}")); + } + if let Some(task_id) = &result.task_id { + parts.push(format!("task {task_id}")); + } + if let Some(reason) = &result.reason { + parts.push(format!("reason {reason}")); + } + parts.join(" | ") +} + +// shared stream formatters live in ui/widgets/stream.rs + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn guidance_success_clears_input() { + let mut state = State::default(); + state.guidance_input.input = "keep".to_string(); + state.guidance_input.cursor = 4; + state.guidance_pending = true; + + let result = GuidanceDeliveryResult { + status: "injected".to_string(), + reason: None, + method: Some("pty".to_string()), + task_id: Some("task-1".to_string()), + }; + + state.apply_guidance_result(result.clone()); + + assert_eq!(state.guidance_input.input, ""); + assert_eq!(state.guidance_input.cursor, 0); + assert!(!state.guidance_pending); + assert_eq!(state.last_guidance, Some(result)); + assert_eq!(state.last_guidance_error, None); + } + + #[test] + fn guidance_error_preserves_input() { + let mut state = State::default(); + state.guidance_input.input = "stay".to_string(); + state.guidance_input.cursor = 4; + state.guidance_pending = true; + + state.apply_guidance_error("network".to_string()); + + assert_eq!(state.guidance_input.input, "stay"); + assert_eq!(state.guidance_input.cursor, 4); + assert!(!state.guidance_pending); + assert_eq!(state.last_guidance_error, Some("network".to_string())); + } + + #[test] + fn guidance_status_format_includes_details() { + let result = GuidanceDeliveryResult { + status: "queued".to_string(), + reason: Some("no tty".to_string()), + method: Some("queue".to_string()), + task_id: Some("task-9".to_string()), + }; + + let formatted = format_guidance_result(&result); + + assert!(formatted.contains("queued")); + assert!(formatted.contains("via queue")); + assert!(formatted.contains("task task-9")); + assert!(formatted.contains("reason no tty")); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/screens/agent_microscope.rs b/tui-rs/crates/zeroshot-tui/src/screens/agent_microscope.rs new file mode 100644 index 00000000..6855008a --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/screens/agent_microscope.rs @@ -0,0 +1,511 @@ +use ratatui::layout::Rect; +use ratatui::style::Style; +use ratatui::text::{Line, Span}; +use ratatui::widgets::{Block, Borders, Paragraph}; +use ratatui::Frame; + +use crate::app::{agent_microscope, TimeCursor}; +use crate::protocol::{ClusterLogLine, TimelineEvent}; +use crate::ui::shared::TimeIndexedBuffer; +use crate::ui::theme; +use crate::ui::widgets::stream::{self, StreamOverlay}; + +pub fn render( + frame: &mut Frame<'_>, + area: Rect, + cluster_id: &str, + agent_id: &str, + cluster_timeline: Option<&TimeIndexedBuffer>, + microscope_state: Option<&agent_microscope::State>, + time_cursor: &TimeCursor, +) { + if area.width == 0 || area.height == 0 { + return; + } + + let metadata = build_metadata_overlay(area, cluster_id, agent_id, microscope_state); + let reserved_lines = metadata + .as_ref() + .map(|overlay| overlay.reserved_lines) + .unwrap_or(0); + + let max_lines = area.height.saturating_sub(2) as usize; + let window_max = max_lines.saturating_sub(reserved_lines); + let log_entries = microscope_state + .map(|state| { + stream::select_time_window(&state.logs_time, time_cursor, window_max, |_| true) + }) + .unwrap_or_default() + .into_iter() + .collect::>(); + + let marker_margin = + build_phase_marker_margin(area, cluster_timeline, time_cursor, &log_entries); + + let mut content_lines = if log_entries.is_empty() { + stream::log_placeholder_lines(stream::LogPlaceholderContext::Agent) + } else { + let log_lines = log_entries + .iter() + .map(|line| stream::format_log_line_styled(line)) + .collect::>(); + if let Some(marker_margin) = marker_margin { + apply_phase_marker_margin(log_lines, &marker_margin) + } else { + log_lines + } + }; + + if reserved_lines > 0 { + let mut padded = Vec::with_capacity(reserved_lines + content_lines.len()); + for _ in 0..reserved_lines { + padded.push(Line::from("")); + } + padded.extend(content_lines); + content_lines = padded; + } + + let title = stream::overlay_title("Stream", time_cursor); + let overlay = + StreamOverlay::new(title, content_lines).border_style(theme::focus_border_style()); + frame.render_widget(overlay, area); + + if let Some(metadata) = metadata { + render_metadata_overlay(frame, metadata); + } +} + +struct PhaseMarkerMargin { + labels: Vec>, + margin_width: usize, +} + +struct MetadataOverlay { + area: Rect, + lines: Vec>, + reserved_lines: usize, +} + +fn build_metadata_overlay( + area: Rect, + cluster_id: &str, + agent_id: &str, + microscope_state: Option<&agent_microscope::State>, +) -> Option { + let available_width = area.width.saturating_sub(2); + let available_height = area.height.saturating_sub(2); + if available_width < 6 || available_height < 4 { + return None; + } + + let role = microscope_state + .and_then(|state| state.role.as_deref()) + .unwrap_or("unknown"); + let status = microscope_state + .and_then(|state| state.status.as_deref()) + .unwrap_or("unknown"); + + let raw_lines = [ + format!("Agent: {agent_id}"), + format!("Role: {role}"), + format!("Status: {status}"), + format!("Cluster: {cluster_id}"), + ]; + + let mut lines = vec![ + Line::from(vec![ + Span::styled("Agent: ", theme::muted_style()), + Span::styled( + agent_id.to_string(), + Style::default().fg(theme::agent_color(agent_id)), + ), + ]), + Line::from(vec![ + Span::styled("Role: ", theme::muted_style()), + Span::styled(role.to_string(), theme::dim_style()), + ]), + Line::from(vec![ + Span::styled("Status: ", theme::muted_style()), + Span::styled(status.to_string(), theme::status_style(status)), + ]), + Line::from(vec![ + Span::styled("Cluster: ", theme::muted_style()), + Span::styled(cluster_id.to_string(), theme::muted_style()), + ]), + ]; + + let max_line_len = raw_lines + .iter() + .map(|line| line.chars().count()) + .max() + .unwrap_or_default() as u16; + + let overlay_width = (max_line_len + 2).min(available_width); + let overlay_height = (lines.len() as u16 + 2).min(available_height); + if overlay_width < 4 || overlay_height < 3 || overlay_height >= available_height { + return None; + } + + let max_lines = overlay_height.saturating_sub(2) as usize; + if lines.len() > max_lines { + lines.truncate(max_lines); + } + + let overlay_area = Rect { + x: area.x + 1, + y: area.y + 1, + width: overlay_width, + height: overlay_height, + }; + + Some(MetadataOverlay { + area: overlay_area, + lines, + reserved_lines: overlay_height as usize, + }) +} + +fn render_metadata_overlay(frame: &mut Frame<'_>, metadata: MetadataOverlay) { + let block = Block::default() + .borders(Borders::ALL) + .border_style(theme::unfocus_border_style()); + let widget = Paragraph::new(metadata.lines).block(block); + frame.render_widget(widget, metadata.area); +} + +fn build_phase_marker_margin( + area: Rect, + timeline: Option<&TimeIndexedBuffer>, + time_cursor: &TimeCursor, + log_entries: &[&ClusterLogLine], +) -> Option { + let timeline = timeline?; + if log_entries.is_empty() { + return None; + } + let available_width = area.width.saturating_sub(2); + let margin_width = phase_marker_margin_width(available_width); + if margin_width == 0 { + return None; + } + + let markers = stream::derive_phase_markers(timeline, time_cursor, stream::PHASE_MARKER_LIMIT); + if markers.is_empty() { + return None; + } + + let mut labels = vec![None; log_entries.len()]; + for marker in markers { + let index = marker_line_index(log_entries, marker.timestamp_ms); + let raw = stream::format_phase_marker_label(&marker.topic, &marker.label); + let truncated = stream::truncate_marker_label(&raw, margin_width); + labels[index] = Some(pad_phase_label(&truncated, margin_width)); + } + + Some(PhaseMarkerMargin { + labels, + margin_width, + }) +} + +fn phase_marker_margin_width(available_width: u16) -> usize { + let available = available_width as usize; + let min_content = 16usize; + let min_margin = 8usize; + if available < min_content + min_margin { + return 0; + } + let mut margin = available / 4; + if margin < min_margin { + margin = min_margin; + } + if margin > 14 { + margin = 14; + } + if available.saturating_sub(margin) < min_content { + return 0; + } + margin +} + +fn marker_line_index(log_entries: &[&ClusterLogLine], timestamp: i64) -> usize { + if log_entries.is_empty() { + return 0; + } + let mut left = 0usize; + let mut right = log_entries.len(); + while left < right { + let mid = left + (right - left) / 2; + if log_entries[mid].timestamp < timestamp { + left = mid + 1; + } else { + right = mid; + } + } + if left >= log_entries.len() { + log_entries.len().saturating_sub(1) + } else { + left + } +} + +fn pad_phase_label(label: &str, width: usize) -> String { + let len = label.chars().count(); + if len >= width { + return label.to_string(); + } + let mut out = String::with_capacity(width); + out.push_str(label); + for _ in 0..(width - len) { + out.push(' '); + } + out +} + +fn apply_phase_marker_margin<'a>( + lines: Vec>, + marker_margin: &PhaseMarkerMargin, +) -> Vec> { + let empty_label = " ".repeat(marker_margin.margin_width); + lines + .into_iter() + .enumerate() + .map(|(idx, line)| { + let label = marker_margin + .labels + .get(idx) + .and_then(|label| label.as_ref()) + .unwrap_or(&empty_label); + prepend_phase_marker_label(line, label, marker_margin.margin_width) + }) + .collect() +} + +fn prepend_phase_marker_label<'a>(line: Line<'a>, label: &str, margin_width: usize) -> Line<'a> { + let mut spans = Vec::with_capacity(line.spans.len() + 2); + let mut label_text = label.to_string(); + if label_text.chars().count() < margin_width { + label_text = pad_phase_label(&label_text, margin_width); + } + spans.push(Span::styled(label_text, theme::muted_style())); + spans.push(Span::raw(" ")); + spans.extend(line.spans); + Line { + style: line.style, + alignment: line.alignment, + spans, + } +} + +#[cfg(test)] +mod tests { + use super::*; + use ratatui::backend::TestBackend; + use ratatui::Terminal; + + use crate::app::TimeCursorMode; + use crate::protocol::ClusterLogLine; + use crate::ui::widgets::test_utils::line_text; + + fn buffer_contains(terminal: &Terminal, needle: &str) -> bool { + let buffer = terminal.backend().buffer(); + for y in 0..buffer.area.height { + if line_text(buffer, y).contains(needle) { + return true; + } + } + false + } + + fn sample_state(lines: Vec) -> agent_microscope::State { + let mut state = agent_microscope::State::default(); + state.push_log_lines(lines, None); + state + } + + fn sample_timeline_event(id: &str, timestamp: i64, topic: &str, label: &str) -> TimelineEvent { + TimelineEvent { + id: id.to_string(), + timestamp, + topic: topic.to_string(), + label: label.to_string(), + approved: None, + sender: None, + } + } + + #[test] + fn agent_microscope_renders_empty_state() { + let backend = TestBackend::new(70, 12); + let mut terminal = Terminal::new(backend).expect("terminal"); + let cursor = TimeCursor { + mode: TimeCursorMode::Live, + t_ms: 0, + window_ms: 60, + }; + + terminal + .draw(|frame| { + let area = frame.area(); + render(frame, area, "cluster-1", "agent-1", None, None, &cursor); + }) + .expect("draw"); + + assert!(buffer_contains(&terminal, "No logs yet.")); + assert!(buffer_contains(&terminal, "Agent: agent-1")); + } + + #[test] + fn agent_microscope_renders_live_mode() { + let backend = TestBackend::new(70, 12); + let mut terminal = Terminal::new(backend).expect("terminal"); + let state = sample_state(vec![ + ClusterLogLine { + id: "old".to_string(), + timestamp: 100, + text: "old-line".to_string(), + agent: Some("agent-1".to_string()), + role: None, + sender: None, + }, + ClusterLogLine { + id: "new".to_string(), + timestamp: 300, + text: "new-line".to_string(), + agent: Some("agent-1".to_string()), + role: None, + sender: None, + }, + ]); + + let cursor = TimeCursor { + mode: TimeCursorMode::Live, + t_ms: 300, + window_ms: 60, + }; + + terminal + .draw(|frame| { + let area = frame.area(); + render( + frame, + area, + "cluster-1", + "agent-1", + None, + Some(&state), + &cursor, + ); + }) + .expect("draw"); + + assert!(buffer_contains(&terminal, "new-line")); + } + + #[test] + fn agent_microscope_renders_scrub_window() { + let backend = TestBackend::new(70, 14); + let mut terminal = Terminal::new(backend).expect("terminal"); + let state = sample_state(vec![ + ClusterLogLine { + id: "old".to_string(), + timestamp: 100, + text: "old-line".to_string(), + agent: Some("agent-1".to_string()), + role: None, + sender: None, + }, + ClusterLogLine { + id: "mid".to_string(), + timestamp: 220, + text: "mid-line".to_string(), + agent: Some("agent-1".to_string()), + role: None, + sender: None, + }, + ClusterLogLine { + id: "new".to_string(), + timestamp: 400, + text: "new-line".to_string(), + agent: Some("agent-1".to_string()), + role: None, + sender: None, + }, + ]); + + let cursor = TimeCursor { + mode: TimeCursorMode::Scrub, + t_ms: 230, + window_ms: 60, + }; + + terminal + .draw(|frame| { + let area = frame.area(); + render( + frame, + area, + "cluster-1", + "agent-1", + None, + Some(&state), + &cursor, + ); + }) + .expect("draw"); + + assert!(buffer_contains(&terminal, "mid-line")); + assert!(!buffer_contains(&terminal, "old-line")); + assert!(!buffer_contains(&terminal, "new-line")); + } + + #[test] + fn agent_microscope_renders_phase_markers_margin() { + let backend = TestBackend::new(70, 12); + let mut terminal = Terminal::new(backend).expect("terminal"); + let state = sample_state(vec![ + ClusterLogLine { + id: "line-1".to_string(), + timestamp: 100, + text: "first-line".to_string(), + agent: Some("agent-1".to_string()), + role: None, + sender: None, + }, + ClusterLogLine { + id: "line-2".to_string(), + timestamp: 220, + text: "second-line".to_string(), + agent: Some("agent-1".to_string()), + role: None, + sender: None, + }, + ]); + let mut timeline = TimeIndexedBuffer::new(16); + timeline.push_many(vec![sample_timeline_event("event-1", 150, "plan", "start")]); + + let cursor = TimeCursor { + mode: TimeCursorMode::Live, + t_ms: 220, + window_ms: 60, + }; + + terminal + .draw(|frame| { + let area = frame.area(); + render( + frame, + area, + "cluster-1", + "agent-1", + Some(&timeline), + Some(&state), + &cursor, + ); + }) + .expect("draw"); + + assert!(buffer_contains(&terminal, "plan: start")); + assert!(buffer_contains(&terminal, "second-line")); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/screens/cluster.rs b/tui-rs/crates/zeroshot-tui/src/screens/cluster.rs new file mode 100644 index 00000000..e63d6b8c --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/screens/cluster.rs @@ -0,0 +1,414 @@ +use ratatui::layout::{Constraint, Layout, Rect}; +use ratatui::style::Style; +use ratatui::text::{Line, Span}; +use ratatui::widgets::{ + List, ListItem, ListState, Paragraph, Scrollbar, ScrollbarOrientation, ScrollbarState, Wrap, +}; +use ratatui::Frame; + +use crate::protocol::{ + ClusterLogLine, ClusterMetrics, ClusterSummary, ClusterTopology, TimelineEvent, +}; +use crate::screens::metrics; +use crate::ui::shared::{pane_block, HasTimestamp, ScrollableBuffer, TimeIndexedBuffer}; +use crate::ui::theme; +use crate::ui::widgets::{stream, topology}; + +pub const MAX_LOG_LINES: usize = 1000; +pub const MAX_TIMELINE_EVENTS: usize = 500; + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum FocusDirection { + Next, + Prev, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum ClusterPane { + Topology, + Logs, + Timeline, + Agents, +} + +impl ClusterPane { + fn next(&self) -> Self { + match self { + ClusterPane::Topology => ClusterPane::Logs, + ClusterPane::Logs => ClusterPane::Timeline, + ClusterPane::Timeline => ClusterPane::Agents, + ClusterPane::Agents => ClusterPane::Topology, + } + } + + fn prev(&self) -> Self { + match self { + ClusterPane::Topology => ClusterPane::Agents, + ClusterPane::Logs => ClusterPane::Topology, + ClusterPane::Timeline => ClusterPane::Logs, + ClusterPane::Agents => ClusterPane::Timeline, + } + } +} + +#[derive(Debug, Clone)] +pub struct State { + pub focus: ClusterPane, + pub summary: Option, + pub topology: Option, + pub topology_error: Option, + pub logs: ScrollableBuffer, + pub logs_time: TimeIndexedBuffer, + pub timeline: ScrollableBuffer, + pub timeline_time: TimeIndexedBuffer, + pub agents: Vec, + pub selected_agent: usize, + pub log_drop_seq: u64, + pub log_subscription: Option, + pub timeline_subscription: Option, +} + +impl Default for State { + fn default() -> Self { + Self { + focus: ClusterPane::Topology, + summary: None, + topology: None, + topology_error: None, + logs: ScrollableBuffer::new(MAX_LOG_LINES), + logs_time: TimeIndexedBuffer::new(MAX_LOG_LINES), + timeline: ScrollableBuffer::new(MAX_TIMELINE_EVENTS), + timeline_time: TimeIndexedBuffer::new(MAX_TIMELINE_EVENTS), + agents: Vec::new(), + selected_agent: 0, + log_drop_seq: 0, + log_subscription: None, + timeline_subscription: None, + } + } +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct AgentInfo { + pub id: String, + pub role: Option, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum Action { + CycleFocus(FocusDirection), + MoveFocused(i32), + ActivateFocused, + OpenAgent(String), +} + +impl State { + pub fn cycle_focus(&mut self, direction: FocusDirection) { + self.focus = match direction { + FocusDirection::Next => self.focus.next(), + FocusDirection::Prev => self.focus.prev(), + }; + } + + pub fn move_focused(&mut self, delta: i32) { + match self.focus { + ClusterPane::Topology => {} + ClusterPane::Logs => self.logs.move_scroll(delta), + ClusterPane::Timeline => self.timeline.move_scroll(delta), + ClusterPane::Agents => self.move_agent_selection(delta), + } + } + + pub fn activate_focused(&self) -> Option { + match self.focus { + ClusterPane::Agents => self.selected_agent_id(), + _ => None, + } + } + + pub fn push_log_lines(&mut self, mut lines: Vec, dropped_count: Option) { + self.update_agents_from_logs(&lines); + + let mut to_push = Vec::new(); + if let Some(count) = dropped_count { + if count > 0 { + let line = ClusterLogLine { + id: format!("dropped-{}", self.log_drop_seq), + timestamp: lines.first().map(|line| line.timestamp).unwrap_or(0), + text: format!("[dropped {} log lines]", count), + agent: None, + role: None, + sender: None, + }; + self.log_drop_seq = self.log_drop_seq.saturating_add(1); + to_push.push(line); + } + } + + to_push.append(&mut lines); + let time_lines = to_push.clone(); + self.logs.push_many(to_push); + self.logs_time.push_many(time_lines); + } + + pub fn push_timeline_events(&mut self, events: Vec) { + let time_events = events.clone(); + self.timeline.push_many(events); + self.timeline_time.push_many(time_events); + } + + fn move_agent_selection(&mut self, delta: i32) { + if self.agents.is_empty() { + self.selected_agent = 0; + return; + } + + let len = self.agents.len() as i32; + let mut next = self.selected_agent as i32 + delta; + if next < 0 { + next = 0; + } + if next >= len { + next = len - 1; + } + self.selected_agent = next as usize; + } + + fn update_agents_from_logs(&mut self, lines: &[ClusterLogLine]) { + let selected_id = self.selected_agent_id(); + for line in lines { + let Some(agent_id) = line.agent.as_ref() else { + continue; + }; + + match self.agents.iter_mut().find(|agent| agent.id == *agent_id) { + Some(agent) => { + if agent.role.is_none() { + agent.role = line.role.clone(); + } + } + None => { + self.agents.push(AgentInfo { + id: agent_id.clone(), + role: line.role.clone(), + }); + } + } + } + self.reconcile_agent_selection(selected_id); + } + + fn reconcile_agent_selection(&mut self, selected_id: Option) { + if let Some(id) = selected_id { + if let Some(index) = self.agents.iter().position(|agent| agent.id == id) { + self.selected_agent = index; + return; + } + } + + if self.selected_agent >= self.agents.len() { + self.selected_agent = self.agents.len().saturating_sub(1); + } + } + + fn selected_agent_id(&self) -> Option { + self.agents + .get(self.selected_agent) + .map(|agent| agent.id.clone()) + } +} + +impl HasTimestamp for ClusterLogLine { + fn timestamp_ms(&self) -> i64 { + self.timestamp + } +} + +impl HasTimestamp for TimelineEvent { + fn timestamp_ms(&self) -> i64 { + self.timestamp + } +} + +pub fn render(frame: &mut Frame<'_>, area: Rect, state: &State, metrics: Option<&ClusterMetrics>) { + let [metrics_area, content] = + Layout::vertical([Constraint::Length(1), Constraint::Min(1)]).areas(area); + + render_metrics_line(frame, metrics_area, metrics); + + let [top, bottom] = + Layout::vertical([Constraint::Percentage(30), Constraint::Percentage(70)]).areas(content); + + let [topo_area, agents_area] = + Layout::horizontal([Constraint::Percentage(50), Constraint::Percentage(50)]).areas(top); + + let [logs_area, timeline_area] = + Layout::horizontal([Constraint::Percentage(50), Constraint::Percentage(50)]).areas(bottom); + + render_topology(frame, topo_area, state); + render_agents(frame, agents_area, state); + render_logs(frame, logs_area, state); + render_timeline(frame, timeline_area, state); +} + +fn render_metrics_line(frame: &mut Frame<'_>, area: Rect, metrics: Option<&ClusterMetrics>) { + let line = Line::from(vec![ + Span::styled("Metrics:", theme::dim_style()), + Span::raw(" "), + Span::styled(metrics::format_metrics_line(metrics), theme::dim_style()), + ]); + let widget = Paragraph::new(line); + frame.render_widget(widget, area); +} + +fn render_topology(frame: &mut Frame<'_>, area: Rect, state: &State) { + let block = pane_block("Topology", state.focus == ClusterPane::Topology); + topology::render( + frame, + area, + block, + state.summary.as_ref(), + state.topology.as_ref(), + state.topology_error.as_deref(), + ); +} + +fn render_agents(frame: &mut Frame<'_>, area: Rect, state: &State) { + let title = format!("Agents ({})", state.agents.len()); + let block = pane_block(title, state.focus == ClusterPane::Agents); + let inner = block.inner(area); + + if state.agents.is_empty() || inner.height == 0 { + let lines = vec![ + Line::from(Span::styled("No agents yet.", theme::muted_style())), + Line::from(Span::styled( + "Wait for logs to identify agents.", + theme::muted_style(), + )), + ]; + let widget = Paragraph::new(lines) + .block(block) + .wrap(Wrap { trim: false }); + frame.render_widget(widget, area); + return; + } + + let items: Vec = state + .agents + .iter() + .map(|agent| { + let agent_color = theme::agent_color(&agent.id); + let text = match &agent.role { + Some(role) => format!("{} ({role})", agent.id), + None => agent.id.clone(), + }; + ListItem::new(text).style(Style::default().fg(agent_color)) + }) + .collect(); + + let list = List::new(items) + .block(block) + .highlight_style(theme::selected_style()) + .highlight_symbol(" > "); + + let mut list_state = ListState::default(); + list_state.select(Some(state.selected_agent)); + + frame.render_stateful_widget(list, area, &mut list_state); +} + +fn render_logs(frame: &mut Frame<'_>, area: Rect, state: &State) { + let title = if state.logs.scroll_offset > 0 { + format!("Logs (up {})", state.logs.scroll_offset) + } else { + "Logs".to_string() + }; + let block = pane_block(title, state.focus == ClusterPane::Logs); + let inner = block.inner(area); + let height = inner.height as usize; + + let lines: Vec = if state.logs.is_empty() || height == 0 { + stream::log_placeholder_lines(stream::LogPlaceholderContext::Cluster) + } else { + let total = state.logs.len(); + let max_start = total.saturating_sub(height); + let start = max_start.saturating_sub(state.logs.scroll_offset.min(max_start)); + state + .logs + .items + .iter() + .skip(start) + .take(height) + .map(stream::format_log_line_styled) + .collect() + }; + + let widget = Paragraph::new(lines) + .block(block) + .wrap(Wrap { trim: false }); + frame.render_widget(widget, area); + + // Scrollbar + if !state.logs.is_empty() && height > 0 { + let total = state.logs.len(); + let position = total + .saturating_sub(height) + .saturating_sub(state.logs.scroll_offset); + let mut scrollbar_state = + ScrollbarState::new(total.saturating_sub(height)).position(position); + frame.render_stateful_widget( + Scrollbar::new(ScrollbarOrientation::VerticalRight), + inner, + &mut scrollbar_state, + ); + } +} + +fn render_timeline(frame: &mut Frame<'_>, area: Rect, state: &State) { + let title = if state.timeline.scroll_offset > 0 { + format!("Timeline (up {})", state.timeline.scroll_offset) + } else { + "Timeline".to_string() + }; + let block = pane_block(title, state.focus == ClusterPane::Timeline); + let inner = block.inner(area); + let height = inner.height as usize; + + let lines: Vec = if state.timeline.is_empty() || height == 0 { + stream::timeline_placeholder_lines() + } else { + let total = state.timeline.len(); + let max_start = total.saturating_sub(height); + let start = max_start.saturating_sub(state.timeline.scroll_offset.min(max_start)); + state + .timeline + .items + .iter() + .skip(start) + .take(height) + .map(stream::format_timeline_event_styled) + .collect() + }; + + let widget = Paragraph::new(lines) + .block(block) + .wrap(Wrap { trim: false }); + frame.render_widget(widget, area); + + // Scrollbar + if !state.timeline.is_empty() && height > 0 { + let total = state.timeline.len(); + let position = total + .saturating_sub(height) + .saturating_sub(state.timeline.scroll_offset); + let mut scrollbar_state = + ScrollbarState::new(total.saturating_sub(height)).position(position); + frame.render_stateful_widget( + Scrollbar::new(ScrollbarOrientation::VerticalRight), + inner, + &mut scrollbar_state, + ); + } +} + +// shared stream formatters live in ui/widgets/stream.rs diff --git a/tui-rs/crates/zeroshot-tui/src/screens/cluster_canvas.rs b/tui-rs/crates/zeroshot-tui/src/screens/cluster_canvas.rs new file mode 100644 index 00000000..beec1e64 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/screens/cluster_canvas.rs @@ -0,0 +1,1688 @@ +use std::collections::HashMap; + +use ratatui::layout::Rect; +use ratatui::style::{Modifier, Style}; +use ratatui::symbols::Marker; +use ratatui::text::{Line, Span}; +use ratatui::widgets::canvas::{Canvas, Circle, Line as CanvasLine, Points}; +use ratatui::widgets::{Block, Borders}; +use ratatui::Frame; + +use crate::app::animation::AnimClock; +use crate::app::{animation, FocusTarget, TimeCursor}; +use crate::protocol::{ClusterTopology, TopologyAgent}; +use crate::screens::cluster; +use crate::ui::shared::calm_empty_state; +use crate::ui::theme; +use crate::ui::widgets::stream::{self, StreamOverlay}; + +const WORLD_RADIUS: f64 = 48.0; +const AGENT_RING_RADIUS: f64 = 28.0; +const TOPIC_RING_RADIUS: f64 = 14.0; +const AGENT_ORB_RADIUS: f64 = 1.8; +const TOPIC_ORB_RADIUS: f64 = 1.2; +const LABEL_OFFSET: f64 = 4.0; +const LABEL_RADIAL_OFFSET: f64 = 1.4; +const LABEL_LIMIT: usize = 14; +const PENDING_MESSAGE: &str = "Topology pending"; +const UNAVAILABLE_MESSAGE: &str = "Topology unavailable"; +const FOCUS_EPSILON: f64 = 0.0001; +const CANVAS_CAMERA_ACCEL: f64 = 0.16; +const CANVAS_CAMERA_FRICTION: f64 = 0.82; +const CANVAS_CAMERA_SNAP_EPSILON: f64 = 0.08; + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum NodeKind { + Agent, + Topic, +} + +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub enum Direction { + Left, + Right, + Up, + Down, +} + +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub enum MoveSpeed { + Step, + Fast, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum Action { + MoveFocus { + direction: Direction, + speed: MoveSpeed, + }, + ZoomIn, +} + +#[derive(Debug, Clone, PartialEq)] +pub struct NodeLayout { + pub id: String, + pub label: String, + pub x: f64, + pub y: f64, + pub kind: NodeKind, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct LayoutEdge { + pub from: String, + pub to: String, +} + +#[derive(Debug, Clone, PartialEq)] +pub struct LayoutBounds { + pub min_x: f64, + pub max_x: f64, + pub min_y: f64, + pub max_y: f64, +} + +#[derive(Debug, Clone, PartialEq)] +pub struct LayoutCache { + pub nodes: HashMap, + pub edges: Vec, + pub bounds: LayoutBounds, +} + +#[derive(Debug, Clone, Default)] +pub struct State { + pub focused_id: Option, + pub layout: Option, + pub log_subscription: Option, + pub timeline_subscription: Option, + pub camera: (f64, f64), + pub camera_target: (f64, f64), + pub camera_velocity: (f64, f64), +} + +impl State { + pub fn update_layout(&mut self, topology: &ClusterTopology) { + let had_layout = self.layout.is_some(); + self.layout = Some(layout_for(topology)); + self.ensure_focus(topology); + if had_layout { + self.set_camera_target_to_focus(); + } else { + self.snap_camera_to_focus(); + } + } + + pub fn ensure_focus(&mut self, topology: &ClusterTopology) { + let mut needs_focus = self.focused_id.is_none(); + if let Some(focused) = self.focused_id.as_ref() { + let in_agents = topology.agents.iter().any(|agent| agent.id == *focused); + let in_topics = topology.topics.iter().any(|topic| topic == focused); + if !in_agents && !in_topics { + needs_focus = true; + } + } + if needs_focus { + self.focused_id = default_focus_id(topology); + } + } + + pub fn move_focus(&mut self, direction: Direction, speed: MoveSpeed) { + if self.layout.is_none() { + return; + } + if let Some(layout) = self.layout.as_ref() { + match self.focused_id.as_ref() { + Some(focused) => { + if !layout.nodes.contains_key(focused) { + self.focused_id = default_focus_id_from_layout(layout); + } + } + None => { + self.focused_id = default_focus_id_from_layout(layout); + } + } + } + if self.focused_id.is_none() { + return; + } + + self.set_camera_target_to_focus(); + + let mut steps = match speed { + MoveSpeed::Step => 1, + MoveSpeed::Fast => 2, + }; + while steps > 0 { + let Some(next) = self.next_focus_id(direction) else { + break; + }; + self.focused_id = Some(next); + self.set_camera_target_to_focus(); + steps -= 1; + } + } + + pub fn focused_agent_id(&self) -> Option { + let layout = self.layout.as_ref()?; + let focused = self.focused_id.as_ref()?; + let node = layout.nodes.get(focused)?; + if node.kind == NodeKind::Agent { + Some(node.id.clone()) + } else { + None + } + } + + pub fn clear_layout(&mut self) { + self.layout = None; + self.camera = (0.0, 0.0); + self.camera_target = (0.0, 0.0); + self.camera_velocity = (0.0, 0.0); + } + + pub fn tick_camera(&mut self, dt_ms: i64) { + let (position, velocity) = animation::step_spring_f64( + self.camera, + self.camera_velocity, + self.camera_target, + dt_ms, + CANVAS_CAMERA_ACCEL, + CANVAS_CAMERA_FRICTION, + ); + self.camera = position; + self.camera_velocity = velocity; + + let dx = self.camera.0 - self.camera_target.0; + let dy = self.camera.1 - self.camera_target.1; + if dx.abs() <= CANVAS_CAMERA_SNAP_EPSILON && dy.abs() <= CANVAS_CAMERA_SNAP_EPSILON { + self.camera = self.camera_target; + self.camera_velocity = (0.0, 0.0); + } + } + + fn next_focus_id(&self, direction: Direction) -> Option { + let layout = self.layout.as_ref()?; + let focused_id = self.focused_id.as_ref()?; + let focused = layout.nodes.get(focused_id)?; + let (dir_x, dir_y) = direction_vector(direction); + let mut best: Option<(f64, String)> = None; + + for node in layout.nodes.values() { + if node.id == focused.id { + continue; + } + let dx = node.x - focused.x; + let dy = node.y - focused.y; + let dot = dx * dir_x + dy * dir_y; + if dot <= FOCUS_EPSILON { + continue; + } + let dist = dx * dx + dy * dy; + match &mut best { + Some((best_dist, best_id)) => { + if dist + FOCUS_EPSILON < *best_dist + || (dist - *best_dist).abs() <= FOCUS_EPSILON && node.id < *best_id + { + *best_dist = dist; + *best_id = node.id.clone(); + } + } + None => best = Some((dist, node.id.clone())), + } + } + + best.map(|(_, id)| id) + } + + fn set_camera_target_to_focus(&mut self) { + let Some((x, y)) = self.focus_position() else { + return; + }; + self.camera_target = (x, y); + } + + fn snap_camera_to_focus(&mut self) { + let Some((x, y)) = self.focus_position() else { + return; + }; + self.camera = (x, y); + self.camera_target = (x, y); + self.camera_velocity = (0.0, 0.0); + } + + fn focus_position(&self) -> Option<(f64, f64)> { + let Some(layout) = self.layout.as_ref() else { + return None; + }; + let Some(focused_id) = self.focused_id.as_ref() else { + return None; + }; + let Some(node) = layout.nodes.get(focused_id) else { + return None; + }; + Some((node.x, node.y)) + } +} + +struct OverlayTarget<'a> { + id: &'a str, + force_cluster: bool, +} + +pub struct RenderContext<'a> { + pub cluster_id: &'a str, + pub cluster_state: Option<&'a cluster::State>, + pub canvas_state: Option<&'a State>, + pub time_cursor: &'a TimeCursor, + pub anim_clock: &'a AnimClock, + pub pinned_target: Option<&'a FocusTarget>, +} + +pub fn render(frame: &mut Frame<'_>, area: Rect, context: RenderContext<'_>) { + let RenderContext { + cluster_id, + cluster_state, + canvas_state, + time_cursor, + anim_clock, + pinned_target, + } = context; + let block = Block::default() + .borders(Borders::ALL) + .title("Cluster Canvas"); + + let Some(cluster_state) = cluster_state else { + render_placeholder(frame, area, UNAVAILABLE_MESSAGE, None); + return; + }; + + if let Some(error) = cluster_state.topology_error.as_deref() { + render_placeholder(frame, area, UNAVAILABLE_MESSAGE, Some(error)); + return; + } + + let Some(topology) = cluster_state.topology.as_ref() else { + render_placeholder(frame, area, PENDING_MESSAGE, None); + return; + }; + + let fallback_layout; + let layout = match canvas_state.and_then(|state| state.layout.as_ref()) { + Some(layout) => layout, + None => { + fallback_layout = layout_for(topology); + &fallback_layout + } + }; + + let focused = canvas_state.and_then(|state| state.focused_id.as_deref()); + let pinned_highlight = match pinned_target { + Some(FocusTarget::Agent { + cluster_id: pinned_cluster_id, + agent_id, + }) if *pinned_cluster_id == cluster_id => Some(agent_id.as_str()), + _ => None, + }; + let pinned_overlay = + resolve_pinned_overlay(pinned_target, cluster_id, canvas_state, Some(cluster_state)); + let camera = canvas_state.map(|state| state.camera).unwrap_or((0.0, 0.0)); + let focus_pulse = animation::pulse_factor(anim_clock.phase) as f64; + + render_canvas( + frame, + CanvasRenderContext { + area, + cluster_id, + cluster_state, + topology, + layout, + focused, + pinned_highlight, + pinned_overlay, + camera, + block, + time_cursor, + focus_pulse, + }, + ); +} + +fn render_placeholder(frame: &mut Frame<'_>, area: Rect, headline: &str, detail: Option<&str>) { + let widget = calm_empty_state( + "Cluster Canvas", + headline, + detail, + Some("Press Esc to return to Fleet Radar"), + ); + frame.render_widget(widget, area); +} + +fn resolve_pinned_overlay<'a>( + pinned_target: Option<&'a FocusTarget>, + cluster_id: &str, + canvas_state: Option<&'a State>, + cluster_state: Option<&'a cluster::State>, +) -> Option> { + let pinned_target = pinned_target?; + match pinned_target { + FocusTarget::Agent { + cluster_id: pinned_cluster_id, + agent_id, + } if pinned_cluster_id == cluster_id => Some(OverlayTarget { + id: agent_id.as_str(), + force_cluster: false, + }), + FocusTarget::Cluster { id } if id == cluster_id => { + if let Some(focused_id) = canvas_state.and_then(|state| state.focused_id.as_deref()) { + return Some(OverlayTarget { + id: focused_id, + force_cluster: true, + }); + } + if let Some(layout) = canvas_state.and_then(|state| state.layout.as_ref()) { + if let Some(id) = layout.nodes.keys().next() { + return Some(OverlayTarget { + id: id.as_str(), + force_cluster: true, + }); + } + } + let topology = cluster_state.and_then(|state| state.topology.as_ref())?; + if let Some(agent) = topology.agents.first() { + return Some(OverlayTarget { + id: agent.id.as_str(), + force_cluster: true, + }); + } + topology.topics.first().map(|topic| OverlayTarget { + id: topic.as_str(), + force_cluster: true, + }) + } + _ => None, + } +} + +struct CanvasRenderContext<'a> { + area: Rect, + cluster_id: &'a str, + cluster_state: &'a cluster::State, + topology: &'a ClusterTopology, + layout: &'a LayoutCache, + focused: Option<&'a str>, + pinned_highlight: Option<&'a str>, + pinned_overlay: Option>, + camera: (f64, f64), + block: Block<'a>, + time_cursor: &'a TimeCursor, + focus_pulse: f64, +} + +fn render_canvas(frame: &mut Frame<'_>, canvas_ctx: CanvasRenderContext<'_>) { + let title = format!("Cluster Canvas {}", canvas_ctx.cluster_id); + let render_bounds = camera_bounds(canvas_ctx.layout, canvas_ctx.camera); + let block = canvas_ctx.block.title(title); + let canvas_inner = block.inner(canvas_ctx.area); + let layout = canvas_ctx.layout; + let focused = canvas_ctx.focused; + let pinned_highlight = canvas_ctx.pinned_highlight; + let topology = canvas_ctx.topology; + let focus_pulse = canvas_ctx.focus_pulse; + let canvas = Canvas::default() + .block(block) + .x_bounds([render_bounds.min_x, render_bounds.max_x]) + .y_bounds([render_bounds.min_y, render_bounds.max_y]) + .marker(Marker::Braille) + .paint(|ctx| { + for edge in &layout.edges { + let Some(from) = layout.nodes.get(&edge.from) else { + continue; + }; + let Some(to) = layout.nodes.get(&edge.to) else { + continue; + }; + ctx.draw(&CanvasLine { + x1: from.x, + y1: from.y, + x2: to.x, + y2: to.y, + color: theme::FG_DIM, + }); + } + + for node in layout.nodes.values() { + let color = match node.kind { + NodeKind::Agent => theme::agent_color(node.id.as_str()), + NodeKind::Topic => theme::FG_MUTED, + }; + let orb_radius = match node.kind { + NodeKind::Agent => AGENT_ORB_RADIUS, + NodeKind::Topic => TOPIC_ORB_RADIUS, + }; + + if focused == Some(node.id.as_str()) { + ctx.draw(&Circle { + x: node.x, + y: node.y, + radius: orb_radius + 0.8 + 0.25 * focus_pulse, + color: theme::ACCENT, + }); + } + + if pinned_highlight == Some(node.id.as_str()) && focused != Some(node.id.as_str()) { + ctx.draw(&Circle { + x: node.x, + y: node.y, + radius: orb_radius + 1.2, + color: theme::ACCENT2, + }); + } + + ctx.draw(&Circle { + x: node.x, + y: node.y, + radius: orb_radius, + color, + }); + + ctx.draw(&Points { + coords: &[(node.x, node.y)], + color, + }); + + let (label_x, label_y) = label_position(node.x, node.y, node.id.as_str()); + let label_style = Style::default().fg(color); + let label = node.label.clone(); + let line = Line::from(Span::styled(label, label_style)); + ctx.print(label_x, label_y, line); + } + + let summary_line = topology_summary(topology); + if let Some(summary_line) = summary_line { + let line = Line::from(Span::styled(summary_line, theme::dim_style())); + ctx.print(render_bounds.min_x + 1.0, render_bounds.max_y - 1.0, line); + } + }); + + frame.render_widget(canvas, canvas_ctx.area); + let overlay_layout = StreamOverlayLayout { + area: canvas_inner, + layout: canvas_ctx.layout, + focused: canvas_ctx.focused, + pinned_overlay: canvas_ctx.pinned_overlay, + render_bounds: &render_bounds, + spine_area: None, + }; + render_stream_overlay( + frame, + overlay_layout, + canvas_ctx.cluster_state, + canvas_ctx.time_cursor, + ); +} + +struct StreamOverlayLayout<'a> { + area: Rect, + layout: &'a LayoutCache, + focused: Option<&'a str>, + pinned_overlay: Option>, + render_bounds: &'a LayoutBounds, + spine_area: Option, +} + +fn render_stream_overlay( + frame: &mut Frame<'_>, + layout_ctx: StreamOverlayLayout<'_>, + cluster_state: &cluster::State, + time_cursor: &TimeCursor, +) { + if layout_ctx.area.width < 6 || layout_ctx.area.height < 4 { + return; + } + + if let Some(focused_id) = layout_ctx.focused { + if let Some(node) = layout_ctx.layout.nodes.get(focused_id) { + render_overlay_for_node( + frame, + &layout_ctx, + node, + cluster_state, + time_cursor, + theme::focus_border_style(), + false, + ); + } + } + + if let Some(pinned) = layout_ctx.pinned_overlay.as_ref() { + if Some(pinned.id) != layout_ctx.focused || pinned.force_cluster { + if let Some(node) = layout_ctx.layout.nodes.get(pinned.id) { + let style = Style::default() + .fg(theme::ACCENT2) + .add_modifier(Modifier::BOLD); + render_overlay_for_node( + frame, + &layout_ctx, + node, + cluster_state, + time_cursor, + style, + pinned.force_cluster, + ); + } + } + } +} + +fn render_overlay_for_node( + frame: &mut Frame<'_>, + layout_ctx: &StreamOverlayLayout<'_>, + node: &NodeLayout, + cluster_state: &cluster::State, + time_cursor: &TimeCursor, + border_style: Style, + force_cluster: bool, +) { + let focus_point = world_to_screen(layout_ctx.area, layout_ctx.render_bounds, node.x, node.y); + let overlay_size = overlay_dimensions(layout_ctx.area); + if overlay_size.0 == 0 || overlay_size.1 == 0 { + return; + } + + let overlay_rect = overlay_rect_near_focus( + layout_ctx.area, + focus_point, + overlay_size, + layout_ctx.spine_area, + ); + let inner = Block::default().borders(Borders::ALL).inner(overlay_rect); + let max_lines = inner.height as usize; + if max_lines == 0 { + return; + } + + let (title, lines) = + build_overlay_lines(cluster_state, node, time_cursor, max_lines, force_cluster); + let overlay = StreamOverlay::new(title, lines) + .placeholder_lines(stream::log_placeholder_lines( + stream::LogPlaceholderContext::Overlay, + )) + .border_style(border_style); + frame.render_widget(overlay, overlay_rect); +} + +fn build_overlay_lines<'a>( + cluster_state: &'a cluster::State, + node: &NodeLayout, + time_cursor: &TimeCursor, + max_lines: usize, + force_cluster: bool, +) -> (Line<'a>, Vec>) { + let is_agent = if force_cluster { + false + } else { + node.kind == NodeKind::Agent + }; + let log_title = if is_agent { + stream::overlay_title(format!("Logs - agent {}", node.id), time_cursor) + } else { + stream::overlay_title("Logs - cluster", time_cursor) + }; + let timeline_title = if is_agent { + stream::overlay_title(format!("Timeline - agent {}", node.id), time_cursor) + } else { + stream::overlay_title("Timeline - cluster", time_cursor) + }; + + let log_lines = collect_log_lines( + cluster_state, + time_cursor, + is_agent.then_some(node.id.as_str()), + max_lines, + ); + if !log_lines.is_empty() { + return (log_title, log_lines); + } + + let timeline_lines = collect_timeline_lines(cluster_state, time_cursor, max_lines); + if !timeline_lines.is_empty() { + return (timeline_title, timeline_lines); + } + + (log_title, Vec::new()) +} + +fn collect_log_lines<'a>( + cluster_state: &'a cluster::State, + time_cursor: &TimeCursor, + agent_id: Option<&str>, + max_lines: usize, +) -> Vec> { + if max_lines == 0 { + return Vec::new(); + } + let collected = + stream::select_time_window(&cluster_state.logs_time, time_cursor, max_lines, |line| { + if let Some(agent_id) = agent_id { + line.agent.as_deref() == Some(agent_id) || line.sender.as_deref() == Some(agent_id) + } else { + true + } + }); + collected + .into_iter() + .map(stream::format_log_line_styled) + .collect() +} + +fn collect_timeline_lines<'a>( + cluster_state: &'a cluster::State, + time_cursor: &TimeCursor, + max_lines: usize, +) -> Vec> { + if max_lines == 0 { + return Vec::new(); + } + let collected = + stream::select_time_window(&cluster_state.timeline_time, time_cursor, max_lines, |_| { + true + }); + collected + .into_iter() + .map(stream::format_timeline_event_styled) + .collect() +} + +fn overlay_dimensions(area: Rect) -> (u16, u16) { + if area.width < 6 || area.height < 4 { + return (0, 0); + } + let max_width = area.width.saturating_sub(2); + let max_height = area.height.saturating_sub(2); + if max_width == 0 || max_height == 0 { + return (0, 0); + } + + let mut width = ((area.width as f32) * 0.45).round() as u16; + let mut height = ((area.height as f32) * 0.35).round() as u16; + width = width.clamp(18, 52).min(max_width); + height = height.clamp(5, 12).min(max_height); + + if width == 0 || height == 0 { + return (0, 0); + } + (width, height) +} + +fn world_to_screen( + area: Rect, + render_bounds: &LayoutBounds, + world_x: f64, + world_y: f64, +) -> (u16, u16) { + if area.width == 0 || area.height == 0 { + return (area.x, area.y); + } + let width = (render_bounds.max_x - render_bounds.min_x).max(1.0); + let height = (render_bounds.max_y - render_bounds.min_y).max(1.0); + let mut rel_x = (world_x - render_bounds.min_x) / width; + let mut rel_y = (render_bounds.max_y - world_y) / height; + rel_x = rel_x.clamp(0.0, 1.0); + rel_y = rel_y.clamp(0.0, 1.0); + let x = area.x as f64 + rel_x * ((area.width - 1) as f64); + let y = area.y as f64 + rel_y * ((area.height - 1) as f64); + (x.round() as u16, y.round() as u16) +} + +fn overlay_rect_near_focus( + bounds: Rect, + focus: (u16, u16), + size: (u16, u16), + spine: Option, +) -> Rect { + let width = size.0.min(bounds.width); + let height = size.1.min(bounds.height); + let candidates = [(true, true), (true, false), (false, true), (false, false)]; + + for (right, down) in candidates { + let x = if right { + focus.0.saturating_add(1) + } else { + focus.0.saturating_sub(width.saturating_add(1)) + }; + let y = if down { + focus.1.saturating_add(1) + } else { + focus.1.saturating_sub(height.saturating_add(1)) + }; + let rect = clamp_rect_to_bounds( + Rect { + x, + y, + width, + height, + }, + bounds, + ); + let rect = avoid_spine(rect, bounds, spine); + if !rect_intersects_spine(rect, spine) { + return rect; + } + } + + let rect = clamp_rect_to_bounds( + Rect { + x: bounds.x, + y: bounds.y, + width, + height, + }, + bounds, + ); + avoid_spine(rect, bounds, spine) +} + +fn clamp_rect_to_bounds(rect: Rect, bounds: Rect) -> Rect { + let width = rect.width.min(bounds.width); + let height = rect.height.min(bounds.height); + if bounds.width == 0 || bounds.height == 0 || width == 0 || height == 0 { + return Rect { + x: bounds.x, + y: bounds.y, + width, + height, + }; + } + + let max_x = bounds.x.saturating_add(bounds.width.saturating_sub(width)); + let max_y = bounds + .y + .saturating_add(bounds.height.saturating_sub(height)); + let mut x = rect.x; + let mut y = rect.y; + if x < bounds.x { + x = bounds.x; + } else if x > max_x { + x = max_x; + } + if y < bounds.y { + y = bounds.y; + } else if y > max_y { + y = max_y; + } + + Rect { + x, + y, + width, + height, + } +} + +fn rect_intersects_spine(rect: Rect, spine: Option) -> bool { + spine.is_some_and(|spine| rects_intersect(rect, spine)) +} + +fn rects_intersect(a: Rect, b: Rect) -> bool { + let a_right = a.x.saturating_add(a.width); + let a_bottom = a.y.saturating_add(a.height); + let b_right = b.x.saturating_add(b.width); + let b_bottom = b.y.saturating_add(b.height); + a.x < b_right && a_right > b.x && a.y < b_bottom && a_bottom > b.y +} + +fn avoid_spine(rect: Rect, bounds: Rect, spine: Option) -> Rect { + let Some(spine) = spine else { + return rect; + }; + if !rects_intersect(rect, spine) { + return rect; + } + + let options = vec![ + Rect { + x: rect.x, + y: spine.y.saturating_sub(rect.height.saturating_add(1)), + width: rect.width, + height: rect.height, + }, + Rect { + x: rect.x, + y: spine.y.saturating_add(spine.height).saturating_add(1), + width: rect.width, + height: rect.height, + }, + Rect { + x: spine.x.saturating_sub(rect.width.saturating_add(1)), + y: rect.y, + width: rect.width, + height: rect.height, + }, + Rect { + x: spine.x.saturating_add(spine.width).saturating_add(1), + y: rect.y, + width: rect.width, + height: rect.height, + }, + ]; + + for option in options { + let candidate = clamp_rect_to_bounds(option, bounds); + if !rects_intersect(candidate, spine) { + return candidate; + } + } + + rect +} + +fn topology_summary(topology: &ClusterTopology) -> Option { + if topology.agents.is_empty() && topology.topics.is_empty() && topology.edges.is_empty() { + return None; + } + Some(format!( + "{} agents, {} topics, {} edges", + topology.agents.len(), + topology.topics.len(), + topology.edges.len() + )) +} + +fn camera_bounds(layout: &LayoutCache, camera: (f64, f64)) -> LayoutBounds { + let width = layout.bounds.max_x - layout.bounds.min_x; + let height = layout.bounds.max_y - layout.bounds.min_y; + let half_w = width / 2.0; + let half_h = height / 2.0; + LayoutBounds { + min_x: camera.0 - half_w, + max_x: camera.0 + half_w, + min_y: camera.1 - half_h, + max_y: camera.1 + half_h, + } +} + +fn direction_vector(direction: Direction) -> (f64, f64) { + match direction { + Direction::Left => (-1.0, 0.0), + Direction::Right => (1.0, 0.0), + Direction::Up => (0.0, 1.0), + Direction::Down => (0.0, -1.0), + } +} + +fn default_focus_id(topology: &ClusterTopology) -> Option { + let mut agent_ids: Vec<&String> = topology.agents.iter().map(|agent| &agent.id).collect(); + agent_ids.sort(); + if let Some(id) = agent_ids.first() { + return Some((*id).clone()); + } + let mut topics: Vec<&String> = topology.topics.iter().collect(); + topics.sort(); + topics.first().map(|id| (*id).clone()) +} + +fn default_focus_id_from_layout(layout: &LayoutCache) -> Option { + let mut agent_ids: Vec<&String> = layout + .nodes + .values() + .filter(|node| node.kind == NodeKind::Agent) + .map(|node| &node.id) + .collect(); + agent_ids.sort(); + if let Some(id) = agent_ids.first() { + return Some((*id).clone()); + } + let mut topic_ids: Vec<&String> = layout + .nodes + .values() + .filter(|node| node.kind == NodeKind::Topic) + .map(|node| &node.id) + .collect(); + topic_ids.sort(); + topic_ids.first().map(|id| (*id).clone()) +} + +fn layout_for(topology: &ClusterTopology) -> LayoutCache { + let mut nodes = HashMap::new(); + + for agent in &topology.agents { + let (x, y) = ring_position(&agent.id, AGENT_RING_RADIUS); + let label = truncate_label(&agent_label(agent)); + nodes.insert( + agent.id.clone(), + NodeLayout { + id: agent.id.clone(), + label, + x, + y, + kind: NodeKind::Agent, + }, + ); + } + + for topic in &topology.topics { + let (x, y) = ring_position(topic, TOPIC_RING_RADIUS); + let label = truncate_label(topic); + nodes.insert( + topic.clone(), + NodeLayout { + id: topic.clone(), + label, + x, + y, + kind: NodeKind::Topic, + }, + ); + } + + let mut edges = Vec::new(); + for edge in &topology.edges { + if nodes.contains_key(&edge.from) && nodes.contains_key(&edge.to) { + edges.push(LayoutEdge { + from: edge.from.clone(), + to: edge.to.clone(), + }); + } + } + + LayoutCache { + nodes, + edges, + bounds: LayoutBounds { + min_x: -WORLD_RADIUS, + max_x: WORLD_RADIUS, + min_y: -WORLD_RADIUS, + max_y: WORLD_RADIUS, + }, + } +} + +fn ring_position(id: &str, radius: f64) -> (f64, f64) { + let angle = stable_angle(id); + let jitter = jitter_offset(id); + let r = (radius + jitter).max(4.0); + (r * angle.cos(), r * angle.sin()) +} + +fn jitter_offset(id: &str) -> f64 { + let hash = stable_hash(id); + let step = (hash % 7) as f64 - 3.0; + step * 0.35 +} + +fn agent_label(agent: &TopologyAgent) -> String { + match agent.role.as_ref() { + Some(role) if !role.trim().is_empty() => format!("{} ({})", agent.id, role), + _ => agent.id.clone(), + } +} + +fn label_position(x: f64, y: f64, id: &str) -> (f64, f64) { + let hash = stable_hash(id); + let side = if hash & 1 == 0 { 1.0 } else { -1.0 }; + let angle = if x == 0.0 && y == 0.0 { + 0.0 + } else { + y.atan2(x) + }; + let tangent_x = -angle.sin(); + let tangent_y = angle.cos(); + let radial_x = angle.cos(); + let radial_y = angle.sin(); + let mut lx = x + tangent_x * LABEL_OFFSET * side + radial_x * LABEL_RADIAL_OFFSET; + let mut ly = y + tangent_y * LABEL_OFFSET * side + radial_y * LABEL_RADIAL_OFFSET; + let min = -WORLD_RADIUS + 1.0; + let max = WORLD_RADIUS - 1.0; + if lx < min { + lx = min; + } else if lx > max { + lx = max; + } + if ly < min { + ly = min; + } else if ly > max { + ly = max; + } + (lx, ly) +} + +fn truncate_label(label: &str) -> String { + let mut iter = label.chars(); + let mut out = String::new(); + for _ in 0..LABEL_LIMIT { + match iter.next() { + Some(ch) => out.push(ch), + None => return label.to_string(), + } + } + if iter.next().is_some() { + out.push_str(".."); + } + out +} + +fn stable_angle(input: &str) -> f64 { + let hash = stable_hash(input); + let fraction = (hash % 3600) as f64 / 3600.0; + std::f64::consts::TAU * fraction +} + +fn stable_hash(input: &str) -> u64 { + const FNV_OFFSET: u64 = 0xcbf29ce484222325; + const FNV_PRIME: u64 = 0x00000100000001b3; + let mut hash = FNV_OFFSET; + for byte in input.as_bytes() { + hash ^= *byte as u64; + hash = hash.wrapping_mul(FNV_PRIME); + } + hash +} + +#[cfg(test)] +mod tests { + use super::*; + use ratatui::backend::TestBackend; + use ratatui::buffer::Buffer; + use ratatui::Terminal; + + use crate::app::animation::AnimClock; + use crate::protocol::{ClusterLogLine, TopologyAgent, TopologyEdge, TopologyEdgeKind}; + use crate::ui::widgets::test_utils::line_text; + use std::collections::HashMap; + + fn sample_topology() -> ClusterTopology { + ClusterTopology { + agents: vec![ + TopologyAgent { + id: "agent-alpha".to_string(), + role: Some("planner".to_string()), + }, + TopologyAgent { + id: "agent-bravo".to_string(), + role: Some("worker".to_string()), + }, + TopologyAgent { + id: "agent-charlie".to_string(), + role: None, + }, + ], + topics: vec!["ISSUE_OPENED".to_string()], + edges: vec![TopologyEdge { + from: "agent-alpha".to_string(), + to: "agent-bravo".to_string(), + topic: "ISSUE_OPENED".to_string(), + kind: TopologyEdgeKind::Publish, + dynamic: None, + }], + } + } + + fn buffer_contains(buffer: &Buffer, needle: &str) -> bool { + for y in 0..buffer.area.height { + if line_text(buffer, y).contains(needle) { + return true; + } + } + false + } + + fn buffers_differ(first: &Buffer, second: &Buffer) -> bool { + if first.area != second.area { + return true; + } + for y in first.area.top()..first.area.bottom() { + for x in first.area.left()..first.area.right() { + let first_cell = first.cell((x, y)); + let second_cell = second.cell((x, y)); + if first_cell != second_cell { + return true; + } + } + } + false + } + + fn rect_within_bounds(rect: Rect, bounds: Rect) -> bool { + let rect_right = rect.x.saturating_add(rect.width); + let rect_bottom = rect.y.saturating_add(rect.height); + let bounds_right = bounds.x.saturating_add(bounds.width); + let bounds_bottom = bounds.y.saturating_add(bounds.height); + rect.x >= bounds.x + && rect.y >= bounds.y + && rect_right <= bounds_right + && rect_bottom <= bounds_bottom + } + + fn layout_with_nodes(nodes: Vec) -> LayoutCache { + let mut map = HashMap::new(); + for node in nodes { + map.insert(node.id.clone(), node); + } + LayoutCache { + nodes: map, + edges: Vec::new(), + bounds: LayoutBounds { + min_x: -WORLD_RADIUS, + max_x: WORLD_RADIUS, + min_y: -WORLD_RADIUS, + max_y: WORLD_RADIUS, + }, + } + } + + #[test] + fn layout_is_deterministic() { + let topology = sample_topology(); + let first = layout_for(&topology); + let second = layout_for(&topology); + assert_eq!(first, second); + } + + #[test] + fn render_pending_topology() { + let backend = TestBackend::new(60, 12); + let mut terminal = Terminal::new(backend).expect("terminal"); + let cluster_state = cluster::State::default(); + let canvas_state = State::default(); + let anim_clock = AnimClock::default(); + let time_cursor = TimeCursor::default(); + + terminal + .draw(|frame| { + let area = frame.area(); + render( + frame, + area, + RenderContext { + cluster_id: "cluster-1", + cluster_state: Some(&cluster_state), + canvas_state: Some(&canvas_state), + time_cursor: &time_cursor, + anim_clock: &anim_clock, + pinned_target: None, + }, + ); + }) + .expect("draw"); + + let buffer = terminal.backend().buffer(); + assert!(buffer_contains(buffer, PENDING_MESSAGE)); + } + + #[test] + fn render_missing_cluster_as_unavailable() { + let backend = TestBackend::new(60, 12); + let mut terminal = Terminal::new(backend).expect("terminal"); + let canvas_state = State::default(); + let anim_clock = AnimClock::default(); + let time_cursor = TimeCursor::default(); + + terminal + .draw(|frame| { + let area = frame.area(); + render( + frame, + area, + RenderContext { + cluster_id: "cluster-missing", + cluster_state: None, + canvas_state: Some(&canvas_state), + time_cursor: &time_cursor, + anim_clock: &anim_clock, + pinned_target: None, + }, + ); + }) + .expect("draw"); + + let buffer = terminal.backend().buffer(); + assert!(buffer_contains(buffer, UNAVAILABLE_MESSAGE)); + } + + #[test] + fn render_topology_unavailable_on_error() { + let backend = TestBackend::new(60, 12); + let mut terminal = Terminal::new(backend).expect("terminal"); + let mut cluster_state = cluster::State::default(); + cluster_state.topology_error = Some("backend timeout".to_string()); + let canvas_state = State::default(); + let anim_clock = AnimClock::default(); + let time_cursor = TimeCursor::default(); + + terminal + .draw(|frame| { + let area = frame.area(); + render( + frame, + area, + RenderContext { + cluster_id: "cluster-2", + cluster_state: Some(&cluster_state), + canvas_state: Some(&canvas_state), + time_cursor: &time_cursor, + anim_clock: &anim_clock, + pinned_target: None, + }, + ); + }) + .expect("draw"); + + let buffer = terminal.backend().buffer(); + assert!(buffer_contains(buffer, UNAVAILABLE_MESSAGE)); + assert!(buffer_contains(buffer, "backend timeout")); + } + + #[test] + fn render_basic_topology() { + let backend = TestBackend::new(80, 20); + let mut terminal = Terminal::new(backend).expect("terminal"); + let topology = sample_topology(); + let mut cluster_state = cluster::State::default(); + cluster_state.topology = Some(topology.clone()); + let mut canvas_state = State::default(); + canvas_state.update_layout(&topology); + let anim_clock = AnimClock::default(); + let time_cursor = TimeCursor::default(); + + terminal + .draw(|frame| { + let area = frame.area(); + render( + frame, + area, + RenderContext { + cluster_id: "cluster-3", + cluster_state: Some(&cluster_state), + canvas_state: Some(&canvas_state), + time_cursor: &time_cursor, + anim_clock: &anim_clock, + pinned_target: None, + }, + ); + }) + .expect("draw"); + + let buffer = terminal.backend().buffer(); + assert!(buffer_contains(buffer, "agent-alpha")); + assert!(buffer_contains(buffer, "ISSUE_OPENED")); + } + + #[test] + fn default_focus_prefers_agents() { + let topology = ClusterTopology { + agents: vec![ + TopologyAgent { + id: "worker".to_string(), + role: None, + }, + TopologyAgent { + id: "planner".to_string(), + role: None, + }, + ], + topics: vec!["topic-b".to_string(), "topic-a".to_string()], + edges: Vec::new(), + }; + assert_eq!(default_focus_id(&topology), Some("planner".to_string())); + + let topology = ClusterTopology { + agents: Vec::new(), + topics: vec!["topic-b".to_string(), "topic-a".to_string()], + edges: Vec::new(), + }; + assert_eq!(default_focus_id(&topology), Some("topic-a".to_string())); + } + + #[test] + fn move_focus_direction() { + let layout = layout_with_nodes(vec![ + NodeLayout { + id: "center".to_string(), + label: "center".to_string(), + x: 0.0, + y: 0.0, + kind: NodeKind::Agent, + }, + NodeLayout { + id: "right".to_string(), + label: "right".to_string(), + x: 10.0, + y: 0.0, + kind: NodeKind::Agent, + }, + NodeLayout { + id: "right-far".to_string(), + label: "right-far".to_string(), + x: 18.0, + y: 2.0, + kind: NodeKind::Agent, + }, + NodeLayout { + id: "left".to_string(), + label: "left".to_string(), + x: -10.0, + y: 0.0, + kind: NodeKind::Agent, + }, + NodeLayout { + id: "up".to_string(), + label: "up".to_string(), + x: 0.0, + y: 10.0, + kind: NodeKind::Agent, + }, + NodeLayout { + id: "down".to_string(), + label: "down".to_string(), + x: 0.0, + y: -10.0, + kind: NodeKind::Agent, + }, + ]); + + let mut state = State { + focused_id: Some("center".to_string()), + layout: Some(layout), + ..State::default() + }; + state.move_focus(Direction::Right, MoveSpeed::Step); + assert_eq!(state.focused_id.as_deref(), Some("right")); + } + + #[test] + fn move_focus_fast() { + let layout = layout_with_nodes(vec![ + NodeLayout { + id: "a".to_string(), + label: "a".to_string(), + x: 0.0, + y: 0.0, + kind: NodeKind::Agent, + }, + NodeLayout { + id: "b".to_string(), + label: "b".to_string(), + x: 10.0, + y: 0.0, + kind: NodeKind::Agent, + }, + NodeLayout { + id: "c".to_string(), + label: "c".to_string(), + x: 20.0, + y: 0.0, + kind: NodeKind::Agent, + }, + ]); + + let mut state = State { + focused_id: Some("a".to_string()), + layout: Some(layout), + ..State::default() + }; + state.move_focus(Direction::Right, MoveSpeed::Fast); + assert_eq!(state.focused_id.as_deref(), Some("c")); + } + + #[test] + fn render_focus_ring_changes_output() { + let backend = TestBackend::new(80, 20); + let mut terminal = Terminal::new(backend).expect("terminal"); + let topology = sample_topology(); + let layout = layout_for(&topology); + let mut cluster_state = cluster::State::default(); + cluster_state.topology = Some(topology.clone()); + let anim_clock = AnimClock::default(); + let time_cursor = TimeCursor::default(); + + let canvas_state = State { + focused_id: Some("agent-alpha".to_string()), + layout: Some(layout.clone()), + ..State::default() + }; + terminal + .draw(|frame| { + let area = frame.area(); + render( + frame, + area, + RenderContext { + cluster_id: "cluster-4", + cluster_state: Some(&cluster_state), + canvas_state: Some(&canvas_state), + time_cursor: &time_cursor, + anim_clock: &anim_clock, + pinned_target: None, + }, + ); + }) + .expect("draw"); + let first = terminal.backend().buffer().clone(); + + let backend = TestBackend::new(80, 20); + let mut terminal = Terminal::new(backend).expect("terminal"); + let canvas_state = State { + focused_id: Some("agent-bravo".to_string()), + layout: Some(layout), + ..State::default() + }; + terminal + .draw(|frame| { + let area = frame.area(); + render( + frame, + area, + RenderContext { + cluster_id: "cluster-4", + cluster_state: Some(&cluster_state), + canvas_state: Some(&canvas_state), + time_cursor: &time_cursor, + anim_clock: &anim_clock, + pinned_target: None, + }, + ); + }) + .expect("draw"); + let second = terminal.backend().buffer().clone(); + + assert!(buffers_differ(&first, &second)); + } + + #[test] + fn overlay_rect_clamps_to_bounds() { + let bounds = Rect { + x: 2, + y: 1, + width: 40, + height: 18, + }; + let focus = (41, 18); + let size = (26, 10); + let rect = overlay_rect_near_focus(bounds, focus, size, None); + assert!(rect_within_bounds(rect, bounds)); + } + + #[test] + fn overlay_rect_avoids_spine() { + let bounds = Rect { + x: 0, + y: 0, + width: 60, + height: 20, + }; + let spine = Rect { + x: 22, + y: 12, + width: 12, + height: 4, + }; + let focus = (26, 13); + let rect = overlay_rect_near_focus(bounds, focus, (20, 8), Some(spine)); + assert!(!rects_intersect(rect, spine)); + } + + #[test] + fn cluster_canvas_renders_log_overlay() { + let backend = TestBackend::new(90, 24); + let mut terminal = Terminal::new(backend).expect("terminal"); + let topology = sample_topology(); + let mut cluster_state = cluster::State::default(); + cluster_state.topology = Some(topology.clone()); + cluster_state.push_log_lines( + vec![ClusterLogLine { + id: "line-1".to_string(), + timestamp: 0, + text: "build complete".to_string(), + agent: Some("agent-alpha".to_string()), + role: None, + sender: None, + }], + None, + ); + + let mut canvas_state = State::default(); + canvas_state.update_layout(&topology); + canvas_state.focused_id = Some("agent-alpha".to_string()); + let anim_clock = AnimClock::default(); + let time_cursor = TimeCursor::default(); + + terminal + .draw(|frame| { + let area = frame.area(); + render( + frame, + area, + RenderContext { + cluster_id: "cluster-5", + cluster_state: Some(&cluster_state), + canvas_state: Some(&canvas_state), + time_cursor: &time_cursor, + anim_clock: &anim_clock, + pinned_target: None, + }, + ); + }) + .expect("draw"); + + let buffer = terminal.backend().buffer(); + assert!(buffer_contains(buffer, "Logs - agent agent-alpha")); + assert!(buffer_contains(buffer, "build complete")); + } + + #[test] + fn cluster_canvas_overlay_respects_time_window() { + let backend = TestBackend::new(90, 24); + let mut terminal = Terminal::new(backend).expect("terminal"); + let topology = sample_topology(); + let mut cluster_state = cluster::State::default(); + cluster_state.topology = Some(topology.clone()); + cluster_state.push_log_lines( + vec![ + ClusterLogLine { + id: "line-old".to_string(), + timestamp: 100, + text: "old".to_string(), + agent: Some("agent-alpha".to_string()), + role: None, + sender: None, + }, + ClusterLogLine { + id: "line-mid".to_string(), + timestamp: 200, + text: "mid".to_string(), + agent: Some("agent-alpha".to_string()), + role: None, + sender: None, + }, + ClusterLogLine { + id: "line-new".to_string(), + timestamp: 300, + text: "new".to_string(), + agent: Some("agent-alpha".to_string()), + role: None, + sender: None, + }, + ], + None, + ); + + let mut canvas_state = State::default(); + canvas_state.update_layout(&topology); + canvas_state.focused_id = Some("agent-alpha".to_string()); + + let time_cursor = TimeCursor { + mode: crate::app::TimeCursorMode::Scrub, + t_ms: 250, + window_ms: 120, + }; + let anim_clock = AnimClock::default(); + + terminal + .draw(|frame| { + let area = frame.area(); + render( + frame, + area, + RenderContext { + cluster_id: "cluster-5", + cluster_state: Some(&cluster_state), + canvas_state: Some(&canvas_state), + time_cursor: &time_cursor, + anim_clock: &anim_clock, + pinned_target: None, + }, + ); + }) + .expect("draw"); + + let buffer = terminal.backend().buffer(); + assert!(buffer_contains(buffer, "mid")); + assert!(!buffer_contains(buffer, "old")); + assert!(!buffer_contains(buffer, "new")); + } + + #[test] + fn cluster_canvas_renders_overlay_placeholder() { + let backend = TestBackend::new(80, 20); + let mut terminal = Terminal::new(backend).expect("terminal"); + let topology = sample_topology(); + let mut cluster_state = cluster::State::default(); + cluster_state.topology = Some(topology.clone()); + + let mut canvas_state = State::default(); + canvas_state.update_layout(&topology); + canvas_state.focused_id = Some("agent-alpha".to_string()); + let anim_clock = AnimClock::default(); + let time_cursor = TimeCursor::default(); + + terminal + .draw(|frame| { + let area = frame.area(); + render( + frame, + area, + RenderContext { + cluster_id: "cluster-6", + cluster_state: Some(&cluster_state), + canvas_state: Some(&canvas_state), + time_cursor: &time_cursor, + anim_clock: &anim_clock, + pinned_target: None, + }, + ); + }) + .expect("draw"); + + let buffer = terminal.backend().buffer(); + assert!(buffer_contains(buffer, "No logs yet.")); + } + + #[test] + fn truncate_label_handles_unicode() { + let label = "Ī±Ī²Ī³Ī“ĪµĪ¶Ī·ĪøĪ¹ĪŗĪ»Ī¼Ī½Ī¾ĪæĻ€ĻĻƒ"; + let prefix: String = label.chars().take(LABEL_LIMIT).collect(); + let truncated = truncate_label(label); + assert_eq!(truncated, format!("{}..", prefix)); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/screens/launcher.rs b/tui-rs/crates/zeroshot-tui/src/screens/launcher.rs new file mode 100644 index 00000000..7aa3351c --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/screens/launcher.rs @@ -0,0 +1,97 @@ +#[derive(Debug, Clone, Default)] +pub struct State { + pub input: String, + pub cursor: usize, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum Action { + Submit, + InsertChar(char), + Backspace, + Delete, + MoveCursorLeft, + MoveCursorRight, + MoveCursorHome, + MoveCursorEnd, +} + +impl State { + pub fn insert_char(&mut self, ch: char) { + let idx = self.byte_index(self.cursor); + self.input.insert(idx, ch); + self.cursor = self.cursor.saturating_add(1); + } + + pub fn backspace(&mut self) { + if self.cursor == 0 { + return; + } + let start = self.byte_index(self.cursor - 1); + let end = self.byte_index(self.cursor); + if start < end { + self.input.replace_range(start..end, ""); + self.cursor = self.cursor.saturating_sub(1); + } + } + + pub fn delete(&mut self) { + let len = self.len_chars(); + if self.cursor >= len { + return; + } + let start = self.byte_index(self.cursor); + let end = self.byte_index(self.cursor + 1); + if start < end { + self.input.replace_range(start..end, ""); + } + } + + pub fn move_left(&mut self) { + if self.cursor > 0 { + self.cursor -= 1; + } + } + + pub fn move_right(&mut self) { + let len = self.len_chars(); + if self.cursor < len { + self.cursor += 1; + } + } + + pub fn move_home(&mut self) { + self.cursor = 0; + } + + pub fn move_end(&mut self) { + self.cursor = self.len_chars(); + } + + pub fn clear(&mut self) { + self.input.clear(); + self.cursor = 0; + } + + pub fn clamp_cursor(&mut self) { + let len = self.len_chars(); + if self.cursor > len { + self.cursor = len; + } + } + + fn len_chars(&self) -> usize { + self.input.chars().count() + } + + fn byte_index(&self, char_index: usize) -> usize { + if char_index == 0 { + return 0; + } + self.input + .char_indices() + .nth(char_index) + .map(|(idx, _)| idx) + .unwrap_or_else(|| self.input.len()) + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/screens/metrics.rs b/tui-rs/crates/zeroshot-tui/src/screens/metrics.rs new file mode 100644 index 00000000..e7f18cdb --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/screens/metrics.rs @@ -0,0 +1,90 @@ +use crate::protocol::ClusterMetrics; + +pub const CPU_COLUMN_WIDTH: usize = 6; +pub const MEM_COLUMN_WIDTH: usize = 8; +const PLACEHOLDER: &str = "-"; + +pub fn format_cpu_percent(metrics: Option<&ClusterMetrics>) -> String { + let Some(metrics) = metrics else { + return format_placeholder(CPU_COLUMN_WIDTH); + }; + if !metrics.supported { + return format_placeholder(CPU_COLUMN_WIDTH); + } + let Some(value) = metrics.cpu_percent else { + return format_placeholder(CPU_COLUMN_WIDTH); + }; + if !value.is_finite() { + return format_placeholder(CPU_COLUMN_WIDTH); + } + format!("{:>width$.1}%", value, width = CPU_COLUMN_WIDTH - 1) +} + +pub fn format_memory_mb(metrics: Option<&ClusterMetrics>) -> String { + let Some(metrics) = metrics else { + return format_placeholder(MEM_COLUMN_WIDTH); + }; + if !metrics.supported { + return format_placeholder(MEM_COLUMN_WIDTH); + } + let Some(value) = metrics.memory_mb else { + return format_placeholder(MEM_COLUMN_WIDTH); + }; + if !value.is_finite() { + return format_placeholder(MEM_COLUMN_WIDTH); + } + let rounded = value.round() as i64; + format!("{:>width$}MB", rounded, width = MEM_COLUMN_WIDTH - 2) +} + +pub fn format_metrics_line(metrics: Option<&ClusterMetrics>) -> String { + let cpu = format_cpu_percent(metrics); + let mem = format_memory_mb(metrics); + format!("CPU {cpu} | MEM {mem}") +} + +fn format_placeholder(width: usize) -> String { + format!("{:>width$}", PLACEHOLDER, width = width) +} + +#[cfg(test)] +mod tests { + use super::*; + + fn metrics(supported: bool, cpu: Option, mem: Option) -> ClusterMetrics { + ClusterMetrics { + id: "cluster-1".to_string(), + supported, + cpu_percent: cpu, + memory_mb: mem, + } + } + + #[test] + fn placeholder_when_unsupported() { + let sample = metrics(false, Some(12.3), Some(456.7)); + let cpu = format_cpu_percent(Some(&sample)); + assert_eq!(cpu.trim(), PLACEHOLDER); + assert_eq!(cpu.len(), CPU_COLUMN_WIDTH); + + let mem = format_memory_mb(Some(&sample)); + assert_eq!(mem.trim(), PLACEHOLDER); + assert_eq!(mem.len(), MEM_COLUMN_WIDTH); + } + + #[test] + fn cpu_rounds_to_one_decimal() { + let sample = metrics(true, Some(12.34), None); + assert_eq!(format_cpu_percent(Some(&sample)), " 12.3%"); + let sample = metrics(true, Some(12.36), None); + assert_eq!(format_cpu_percent(Some(&sample)), " 12.4%"); + } + + #[test] + fn memory_rounds_to_nearest_mb() { + let sample = metrics(true, None, Some(256.4)); + assert_eq!(format_memory_mb(Some(&sample)), " 256MB"); + let sample = metrics(true, None, Some(256.6)); + assert_eq!(format_memory_mb(Some(&sample)), " 257MB"); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/screens/mod.rs b/tui-rs/crates/zeroshot-tui/src/screens/mod.rs new file mode 100644 index 00000000..1a3edeea --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/screens/mod.rs @@ -0,0 +1,8 @@ +pub mod agent; +pub mod agent_microscope; +pub mod cluster; +pub mod cluster_canvas; +pub mod launcher; +pub mod metrics; +pub mod monitor; +pub mod radar; diff --git a/tui-rs/crates/zeroshot-tui/src/screens/monitor.rs b/tui-rs/crates/zeroshot-tui/src/screens/monitor.rs new file mode 100644 index 00000000..91d4891a --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/screens/monitor.rs @@ -0,0 +1,231 @@ +use std::collections::HashMap; + +use ratatui::layout::{Alignment, Constraint, Rect}; +use ratatui::style::Style; +use ratatui::text::{Line, Span}; +use ratatui::widgets::{Cell, Paragraph, Row, Table, TableState}; +use ratatui::Frame; + +use crate::protocol::{ClusterMetrics, ClusterSummary}; +use crate::screens::metrics; +use crate::ui::theme; + +const POLL_INTERVAL_MS: i64 = 1000; + +#[derive(Debug, Clone, Default)] +pub struct State { + pub clusters: Vec, + pub selected: usize, + pub last_poll_at: Option, + pub last_message_counts: HashMap, + pub last_activity_at: HashMap, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum Action { + MoveSelection(i32), + OpenSelected, +} + +impl State { + pub fn set_clusters(&mut self, clusters: Vec, now_ms: i64) { + let selected_id = self.selected_cluster_id(); + self.update_activity(&clusters, now_ms); + self.clusters = clusters; + self.reconcile_selection(selected_id); + } + + pub fn move_selection(&mut self, delta: i32) { + if self.clusters.is_empty() { + self.selected = 0; + return; + } + + let len = self.clusters.len() as i32; + let mut next = self.selected as i32 + delta; + if next < 0 { + next = 0; + } + if next >= len { + next = len - 1; + } + self.selected = next as usize; + } + + pub fn selected_cluster_id(&self) -> Option { + self.clusters + .get(self.selected) + .map(|cluster| cluster.id.clone()) + } + + pub fn poll_due(&self, now_ms: i64) -> bool { + match self.last_poll_at { + None => true, + Some(last) => now_ms.saturating_sub(last) >= POLL_INTERVAL_MS, + } + } + + pub fn mark_polled(&mut self, now_ms: i64) { + self.last_poll_at = Some(now_ms); + } + + fn update_activity(&mut self, clusters: &[ClusterSummary], now_ms: i64) { + let mut next_counts = HashMap::new(); + let mut next_activity = HashMap::new(); + + for cluster in clusters { + let prev_count = self.last_message_counts.get(&cluster.id).copied(); + let prev_activity = self.last_activity_at.get(&cluster.id).copied(); + let mut activity = prev_activity; + + if prev_count + .map(|prev| cluster.message_count > prev) + .unwrap_or(true) + { + activity = Some(now_ms); + } + + next_counts.insert(cluster.id.clone(), cluster.message_count); + if let Some(activity_at) = activity { + next_activity.insert(cluster.id.clone(), activity_at); + } + } + + self.last_message_counts = next_counts; + self.last_activity_at = next_activity; + } + + fn reconcile_selection(&mut self, selected_id: Option) { + if let Some(id) = selected_id { + if let Some(index) = self.clusters.iter().position(|cluster| cluster.id == id) { + self.selected = index; + return; + } + } + + if self.selected >= self.clusters.len() { + self.selected = self.clusters.len().saturating_sub(1); + } + } +} + +pub fn render( + frame: &mut Frame<'_>, + area: Rect, + state: &State, + metrics_map: &HashMap, + now_ms: i64, +) { + // Empty state + if state.clusters.is_empty() { + render_empty(frame, area); + return; + } + + let header = Row::new(vec![ + Cell::from("ID"), + Cell::from("STATE"), + Cell::from("PROVIDER"), + Cell::from("CPU%"), + Cell::from("MEM"), + Cell::from("DURATION"), + Cell::from("LAST"), + ]) + .style(theme::table_header_style()); + + let rows: Vec = state + .clusters + .iter() + .map(|cluster| { + let provider = cluster.provider.clone().unwrap_or_else(|| "-".to_string()); + let metrics = metrics_map.get(&cluster.id); + let cpu = metrics::format_cpu_percent(metrics); + let mem = metrics::format_memory_mb(metrics); + let duration = format_duration(now_ms.saturating_sub(cluster.created_at)); + let last_activity = state + .last_activity_at + .get(&cluster.id) + .map(|activity| format_duration(now_ms.saturating_sub(*activity))) + .unwrap_or_else(|| "-".to_string()); + + let state_style = theme::status_style(&cluster.state); + let is_done = matches!( + cluster.state.as_str(), + "done" | "completed" | "complete" | "stopped" + ); + let row_style = if is_done { + theme::done_row_style() + } else { + Style::default() + }; + + Row::new(vec![ + Cell::from(cluster.id.clone()), + Cell::from(Span::styled(cluster.state.clone(), state_style)), + Cell::from(provider), + Cell::from(cpu), + Cell::from(mem), + Cell::from(duration), + Cell::from(last_activity), + ]) + .style(row_style) + }) + .collect(); + + let table = Table::new( + rows, + [ + Constraint::Min(18), + Constraint::Length(10), + Constraint::Length(10), + Constraint::Length(metrics::CPU_COLUMN_WIDTH as u16), + Constraint::Length(metrics::MEM_COLUMN_WIDTH as u16), + Constraint::Length(10), + Constraint::Length(8), + ], + ) + .header(header) + .row_highlight_style(theme::selected_style()) + .highlight_symbol(" > "); + + let mut table_state = TableState::default(); + table_state.select(Some(state.selected)); + + frame.render_stateful_widget(table, area, &mut table_state); +} + +fn render_empty(frame: &mut Frame<'_>, area: Rect) { + let lines = vec![ + Line::from(""), + Line::from(""), + Line::from(Span::styled("No active clusters", theme::muted_style())), + Line::from(""), + Line::from(Span::styled( + "Start a cluster from the Launcher (Esc)", + theme::dim_style(), + )), + Line::from(Span::styled( + "or run: zeroshot run --ship", + theme::dim_style(), + )), + ]; + let widget = Paragraph::new(lines).alignment(Alignment::Center); + frame.render_widget(widget, area); +} + +fn format_duration(delta_ms: i64) -> String { + let seconds = if delta_ms < 0 { 0 } else { delta_ms } / 1000; + if seconds < 60 { + return format!("{}s", seconds); + } + let minutes = seconds / 60; + if minutes < 60 { + return format!("{}m", minutes); + } + let hours = minutes / 60; + if hours < 24 { + return format!("{}h", hours); + } + let days = hours / 24; + format!("{}d", days) +} diff --git a/tui-rs/crates/zeroshot-tui/src/screens/radar.rs b/tui-rs/crates/zeroshot-tui/src/screens/radar.rs new file mode 100644 index 00000000..328a5229 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/screens/radar.rs @@ -0,0 +1,673 @@ +use std::collections::HashMap; + +use ratatui::layout::Rect; +use ratatui::style::{Color, Style}; +use ratatui::symbols::Marker; +use ratatui::text::{Line, Span}; +use ratatui::widgets::canvas::{Canvas, Circle, Points}; +use ratatui::widgets::{Block, Borders}; +use ratatui::Frame; + +use crate::app::animation::{self, AnimClock}; +use crate::app::Camera; +use crate::protocol::ClusterSummary; +use crate::ui::shared::calm_empty_state; +use crate::ui::theme; + +const POLL_INTERVAL_MS: i64 = 1000; +const WORLD_RADIUS: f64 = 48.0; +const LABEL_OFFSET: f64 = 6.0; +const RADIAL_LABEL_OFFSET: f64 = 2.0; +const BASE_ORB_RADIUS: f64 = 1.8; +const MAX_ORB_RADIUS: f64 = 4.6; +const ERROR_PULSE_RADIUS: f64 = 1.6; +const SELECTION_RING_RADIUS: f64 = 1.2; +const PIN_RING_RADIUS: f64 = 2.2; +const MIN_CAMERA_ZOOM: f32 = 0.2; +const ORB_SMOOTH_RATE: f64 = 0.25; + +const ACTIVITY_BANDS_MS: [i64; 4] = [5_000, 30_000, 120_000, 600_000]; +const RING_RADII: [f64; 5] = [10.0, 20.0, 30.0, 40.0, 46.0]; + +#[derive(Debug, Clone, Copy, PartialEq)] +pub struct LayoutPosition { + pub x: f64, + pub y: f64, + pub ring_radius: f64, +} + +#[derive(Debug, Clone, Copy, PartialEq)] +pub struct OrbVisual { + pub radius: f64, + pub intensity: f64, +} + +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub enum Direction { + Left, + Right, + Up, + Down, +} + +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub enum MoveSpeed { + Step, + Fast, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum Action { + MoveSelection { + direction: Direction, + speed: MoveSpeed, + }, + CenterOnSelection, + ResetView, +} + +#[derive(Debug, Clone, Default)] +pub struct FleetRadarState { + pub clusters: Vec, + pub selected: usize, + pub last_poll_at: Option, + pub last_message_counts: HashMap, + pub last_activity_at: HashMap, + pub last_message_deltas: HashMap, + pub layout_angles: HashMap, + pub orb_states: HashMap, +} + +impl FleetRadarState { + pub fn set_clusters(&mut self, clusters: Vec, now_ms: i64) { + let selected_id = self.selected_cluster_id(); + self.update_activity(&clusters, now_ms); + self.ensure_angles(&clusters); + self.ensure_orb_states(&clusters, now_ms); + self.clusters = clusters; + self.reconcile_selection(selected_id); + } + + pub fn poll_due(&self, now_ms: i64) -> bool { + match self.last_poll_at { + None => true, + Some(last) => now_ms.saturating_sub(last) >= POLL_INTERVAL_MS, + } + } + + pub fn mark_polled(&mut self, now_ms: i64) { + self.last_poll_at = Some(now_ms); + } + + pub fn selected_cluster_id(&self) -> Option { + self.clusters + .get(self.selected) + .map(|cluster| cluster.id.clone()) + } + + pub fn selected_layout(&self, now_ms: i64) -> Option { + let cluster = self.clusters.get(self.selected)?; + let age_ms = self.activity_age_ms(cluster, now_ms); + Some(self.layout_for(cluster.id.as_str(), age_ms)) + } + + pub fn move_selection_direction( + &mut self, + now_ms: i64, + direction: Direction, + speed: MoveSpeed, + ) -> bool { + let steps = match speed { + MoveSpeed::Step => 1, + MoveSpeed::Fast => 2, + }; + let mut moved = false; + for _ in 0..steps { + if self.move_selection_step(now_ms, direction) { + moved = true; + } else { + break; + } + } + moved + } + + pub fn activity_age_ms(&self, cluster: &ClusterSummary, now_ms: i64) -> i64 { + let activity_at = self + .last_activity_at + .get(&cluster.id) + .copied() + .unwrap_or(cluster.created_at); + now_ms.saturating_sub(activity_at) + } + + pub fn activity_delta(&self, cluster_id: &str) -> i64 { + self.last_message_deltas + .get(cluster_id) + .copied() + .unwrap_or(0) + } + + pub fn layout_for(&self, cluster_id: &str, activity_age_ms: i64) -> LayoutPosition { + let angle = self + .layout_angles + .get(cluster_id) + .copied() + .unwrap_or_else(|| stable_angle(cluster_id)); + layout_with_angle(angle, activity_age_ms) + } + + pub fn tick_orb_smoothing(&mut self, now_ms: i64, dt_ms: i64) { + if self.clusters.is_empty() { + return; + } + for cluster in &self.clusters { + let (target_radius, target_intensity) = orb_targets(self, cluster, now_ms); + let entry = self + .orb_states + .entry(cluster.id.clone()) + .or_insert(OrbVisual { + radius: target_radius, + intensity: target_intensity, + }); + entry.radius = + animation::smooth_toward_f64(entry.radius, target_radius, dt_ms, ORB_SMOOTH_RATE); + entry.intensity = animation::smooth_toward_f64( + entry.intensity, + target_intensity, + dt_ms, + ORB_SMOOTH_RATE, + ); + } + } + + pub fn orb_visual(&self, cluster: &ClusterSummary, now_ms: i64) -> OrbVisual { + self.orb_states + .get(&cluster.id) + .copied() + .unwrap_or_else(|| { + let (radius, intensity) = orb_targets(self, cluster, now_ms); + OrbVisual { radius, intensity } + }) + } + + fn update_activity(&mut self, clusters: &[ClusterSummary], now_ms: i64) { + let mut next_counts = HashMap::new(); + let mut next_activity = HashMap::new(); + let mut next_deltas = HashMap::new(); + + for cluster in clusters { + let prev_count = self.last_message_counts.get(&cluster.id).copied(); + let prev_activity = self.last_activity_at.get(&cluster.id).copied(); + let delta = prev_count + .map(|prev| cluster.message_count.saturating_sub(prev)) + .unwrap_or(cluster.message_count); + let mut activity = prev_activity; + + if prev_count + .map(|prev| cluster.message_count > prev) + .unwrap_or(true) + { + activity = Some(now_ms); + } + + next_counts.insert(cluster.id.clone(), cluster.message_count); + next_deltas.insert(cluster.id.clone(), delta); + if let Some(activity_at) = activity { + next_activity.insert(cluster.id.clone(), activity_at); + } + } + + self.last_message_counts = next_counts; + self.last_activity_at = next_activity; + self.last_message_deltas = next_deltas; + } + + fn ensure_angles(&mut self, clusters: &[ClusterSummary]) { + for cluster in clusters { + self.layout_angles + .entry(cluster.id.clone()) + .or_insert_with(|| stable_angle(cluster.id.as_str())); + } + self.layout_angles + .retain(|id, _| clusters.iter().any(|cluster| cluster.id == *id)); + } + + fn ensure_orb_states(&mut self, clusters: &[ClusterSummary], now_ms: i64) { + for cluster in clusters { + let (radius, intensity) = orb_targets(self, cluster, now_ms); + self.orb_states + .entry(cluster.id.clone()) + .or_insert(OrbVisual { radius, intensity }); + } + self.orb_states + .retain(|id, _| clusters.iter().any(|cluster| cluster.id == *id)); + } + + fn reconcile_selection(&mut self, selected_id: Option) { + if self.clusters.is_empty() { + self.selected = 0; + return; + } + + if let Some(id) = selected_id { + if let Some(index) = self.clusters.iter().position(|cluster| cluster.id == id) { + self.selected = index; + return; + } + } + + self.selected = 0; + } + + fn move_selection_step(&mut self, now_ms: i64, direction: Direction) -> bool { + if self.clusters.is_empty() { + self.selected = 0; + return false; + } + if self.selected >= self.clusters.len() { + self.selected = self.clusters.len().saturating_sub(1); + } + + let current_cluster = &self.clusters[self.selected]; + let current_layout = self.layout_for( + current_cluster.id.as_str(), + self.activity_age_ms(current_cluster, now_ms), + ); + + let mut best: Option<(usize, f64, f64)> = None; + + for (idx, cluster) in self.clusters.iter().enumerate() { + if idx == self.selected { + continue; + } + let layout = + self.layout_for(cluster.id.as_str(), self.activity_age_ms(cluster, now_ms)); + let dx = layout.x - current_layout.x; + let dy = layout.y - current_layout.y; + + let (axis, off) = match direction { + Direction::Right if dx > 0.0 => (dx, dy.abs()), + Direction::Left if dx < 0.0 => (-dx, dy.abs()), + Direction::Up if dy > 0.0 => (dy, dx.abs()), + Direction::Down if dy < 0.0 => (-dy, dx.abs()), + _ => continue, + }; + + let angle_score = off / axis; + let dist2 = dx * dx + dy * dy; + let replace = match best { + None => true, + Some((_best_idx, best_angle, best_dist)) => { + const EPS: f64 = 1e-6; + if (angle_score - best_angle).abs() > EPS { + angle_score < best_angle + } else if (dist2 - best_dist).abs() > EPS { + dist2 < best_dist + } else { + idx < _best_idx + } + } + }; + + if replace { + best = Some((idx, angle_score, dist2)); + } + } + + if let Some((idx, _, _)) = best { + self.selected = idx; + true + } else { + false + } + } +} + +pub fn layout_position(cluster_id: &str, activity_age_ms: i64) -> LayoutPosition { + let angle = stable_angle(cluster_id); + layout_with_angle(angle, activity_age_ms) +} + +pub fn render( + frame: &mut Frame<'_>, + area: Rect, + state: &FleetRadarState, + camera: &Camera, + now_ms: i64, + anim_clock: &AnimClock, + pinned_cluster_id: Option<&str>, +) { + if state.clusters.is_empty() { + render_empty(frame, area); + return; + } + + let selected_id = state.selected_cluster_id(); + let selected_id = selected_id.as_deref(); + let pinned_id = pinned_cluster_id; + let zoom = camera.zoom.max(MIN_CAMERA_ZOOM); + let half_span = WORLD_RADIUS / zoom as f64; + let center_x = camera.position.0 as f64; + let center_y = camera.position.1 as f64; + let pulse = animation::pulse_factor(anim_clock.phase) as f64; + + let canvas = Canvas::default() + .block(Block::default().borders(Borders::ALL).title("Fleet Radar")) + .x_bounds([center_x - half_span, center_x + half_span]) + .y_bounds([center_y - half_span, center_y + half_span]) + .marker(Marker::Braille) + .paint(|ctx| { + for ring in RING_RADII.iter().take(RING_RADII.len().saturating_sub(1)) { + ctx.draw(&Circle { + x: 0.0, + y: 0.0, + radius: *ring, + color: theme::FG_DIM, + }); + } + + for cluster in &state.clusters { + let age_ms = state.activity_age_ms(cluster, now_ms); + let layout = state.layout_for(cluster.id.as_str(), age_ms); + let color = cluster_color(cluster); + let orb = state.orb_visual(cluster, now_ms); + let orb_radius = orb.radius; + let intensity = orb.intensity; + let is_selected = selected_id == Some(cluster.id.as_str()); + let is_pinned = pinned_id == Some(cluster.id.as_str()); + let is_error = matches!(cluster.state.as_str(), "error" | "failed" | "failure"); + + if is_error { + let intensity_scale = 0.6 + 0.4 * intensity; + let pulse_scale = 0.7 + 0.6 * pulse; + ctx.draw(&Circle { + x: layout.x, + y: layout.y, + radius: orb_radius + ERROR_PULSE_RADIUS * intensity_scale * pulse_scale, + color: theme::STATUS_ERROR, + }); + } + + if is_selected { + ctx.draw(&Circle { + x: layout.x, + y: layout.y, + radius: orb_radius + SELECTION_RING_RADIUS, + color: theme::ACCENT, + }); + } + + if is_pinned { + ctx.draw(&Circle { + x: layout.x, + y: layout.y, + radius: orb_radius + PIN_RING_RADIUS, + color: theme::ACCENT2, + }); + } + + ctx.draw(&Circle { + x: layout.x, + y: layout.y, + radius: orb_radius, + color, + }); + + ctx.draw(&Points { + coords: &[(layout.x, layout.y)], + color, + }); + + let label = truncate_label(cluster.id.as_str()); + let label_style = Style::default().fg(color); + let line = Line::from(Span::styled(label, label_style)); + let (label_x, label_y) = label_position(layout.x, layout.y, cluster.id.as_str()); + ctx.print(label_x, label_y, line); + } + }); + + frame.render_widget(canvas, area); +} + +fn render_empty(frame: &mut Frame<'_>, area: Rect) { + let widget = calm_empty_state( + "Fleet Radar", + "No clusters yet.", + Some("Type an intent in the spine to start a cluster."), + None, + ); + frame.render_widget(widget, area); +} + +fn activity_ring(age_ms: i64) -> f64 { + let age = if age_ms < 0 { 0 } else { age_ms }; + for (index, cutoff) in ACTIVITY_BANDS_MS.iter().enumerate() { + if age <= *cutoff { + return RING_RADII[index]; + } + } + *RING_RADII.last().unwrap_or(&WORLD_RADIUS) +} + +fn layout_with_angle(angle: f64, activity_age_ms: i64) -> LayoutPosition { + let ring_radius = activity_ring(activity_age_ms); + let x = ring_radius * angle.cos(); + let y = ring_radius * angle.sin(); + LayoutPosition { x, y, ring_radius } +} + +fn label_position(x: f64, y: f64, cluster_id: &str) -> (f64, f64) { + let hash = stable_hash(cluster_id); + let side = if hash & 1 == 0 { 1.0 } else { -1.0 }; + let angle = if x == 0.0 && y == 0.0 { + 0.0 + } else { + y.atan2(x) + }; + let tangent_x = -angle.sin(); + let tangent_y = angle.cos(); + let radial_x = -angle.cos(); + let radial_y = -angle.sin(); + let mut lx = x + tangent_x * LABEL_OFFSET * side + radial_x * RADIAL_LABEL_OFFSET; + let mut ly = y + tangent_y * LABEL_OFFSET * side + radial_y * RADIAL_LABEL_OFFSET; + let min = -WORLD_RADIUS + 1.0; + let max = WORLD_RADIUS - 1.0; + if lx < min { + lx = min; + } else if lx > max { + lx = max; + } + if ly < min { + ly = min; + } else if ly > max { + ly = max; + } + (lx, ly) +} + +fn orb_radius(delta: i64, age_ms: i64) -> f64 { + let delta_boost = (delta.max(0) as f64).min(6.0) * 0.35; + let recency_boost = if age_ms <= 5_000 { + 0.9 + } else if age_ms <= 30_000 { + 0.4 + } else { + 0.0 + }; + (BASE_ORB_RADIUS + delta_boost + recency_boost).min(MAX_ORB_RADIUS) +} + +fn orb_intensity(delta: i64, age_ms: i64) -> f64 { + let delta_norm = (delta.max(0) as f64).min(6.0) / 6.0; + let recency = if age_ms <= 5_000 { + 1.0 + } else if age_ms <= 30_000 { + 0.6 + } else { + 0.3 + }; + (0.3 + delta_norm * 0.5 + recency * 0.2).min(1.0) +} + +fn orb_targets(state: &FleetRadarState, cluster: &ClusterSummary, now_ms: i64) -> (f64, f64) { + let age_ms = state.activity_age_ms(cluster, now_ms); + let delta = state.activity_delta(cluster.id.as_str()); + (orb_radius(delta, age_ms), orb_intensity(delta, age_ms)) +} + +fn truncate_label(id: &str) -> String { + const LIMIT: usize = 10; + let mut iter = id.chars(); + let mut out = String::new(); + for _ in 0..LIMIT { + match iter.next() { + Some(ch) => out.push(ch), + None => return id.to_string(), + } + } + if iter.next().is_some() { + out.push_str(".."); + } + out +} + +fn cluster_color(cluster: &ClusterSummary) -> Color { + theme::status_style(&cluster.state) + .fg + .unwrap_or(theme::FG_MUTED) +} + +fn stable_angle(input: &str) -> f64 { + let hash = stable_hash(input); + let fraction = (hash % 3600) as f64 / 3600.0; + std::f64::consts::TAU * fraction +} + +fn stable_hash(input: &str) -> u64 { + const FNV_OFFSET: u64 = 0xcbf29ce484222325; + const FNV_PRIME: u64 = 0x00000100000001b3; + let mut hash = FNV_OFFSET; + for byte in input.as_bytes() { + hash ^= *byte as u64; + hash = hash.wrapping_mul(FNV_PRIME); + } + hash +} + +#[cfg(test)] +mod tests { + use super::*; + use crate::ui::widgets::test_utils::line_text; + use ratatui::backend::TestBackend; + use ratatui::buffer::Buffer; + use ratatui::Terminal; + + fn cluster(id: &str) -> ClusterSummary { + ClusterSummary { + id: id.to_string(), + state: "running".to_string(), + provider: None, + created_at: 0, + agent_count: 1, + message_count: 0, + cwd: None, + } + } + + fn buffer_contains(buffer: &Buffer, needle: &str) -> bool { + for y in 0..buffer.area.height { + if line_text(buffer, y).contains(needle) { + return true; + } + } + false + } + + #[test] + fn fleet_radar_renders_empty_state() { + let backend = TestBackend::new(60, 12); + let mut terminal = Terminal::new(backend).expect("terminal"); + let state = FleetRadarState::default(); + let camera = Camera::default(); + let anim_clock = AnimClock::default(); + + terminal + .draw(|frame| { + let area = frame.area(); + render(frame, area, &state, &camera, 0, &anim_clock, None); + }) + .expect("draw"); + + let buffer = terminal.backend().buffer(); + assert!(buffer_contains(buffer, "No clusters yet.")); + assert!(buffer_contains( + buffer, + "Type an intent in the spine to start a cluster." + )); + } + + #[test] + fn layout_position_is_deterministic() { + let first = layout_position("cluster-1", 10_000); + let second = layout_position("cluster-1", 10_000); + assert_eq!(first, second); + } + + #[test] + fn layout_position_changes_with_age_band() { + let recent = layout_position("cluster-1", 1_000); + let older = layout_position("cluster-1", 1_000_000); + assert!(recent.ring_radius < older.ring_radius); + } + + #[test] + fn truncate_label_handles_unicode() { + let label = "Ī±Ī²Ī³Ī“ĪµĪ¶Ī·ĪøĪ¹ĪŗĪ»Ī¼"; + let truncated = truncate_label(label); + assert_eq!(truncated, "Ī±Ī²Ī³Ī“ĪµĪ¶Ī·ĪøĪ¹Īŗ.."); + } + + #[test] + fn radar_directional_selection() { + let now_ms = 10_000; + let mut state = FleetRadarState::default(); + state.set_clusters( + vec![ + cluster("east"), + cluster("west"), + cluster("north"), + cluster("south"), + ], + now_ms, + ); + state.layout_angles.insert("east".to_string(), 0.0); + state + .layout_angles + .insert("north".to_string(), std::f64::consts::FRAC_PI_2); + state + .layout_angles + .insert("west".to_string(), std::f64::consts::PI); + state + .layout_angles + .insert("south".to_string(), std::f64::consts::TAU * 0.75); + + state.selected = state + .clusters + .iter() + .position(|cluster| cluster.id == "west") + .unwrap(); + + assert!(state.move_selection_direction(now_ms, Direction::Right, MoveSpeed::Step)); + assert_eq!(state.selected_cluster_id().as_deref(), Some("east")); + + assert!(state.move_selection_direction(now_ms, Direction::Up, MoveSpeed::Step)); + assert_eq!(state.selected_cluster_id().as_deref(), Some("north")); + + assert!(state.move_selection_direction(now_ms, Direction::Down, MoveSpeed::Step)); + assert_eq!(state.selected_cluster_id().as_deref(), Some("south")); + + assert!(!state.move_selection_direction(now_ms, Direction::Down, MoveSpeed::Step)); + assert_eq!(state.selected_cluster_id().as_deref(), Some("south")); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/terminal.rs b/tui-rs/crates/zeroshot-tui/src/terminal.rs new file mode 100644 index 00000000..d2006a75 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/terminal.rs @@ -0,0 +1,70 @@ +use std::io::{self, stdout}; +use std::sync::atomic::{AtomicBool, Ordering}; +use std::sync::Arc; + +use crossterm::{cursor, execute, terminal}; + +#[derive(Debug)] +pub struct TerminalGuard { + restored: Arc, +} + +impl TerminalGuard { + pub fn new() -> io::Result { + terminal::enable_raw_mode()?; + if let Err(err) = execute!(stdout(), terminal::EnterAlternateScreen, cursor::Hide) { + if let Err(disable_err) = terminal::disable_raw_mode() { + return Err(disable_err); + } + return Err(err); + } + Ok(Self { + restored: Arc::new(AtomicBool::new(false)), + }) + } + + pub fn install_panic_hook(&self) { + let restored = self.restored.clone(); + let previous = std::panic::take_hook(); + std::panic::set_hook(Box::new(move |info| { + if let Err(err) = restore_terminal(&restored) { + eprintln!("Failed to restore terminal on panic: {err}"); + } + previous(info); + })); + } + + pub fn restore(&self) -> io::Result<()> { + restore_terminal(&self.restored) + } +} + +impl Drop for TerminalGuard { + fn drop(&mut self) { + if let Err(err) = restore_terminal(&self.restored) { + eprintln!("Failed to restore terminal on drop: {err}"); + } + } +} + +fn restore_terminal(restored: &AtomicBool) -> io::Result<()> { + if restored.swap(true, Ordering::SeqCst) { + return Ok(()); + } + + let mut first_error: Option = None; + if let Err(err) = terminal::disable_raw_mode() { + first_error = Some(err); + } + if let Err(err) = execute!(stdout(), terminal::LeaveAlternateScreen, cursor::Show) { + if first_error.is_none() { + first_error = Some(err); + } + } + + if let Some(err) = first_error { + return Err(err); + } + + Ok(()) +} diff --git a/tui-rs/crates/zeroshot-tui/src/ui/launcher.rs b/tui-rs/crates/zeroshot-tui/src/ui/launcher.rs new file mode 100644 index 00000000..30a586b8 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/ui/launcher.rs @@ -0,0 +1,156 @@ +use ratatui::layout::{Alignment, Constraint, Layout, Position, Rect}; +use ratatui::text::{Line, Span}; +use ratatui::widgets::{Block, BorderType, Borders, Paragraph}; +use ratatui::Frame; + +use crate::protocol::ClusterSummary; +use crate::screens::launcher::State; +use crate::ui::theme; + +/// Maximum width for the centered content area. +const MAX_CONTENT_WIDTH: u16 = 60; + +pub fn render( + frame: &mut Frame<'_>, + area: Rect, + state: &State, + provider_override: Option<&str>, + recent_clusters: &[ClusterSummary], +) { + let centered = center_rect(area, MAX_CONTENT_WIDTH); + + // Vertical layout: logo (3) + input (3) + gap (1) + quick actions (6) + gap (1) + recent (rest) + let [logo_area, input_area, _, actions_area, _, recent_area] = Layout::vertical([ + Constraint::Length(3), // logo + Constraint::Length(3), // input + Constraint::Length(1), // gap + Constraint::Length(6), // quick actions + Constraint::Length(1), // gap + Constraint::Min(2), // recent clusters + ]) + .areas(centered); + + render_logo(frame, logo_area, provider_override); + render_input(frame, input_area, state); + render_quick_actions(frame, actions_area); + render_recent(frame, recent_area, recent_clusters); +} + +fn render_logo(frame: &mut Frame<'_>, area: Rect, provider_override: Option<&str>) { + let provider = provider_override.unwrap_or("default"); + let lines = vec![ + Line::from(""), + Line::from(vec![Span::styled( + "\u{25c6} Z E R O S H O T", + theme::logo_style(), + )]), + Line::from(vec![ + Span::styled("Multi-Agent Orchestrator", theme::dim_style()), + Span::raw(" "), + Span::styled(format!("[{provider}]"), theme::muted_style()), + ]), + ]; + let widget = Paragraph::new(lines).alignment(Alignment::Center); + frame.render_widget(widget, area); +} + +fn render_input(frame: &mut Frame<'_>, area: Rect, state: &State) { + let content = if state.input.is_empty() { + Line::from(Span::styled( + "Describe a task or paste an issue URL...", + theme::muted_style(), + )) + } else { + Line::from(Span::styled(state.input.as_str(), theme::title_style())) + }; + + let block = Block::default() + .borders(Borders::ALL) + .border_style(theme::focus_border_style()) + .border_type(BorderType::Rounded); + + let widget = Paragraph::new(content).block(block); + frame.render_widget(widget, area); + + // Set cursor + if area.height > 2 && area.width > 2 { + let max_x = area.x + area.width.saturating_sub(2); + let cursor_x = area.x + 1 + state.cursor as u16; + let cursor_x = cursor_x.min(max_x); + let cursor_y = area.y + 1; + frame.set_cursor_position(Position::new(cursor_x, cursor_y)); + } +} + +fn render_quick_actions(frame: &mut Frame<'_>, area: Rect) { + let lines = vec![ + Line::from(vec![ + Span::raw(" "), + Span::styled("/issue", theme::key_style()), + Span::styled(" org/repo#123 ", theme::dim_style()), + Span::styled("Start from issue", theme::dim_style()), + ]), + Line::from(vec![ + Span::raw(" "), + Span::styled("/monitor", theme::key_style()), + Span::styled(" ", theme::dim_style()), + Span::styled("View active runs", theme::dim_style()), + ]), + Line::from(vec![ + Span::raw(" "), + Span::styled("/provider", theme::key_style()), + Span::styled(" ", theme::dim_style()), + Span::styled("Switch AI model", theme::dim_style()), + ]), + ]; + + let block = Block::default() + .title(Span::styled(" Quick Actions ", theme::dim_style())) + .borders(Borders::ALL) + .border_type(BorderType::Rounded) + .border_style(theme::unfocus_border_style()); + + let widget = Paragraph::new(lines).block(block); + frame.render_widget(widget, area); +} + +fn render_recent(frame: &mut Frame<'_>, area: Rect, clusters: &[ClusterSummary]) { + let mut lines = vec![Line::from(Span::styled("Recent", theme::dim_style()))]; + + if clusters.is_empty() { + lines.push(Line::from(Span::styled( + "(no recent clusters)", + theme::muted_style(), + ))); + } else { + for cluster in clusters.iter().take(3) { + let state_style = theme::status_style(&cluster.state); + lines.push(Line::from(vec![ + Span::raw(" "), + Span::styled(&cluster.id, theme::dim_style()), + Span::raw(" "), + Span::styled(&cluster.state, state_style), + ])); + } + } + + let widget = Paragraph::new(lines); + frame.render_widget(widget, area); +} + +/// Center a rect horizontally within the outer area, capping at max_width. +fn center_rect(outer: Rect, max_width: u16) -> Rect { + let width = outer.width.min(max_width); + let x = outer.x + (outer.width.saturating_sub(width)) / 2; + + // Vertically center if enough space (aim for ~1/3 from top) + let content_height = 17u16; // approximate total height of all sections + let y = if outer.height > content_height + 4 { + outer.y + (outer.height.saturating_sub(content_height)) / 3 + } else { + outer.y + }; + let height = outer.height.saturating_sub(y - outer.y); + + Rect::new(x, y, width, height) +} diff --git a/tui-rs/crates/zeroshot-tui/src/ui/mod.rs b/tui-rs/crates/zeroshot-tui/src/ui/mod.rs new file mode 100644 index 00000000..ca8d74bf --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/ui/mod.rs @@ -0,0 +1,443 @@ +use ratatui::layout::{Alignment, Constraint, Layout, Rect}; +use ratatui::text::{Line, Span}; +use ratatui::widgets::Paragraph; +use ratatui::Frame; + +use crate::app::{ + AppState, BackendStatus, FocusTarget, ScreenId, SpineHint, SpineHintTone, UiVariant, +}; +use crate::screens::{agent, agent_microscope, cluster, cluster_canvas, monitor, radar}; +use crate::ui::widgets::{command_bar, scrub_bar, spine, toast}; + +pub mod launcher; +pub mod scene; +pub mod shared; +pub mod theme; +pub mod widgets; + +const DISRUPTIVE_SPINE_HINT: &str = "/guide /nudge /interrupt /pin / i ? Tab Esc Enter"; + +pub fn render(frame: &mut Frame<'_>, state: &AppState) { + if matches!(state.ui_variant, UiVariant::Disruptive) { + render_disruptive(frame, state); + return; + } + + let size = frame.area(); + let [header_area, content_area, status_area] = Layout::vertical([ + Constraint::Length(1), // header + Constraint::Min(1), // content + Constraint::Length(1), // status bar / command bar + ]) + .areas(size); + + render_header(frame, header_area, state); + + match state.active_screen() { + ScreenId::Launcher | ScreenId::IntentConsole => launcher::render( + frame, + content_area, + &state.launcher, + state.provider_override.as_deref(), + &state.monitor.clusters, + ), + ScreenId::Monitor | ScreenId::FleetRadar => monitor::render( + frame, + content_area, + &state.monitor, + &state.metrics, + state.now_ms, + ), + ScreenId::Cluster { id } => { + if let Some(cluster_state) = state.clusters.get(id) { + let metrics = state.metrics.get(id); + cluster::render(frame, content_area, cluster_state, metrics); + } else { + let default_state = cluster::State::default(); + cluster::render(frame, content_area, &default_state, None); + } + } + ScreenId::ClusterCanvas { id } => { + let cluster_state = state.clusters.get(id); + let canvas_state = state.cluster_canvases.get(id); + cluster_canvas::render( + frame, + content_area, + cluster_canvas::RenderContext { + cluster_id: id, + cluster_state, + canvas_state, + time_cursor: &state.time_cursor, + anim_clock: &state.anim_clock, + pinned_target: state.pinned_target.as_ref(), + }, + ); + } + ScreenId::Agent { + cluster_id, + agent_id, + } => { + let key = crate::app::AgentKey::new(cluster_id.clone(), agent_id.clone()); + if let Some(agent_state) = state.agents.get(&key) { + agent::render(frame, content_area, agent_state, cluster_id, agent_id); + } else { + let default_state = agent::State::default(); + agent::render(frame, content_area, &default_state, cluster_id, agent_id); + } + } + ScreenId::AgentMicroscope { + cluster_id, + agent_id, + } => { + let key = crate::app::AgentKey::new(cluster_id.clone(), agent_id.clone()); + let microscope_state = state.agent_microscopes.get(&key); + let cluster_state = state.clusters.get(cluster_id); + agent_microscope::render( + frame, + content_area, + cluster_id, + agent_id, + cluster_state.map(|state| &state.timeline_time), + microscope_state, + &state.time_cursor, + ); + } + } + + // Status bar: if command bar active, show command input; otherwise show hints + toast + let allow_command_bar = !matches!( + state.active_screen(), + ScreenId::Launcher | ScreenId::IntentConsole + ); + if state.command_bar.active { + command_bar::render(frame, status_area, &state.command_bar, allow_command_bar); + command_bar::set_cursor(frame, status_area, &state.command_bar); + } else { + render_status_bar(frame, status_area, state); + } +} + +fn render_disruptive(frame: &mut Frame<'_>, state: &AppState) { + let size = frame.area(); + let (canvas_area, scrub_area, spine_area) = if size.height >= 4 { + let [canvas_area, scrub_area, spine_area] = Layout::vertical([ + Constraint::Min(1), + Constraint::Length(1), + Constraint::Length(2), + ]) + .areas(size); + (canvas_area, Some(scrub_area), spine_area) + } else { + let [canvas_area, spine_area] = + Layout::vertical([Constraint::Min(1), Constraint::Length(2)]).areas(size); + (canvas_area, None, spine_area) + }; + + match state.active_screen() { + ScreenId::FleetRadar | ScreenId::Launcher | ScreenId::IntentConsole | ScreenId::Monitor => { + let pinned_cluster = match state.pinned_target.as_ref() { + Some(FocusTarget::Cluster { id }) => Some(id.as_str()), + _ => None, + }; + radar::render( + frame, + canvas_area, + &state.fleet_radar, + &state.camera, + state.now_ms, + &state.anim_clock, + pinned_cluster, + ); + } + ScreenId::ClusterCanvas { id } => { + let cluster_state = state.clusters.get(id); + let canvas_state = state.cluster_canvases.get(id); + cluster_canvas::render( + frame, + canvas_area, + cluster_canvas::RenderContext { + cluster_id: id, + cluster_state, + canvas_state, + time_cursor: &state.time_cursor, + anim_clock: &state.anim_clock, + pinned_target: state.pinned_target.as_ref(), + }, + ); + } + ScreenId::Cluster { id } => { + if let Some(cluster_state) = state.clusters.get(id) { + let metrics = state.metrics.get(id); + cluster::render(frame, canvas_area, cluster_state, metrics); + } else { + let default_state = cluster::State::default(); + cluster::render(frame, canvas_area, &default_state, None); + } + } + ScreenId::Agent { + cluster_id, + agent_id, + } => { + let key = crate::app::AgentKey::new(cluster_id.clone(), agent_id.clone()); + if let Some(agent_state) = state.agents.get(&key) { + agent::render(frame, canvas_area, agent_state, cluster_id, agent_id); + } else { + let default_state = agent::State::default(); + agent::render(frame, canvas_area, &default_state, cluster_id, agent_id); + } + } + ScreenId::AgentMicroscope { + cluster_id, + agent_id, + } => { + let key = crate::app::AgentKey::new(cluster_id.clone(), agent_id.clone()); + let microscope_state = state.agent_microscopes.get(&key); + let cluster_state = state.clusters.get(cluster_id); + agent_microscope::render( + frame, + canvas_area, + cluster_id, + agent_id, + cluster_state.map(|state| &state.timeline_time), + microscope_state, + &state.time_cursor, + ); + } + } + + if let Some(scrub_area) = scrub_area { + let scrub_state = match state.active_screen() { + ScreenId::ClusterCanvas { id } => Some(scrub_bar::ScrubBarState { + time_cursor: &state.time_cursor, + logs: state.clusters.get(id).map(|entry| &entry.logs_time), + agent_id: None, + }), + ScreenId::AgentMicroscope { + cluster_id, + agent_id, + } => Some(scrub_bar::ScrubBarState { + time_cursor: &state.time_cursor, + logs: state + .agent_microscopes + .get(&crate::app::AgentKey::new( + cluster_id.clone(), + agent_id.clone(), + )) + .map(|entry| &entry.logs_time), + agent_id: None, + }), + _ => None, + }; + if let Some(scrub_state) = scrub_state { + scrub_bar::render(frame, scrub_area, scrub_state); + } + } + + let mut spine_state = state.spine.clone(); + if let Some(toast_state) = state.toast.as_ref() { + if let Some((toast_text, _)) = toast::format_inline(Some(toast_state)) { + spine_state.hint = SpineHint::from_toast(toast_text, toast_state.level.clone()); + } + } else if spine_state.hint.is_empty() { + if let Some(hint) = backend_status_hint(&state.backend_status) { + spine_state.hint = hint; + } else { + spine_state.hint = SpineHint::new(DISRUPTIVE_SPINE_HINT, SpineHintTone::Muted); + } + } + spine::render(frame, spine_area, &spine_state); + spine::set_cursor(frame, spine_area, &spine_state); +} + +fn backend_status_hint(status: &BackendStatus) -> Option { + match status { + BackendStatus::Connected => None, + BackendStatus::Disconnected => Some(SpineHint::new( + "ā—‹ Backend disconnected", + SpineHintTone::Muted, + )), + BackendStatus::Error(_) => Some(SpineHint::new("āœ— Backend error", SpineHintTone::Error)), + BackendStatus::Exited(_) => Some(SpineHint::new("āœ— Backend exited", SpineHintTone::Error)), + } +} + +fn render_header(frame: &mut Frame<'_>, area: Rect, state: &AppState) { + let screen = state.active_screen(); + let breadcrumb = screen_breadcrumb(screen); + + let (status_dot, status_style) = match &state.backend_status { + BackendStatus::Connected => ("ā—", theme::backend_connected_style()), + BackendStatus::Disconnected => ("ā—‹", theme::backend_error_style()), + BackendStatus::Error(_) => ("āœ—", theme::backend_error_style()), + BackendStatus::Exited(_) => ("āœ—", theme::backend_error_style()), + }; + + let provider_label = state.provider_override.as_deref().unwrap_or("default"); + + // Build left side + let left = Line::from(vec![ + Span::styled("ā—† ZEROSHOT", theme::logo_style()), + Span::raw(" "), + Span::styled(breadcrumb, theme::title_style()), + ]); + + // Build right side + let right_text = format!("{status_dot} {provider_label}"); + let right_len = right_text.len() as u16 + 1; + let right = Line::from(vec![ + Span::styled(status_dot, status_style), + Span::raw(" "), + Span::styled(provider_label, theme::dim_style()), + ]); + + // Render left-aligned header + let widget = Paragraph::new(left); + frame.render_widget(widget, area); + + // Render right-aligned status + if area.width > right_len + 20 { + let right_area = Rect { + x: area.x + area.width.saturating_sub(right_len), + y: area.y, + width: right_len, + height: 1, + }; + let right_widget = Paragraph::new(right).alignment(Alignment::Right); + frame.render_widget(right_widget, right_area); + } +} + +fn render_status_bar(frame: &mut Frame<'_>, area: Rect, state: &AppState) { + let hints = screen_hints(state.active_screen()); + let toast_msg = toast::format_inline(state.toast.as_ref()); + + let mut spans = Vec::new(); + spans.push(Span::raw(" ")); + for (i, (key, desc)) in hints.iter().enumerate() { + if i > 0 { + spans.push(Span::raw(" ")); + } + spans.push(Span::styled(*key, theme::key_style())); + spans.push(Span::styled(format!(":{desc}"), theme::key_desc_style())); + } + + // Calculate space for toast on right side + if let Some((toast_text, toast_style)) = toast_msg { + let hints_len: usize = spans.iter().map(|s| s.content.len()).sum(); + let toast_len = toast_text.len() + 2; + let available = area.width as usize; + if hints_len + toast_len + 4 < available { + let gap = available.saturating_sub(hints_len + toast_len + 1); + spans.push(Span::raw(" ".repeat(gap))); + spans.push(Span::styled(toast_text, toast_style)); + } + } + + let widget = Paragraph::new(Line::from(spans)); + frame.render_widget(widget, area); +} + +fn screen_breadcrumb(screen: &ScreenId) -> String { + match screen { + ScreenId::Launcher => "Launcher".to_string(), + ScreenId::Monitor => "Monitor".to_string(), + ScreenId::IntentConsole => "Intent Console".to_string(), + ScreenId::FleetRadar => "Fleet Radar".to_string(), + ScreenId::Cluster { id } => format!("Monitor > {}", truncate_id(id)), + ScreenId::ClusterCanvas { id } => format!("Fleet Radar > {}", truncate_id(id)), + ScreenId::Agent { + cluster_id, + agent_id, + } => format!("Monitor > {} > {}", truncate_id(cluster_id), agent_id), + ScreenId::AgentMicroscope { + cluster_id, + agent_id, + } => format!("Fleet Radar > {} > {}", truncate_id(cluster_id), agent_id), + } +} + +fn truncate_id(id: &str) -> String { + const LIMIT: usize = 16; + let mut iter = id.chars(); + let mut out = String::new(); + for _ in 0..LIMIT { + match iter.next() { + Some(ch) => out.push(ch), + None => return id.to_string(), + } + } + out +} + +fn screen_hints(screen: &ScreenId) -> Vec<(&'static str, &'static str)> { + match screen { + ScreenId::Launcher => vec![("Enter", "start"), ("/", "commands"), ("Ctrl+C", "quit")], + ScreenId::IntentConsole => vec![("i", "intent"), ("/", "commands"), ("Esc", "back")], + ScreenId::Monitor => vec![ + ("j/k", "navigate"), + ("Enter", "open"), + ("/", "commands"), + ("Esc", "back"), + ], + ScreenId::FleetRadar => vec![ + ("h/j/k/l", "select"), + ("g/G", "center"), + ("Enter", "zoom"), + ("/", "commands"), + ("Esc", "back"), + ], + ScreenId::Cluster { .. } => vec![ + ("Tab", "pane"), + ("j/k", "scroll"), + ("Enter", "agent"), + ("Esc", "back"), + ], + ScreenId::ClusterCanvas { .. } => vec![ + ("h/j/k/l", "focus"), + ("Shift+h/j/k/l", "fast"), + ("Enter", "zoom"), + ("Esc", "back"), + ], + ScreenId::Agent { .. } => vec![("Enter", "send"), ("j/k", "scroll"), ("Esc", "back")], + ScreenId::AgentMicroscope { .. } => vec![("Esc", "back")], + } +} + +#[cfg(test)] +mod tests { + use super::*; + use crate::ui::widgets::test_utils::line_text; + use ratatui::backend::TestBackend; + use ratatui::buffer::Buffer; + use ratatui::Terminal; + + fn buffer_contains(buffer: &Buffer, needle: &str) -> bool { + for y in 0..buffer.area.height { + if line_text(buffer, y).contains(needle) { + return true; + } + } + false + } + + #[test] + fn disruptive_spine_shows_backend_disconnected() { + let backend = TestBackend::new(80, 8); + let mut terminal = Terminal::new(backend).expect("terminal"); + let mut state = AppState::default(); + state.ui_variant = UiVariant::Disruptive; + state.screen_stack = vec![ScreenId::IntentConsole, ScreenId::FleetRadar]; + state.backend_status = BackendStatus::Disconnected; + state.spine.hint = SpineHint::empty(); + state.toast = None; + + terminal + .draw(|frame| { + render(frame, &state); + }) + .expect("draw"); + + let buffer = terminal.backend().buffer(); + assert!(buffer_contains(buffer, "Backend disconnected")); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/ui/scene.rs b/tui-rs/crates/zeroshot-tui/src/ui/scene.rs new file mode 100644 index 00000000..88c40b88 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/ui/scene.rs @@ -0,0 +1,82 @@ +use std::collections::HashMap; + +use ratatui::style::Style; + +#[derive(Debug, Clone, Copy)] +pub struct WorldBounds { + pub min: (f32, f32), + pub max: (f32, f32), +} + +impl Default for WorldBounds { + fn default() -> Self { + Self { + min: (0.0, 0.0), + max: (0.0, 0.0), + } + } +} + +#[derive(Debug, Clone, Default)] +pub struct Scene { + pub world_bounds: WorldBounds, + pub objects: Vec, + pub overlays: Vec, +} + +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub enum ObjectKind { + Node, + Edge, + Unknown, +} + +#[derive(Debug, Clone)] +pub struct Object { + pub id: String, + pub kind: ObjectKind, + pub pos: (f32, f32), + pub radius: f32, + pub style: Style, + pub label: Option, + pub metadata: HashMap, +} + +impl Default for Object { + fn default() -> Self { + Self { + id: String::new(), + kind: ObjectKind::Unknown, + pos: (0.0, 0.0), + radius: 0.0, + style: Style::default(), + label: None, + metadata: HashMap::new(), + } + } +} + +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub enum OverlayKind { + Label, + Grid, + Cursor, + Unknown, +} + +#[derive(Debug, Clone)] +pub struct Overlay { + pub kind: OverlayKind, + pub label: Option, + pub metadata: HashMap, +} + +impl Default for Overlay { + fn default() -> Self { + Self { + kind: OverlayKind::Unknown, + label: None, + metadata: HashMap::new(), + } + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/ui/shared.rs b/tui-rs/crates/zeroshot-tui/src/ui/shared.rs new file mode 100644 index 00000000..07fe47d2 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/ui/shared.rs @@ -0,0 +1,434 @@ +use std::collections::VecDeque; + +use ratatui::layout::Alignment; +use ratatui::text::{Line, Span}; +use ratatui::widgets::{Block, BorderType, Borders, Paragraph}; + +use crate::ui::theme; + +// ā”€ā”€ ScrollableBuffer ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ + +/// A capped VecDeque with scroll offset tracking. +/// +/// Used by cluster logs, timeline events, and agent logs to manage +/// scrollable content with a maximum capacity. +#[derive(Debug, Clone)] +pub struct ScrollableBuffer { + pub items: VecDeque, + pub scroll_offset: usize, + max_capacity: usize, +} + +impl ScrollableBuffer { + pub fn new(max_capacity: usize) -> Self { + Self { + items: VecDeque::new(), + scroll_offset: 0, + max_capacity, + } + } + + pub fn push_many(&mut self, items: impl IntoIterator) { + let before = self.items.len(); + self.items.extend(items); + let added = self.items.len() - before; + self.adjust_scroll_on_append(added); + let dropped = self.trim(); + self.adjust_scroll_on_trim(dropped); + self.clamp_scroll(); + } + + pub fn move_scroll(&mut self, delta: i32) { + let len = self.items.len(); + if len == 0 { + self.scroll_offset = 0; + return; + } + if delta < 0 { + self.scroll_offset = self + .scroll_offset + .saturating_add(delta.unsigned_abs() as usize); + } else { + self.scroll_offset = self.scroll_offset.saturating_sub(delta as usize); + } + self.clamp_scroll(); + } + + pub fn len(&self) -> usize { + self.items.len() + } + + pub fn is_empty(&self) -> bool { + self.items.is_empty() + } + + fn adjust_scroll_on_append(&mut self, added: usize) { + if self.scroll_offset > 0 { + self.scroll_offset = self.scroll_offset.saturating_add(added); + } + } + + fn adjust_scroll_on_trim(&mut self, dropped: usize) { + self.scroll_offset = self.scroll_offset.saturating_sub(dropped); + } + + fn clamp_scroll(&mut self) { + let max_offset = self.items.len().saturating_sub(1); + if self.scroll_offset > max_offset { + self.scroll_offset = max_offset; + } + } + + fn trim(&mut self) -> usize { + if self.items.len() <= self.max_capacity { + return 0; + } + let mut dropped = 0usize; + while self.items.len() > self.max_capacity { + self.items.pop_front(); + dropped += 1; + } + dropped + } +} + +// ā”€ā”€ TimeIndexedBuffer ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ + +pub trait HasTimestamp { + fn timestamp_ms(&self) -> i64; +} + +/// A capped, time-indexed buffer with stable insertion ordering. +/// +/// Optimized for windowed reads by timestamp while maintaining bounded memory. +#[derive(Debug, Clone)] +pub struct TimeIndexedBuffer { + items: VecDeque, + max_capacity: usize, +} + +impl TimeIndexedBuffer { + pub fn new(max_capacity: usize) -> Self { + Self { + items: VecDeque::new(), + max_capacity, + } + } + + pub fn len(&self) -> usize { + self.items.len() + } + + pub fn is_empty(&self) -> bool { + self.items.is_empty() + } + + pub fn push_many(&mut self, items: impl IntoIterator) { + self.items.extend(items); + self.trim(); + } + + pub fn window(&self, t_ms: i64, window_ms: i64) -> Vec<&T> { + if self.items.is_empty() { + return Vec::new(); + } + let window_ms = window_ms.max(0); + let start = t_ms.saturating_sub(window_ms); + let end = t_ms; + let lower = self.lower_bound(start); + let upper = self.upper_bound(end); + let mut out = Vec::with_capacity(upper.saturating_sub(lower)); + for idx in lower..upper { + if let Some(item) = self.items.get(idx) { + out.push(item); + } + } + out + } + + pub fn latest(&self, n: usize) -> Vec<&T> { + if n == 0 || self.items.is_empty() { + return Vec::new(); + } + let len = self.items.len(); + let start = len.saturating_sub(n); + let mut out = Vec::with_capacity(len - start); + for idx in start..len { + if let Some(item) = self.items.get(idx) { + out.push(item); + } + } + out + } + + pub fn iter(&self) -> impl Iterator { + self.items.iter() + } + + pub fn iter_rev(&self) -> impl Iterator { + self.items.iter().rev() + } + + fn lower_bound(&self, target: i64) -> usize { + let mut left = 0usize; + let mut right = self.items.len(); + while left < right { + let mid = left + (right - left) / 2; + let Some(item) = self.items.get(mid) else { + break; + }; + if item.timestamp_ms() < target { + left = mid + 1; + } else { + right = mid; + } + } + left + } + + fn upper_bound(&self, target: i64) -> usize { + let mut left = 0usize; + let mut right = self.items.len(); + while left < right { + let mid = left + (right - left) / 2; + let Some(item) = self.items.get(mid) else { + break; + }; + if item.timestamp_ms() <= target { + left = mid + 1; + } else { + right = mid; + } + } + left + } + + fn trim(&mut self) { + while self.items.len() > self.max_capacity { + self.items.pop_front(); + } + } +} + +// ā”€ā”€ InputState ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ + +/// Character-indexed cursor input state, shared between agent guidance, +/// launcher input, and command bar. +#[derive(Debug, Clone, Default, PartialEq, Eq)] +pub struct InputState { + pub input: String, + pub cursor: usize, +} + +impl InputState { + pub fn insert_char(&mut self, ch: char) { + let idx = self.byte_index(self.cursor); + self.input.insert(idx, ch); + self.cursor = self.cursor.saturating_add(1); + } + + pub fn backspace(&mut self) { + if self.cursor == 0 { + return; + } + let start = self.byte_index(self.cursor - 1); + let end = self.byte_index(self.cursor); + if start < end { + self.input.replace_range(start..end, ""); + self.cursor = self.cursor.saturating_sub(1); + } + } + + pub fn delete(&mut self) { + let len = self.len_chars(); + if self.cursor >= len { + return; + } + let start = self.byte_index(self.cursor); + let end = self.byte_index(self.cursor + 1); + if start < end { + self.input.replace_range(start..end, ""); + } + } + + pub fn move_left(&mut self) { + if self.cursor > 0 { + self.cursor -= 1; + } + } + + pub fn move_right(&mut self) { + let len = self.len_chars(); + if self.cursor < len { + self.cursor += 1; + } + } + + pub fn move_home(&mut self) { + self.cursor = 0; + } + + pub fn move_end(&mut self) { + self.cursor = self.len_chars(); + } + + pub fn clear(&mut self) { + self.input.clear(); + self.cursor = 0; + } + + pub fn clamp_cursor(&mut self) { + let len = self.len_chars(); + if self.cursor > len { + self.cursor = len; + } + } + + fn len_chars(&self) -> usize { + self.input.chars().count() + } + + fn byte_index(&self, char_index: usize) -> usize { + if char_index == 0 { + return 0; + } + self.input + .char_indices() + .nth(char_index) + .map(|(idx, _)| idx) + .unwrap_or_else(|| self.input.len()) + } +} + +// ā”€ā”€ pane_block ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ + +/// Shared pane block with focus-dependent border styling. +pub fn pane_block<'a>(title: impl Into>, focused: bool) -> Block<'a> { + if focused { + Block::default() + .title(title) + .borders(Borders::ALL) + .border_type(BorderType::Double) + .border_style(theme::focus_border_style()) + } else { + Block::default() + .title(title) + .borders(Borders::ALL) + .border_style(theme::unfocus_border_style()) + } +} + +/// Builds a calm, centered empty-state card with optional detail + footer. +pub fn calm_empty_state<'a>( + title: impl Into>, + headline: &'a str, + detail: Option<&'a str>, + footer: Option<&'a str>, +) -> Paragraph<'a> { + let mut lines = Vec::new(); + lines.push(Line::from("")); + lines.push(Line::from(Span::styled(headline, theme::muted_style()))); + if let Some(detail) = detail { + lines.push(Line::from("")); + lines.push(Line::from(Span::styled(detail, theme::dim_style()))); + } + if let Some(footer) = footer { + lines.push(Line::from("")); + lines.push(Line::from(Span::styled(footer, theme::dim_style()))); + } + + Paragraph::new(lines) + .alignment(Alignment::Center) + .block(Block::default().borders(Borders::ALL).title(title)) +} + +#[cfg(test)] +mod tests { + use super::*; + + #[derive(Debug, Clone)] + struct Sample { + ts: i64, + label: &'static str, + } + + impl HasTimestamp for Sample { + fn timestamp_ms(&self) -> i64 { + self.ts + } + } + + #[test] + fn time_indexed_buffer_window_returns_expected_items() { + let mut buffer = TimeIndexedBuffer::new(10); + buffer.push_many([ + Sample { + ts: 100, + label: "a", + }, + Sample { + ts: 110, + label: "b", + }, + Sample { + ts: 120, + label: "c", + }, + Sample { + ts: 130, + label: "d", + }, + Sample { + ts: 140, + label: "e", + }, + ]); + + let window = buffer.window(130, 20); + let labels: Vec<&str> = window.iter().map(|item| item.label).collect(); + assert_eq!(labels, vec!["b", "c", "d"]); + } + + #[test] + fn time_indexed_buffer_trims_to_capacity_preserving_order() { + let mut buffer = TimeIndexedBuffer::new(3); + buffer.push_many([ + Sample { ts: 1, label: "a" }, + Sample { ts: 2, label: "b" }, + Sample { ts: 3, label: "c" }, + Sample { ts: 4, label: "d" }, + Sample { ts: 5, label: "e" }, + ]); + + let latest = buffer.latest(10); + let labels: Vec<&str> = latest.iter().map(|item| item.label).collect(); + assert_eq!(labels, vec!["c", "d", "e"]); + } + + #[test] + fn time_indexed_buffer_window_includes_equal_timestamps() { + let mut buffer = TimeIndexedBuffer::new(10); + buffer.push_many([ + Sample { + ts: 100, + label: "a", + }, + Sample { + ts: 100, + label: "b", + }, + Sample { + ts: 100, + label: "c", + }, + Sample { + ts: 110, + label: "d", + }, + ]); + + let window = buffer.window(100, 0); + let labels: Vec<&str> = window.iter().map(|item| item.label).collect(); + assert_eq!(labels, vec!["a", "b", "c"]); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/ui/theme.rs b/tui-rs/crates/zeroshot-tui/src/ui/theme.rs new file mode 100644 index 00000000..ebe7d48d --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/ui/theme.rs @@ -0,0 +1,190 @@ +//! Centralized color palette and style definitions. +//! +//! All render functions should import styles from here instead of using +//! inline `Style::default().fg(...)`. Based on Catppuccin Mocha palette. + +use ratatui::style::{Color, Modifier, Style}; + +use crate::app::SpineHintTone; + +// ā”€ā”€ Base palette ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ + +pub const ACCENT: Color = Color::Rgb(137, 180, 250); // #89b4fa blue +pub const ACCENT2: Color = Color::Rgb(166, 227, 161); // #a6e3a1 green +pub const FG_PRIMARY: Color = Color::Rgb(205, 214, 244); // #cdd6f4 +pub const FG_DIM: Color = Color::DarkGray; +pub const FG_MUTED: Color = Color::Rgb(108, 112, 134); // #6c7086 +pub const SURFACE: Color = Color::Rgb(30, 30, 46); // #1e1e2e +pub const FOCUS_BORDER: Color = ACCENT; +pub const UNFOCUS_BORDER: Color = Color::Rgb(69, 71, 90); // #45475a + +// ā”€ā”€ Status colors ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ + +pub const STATUS_RUNNING: Color = Color::Green; +pub const STATUS_DONE: Color = Color::Rgb(166, 227, 161); // #a6e3a1 +pub const STATUS_ERROR: Color = Color::Rgb(243, 139, 168); // #f38ba8 +pub const STATUS_PENDING: Color = Color::Yellow; +pub const STATUS_IDLE: Color = Color::DarkGray; + +// ā”€ā”€ Agent colors (rotating) ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ + +const AGENT_COLORS: [Color; 6] = [ + Color::Rgb(137, 180, 250), // blue + Color::Rgb(166, 227, 161), // green + Color::Rgb(249, 226, 175), // yellow #f9e2af + Color::Rgb(203, 166, 247), // mauve #cba6f7 + Color::Rgb(148, 226, 213), // teal #94e2d5 + Color::Rgb(242, 205, 205), // flamingo #f2cdcd +]; + +/// Get a color for an agent by hashing its ID to an index. +pub fn agent_color(agent_id: &str) -> Color { + let hash = agent_id + .bytes() + .fold(0u32, |acc, b| acc.wrapping_add(b as u32)); + AGENT_COLORS[hash as usize % AGENT_COLORS.len()] +} + +// ā”€ā”€ Pre-built styles ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ + +/// Logo / branding text. +pub fn logo_style() -> Style { + Style::default().fg(ACCENT).add_modifier(Modifier::BOLD) +} + +/// Screen title in the header. +pub fn title_style() -> Style { + Style::default().fg(FG_PRIMARY).add_modifier(Modifier::BOLD) +} + +/// Hint / secondary text. +pub fn dim_style() -> Style { + Style::default().fg(FG_DIM) +} + +/// Muted / disabled text. +pub fn muted_style() -> Style { + Style::default().fg(FG_MUTED) +} + +/// Keyboard shortcut key label (e.g., "Enter", "Esc"). +pub fn key_style() -> Style { + Style::default().fg(ACCENT).add_modifier(Modifier::BOLD) +} + +/// Keyboard shortcut description. +pub fn key_desc_style() -> Style { + Style::default().fg(FG_DIM) +} + +/// Spine mode label (Intent/Command/etc). +pub fn spine_mode_style() -> Style { + Style::default().fg(ACCENT).add_modifier(Modifier::BOLD) +} + +/// Spine input text. +pub fn spine_input_style() -> Style { + Style::default().fg(FG_PRIMARY) +} + +/// Spine placeholder text. +pub fn spine_placeholder_style() -> Style { + Style::default().fg(FG_MUTED) +} + +/// Spine completion text (ghost). +pub fn spine_completion_style() -> Style { + Style::default().fg(FG_DIM) +} + +/// Spine right-side hint text. +pub fn spine_hint_style() -> Style { + spine_hint_style_for(SpineHintTone::Muted) +} + +/// Spine hint style by tone. +pub fn spine_hint_style_for(tone: SpineHintTone) -> Style { + match tone { + SpineHintTone::Muted => Style::default().fg(FG_MUTED), + SpineHintTone::Info => Style::default().fg(ACCENT), + SpineHintTone::Success => Style::default().fg(ACCENT2), + SpineHintTone::Error => Style::default().fg(STATUS_ERROR), + } +} + +/// Spine command prefix. +pub fn spine_prefix_style() -> Style { + Style::default().fg(ACCENT).add_modifier(Modifier::BOLD) +} + +/// Focused pane border. +pub fn focus_border_style() -> Style { + Style::default() + .fg(FOCUS_BORDER) + .add_modifier(Modifier::BOLD) +} + +/// Unfocused pane border. +pub fn unfocus_border_style() -> Style { + Style::default().fg(UNFOCUS_BORDER) +} + +/// Spine border style. +pub fn spine_border_style() -> Style { + Style::default().fg(UNFOCUS_BORDER) +} + +/// Selected row in a list/table (accent bg, dark fg). +pub fn selected_style() -> Style { + Style::default() + .fg(SURFACE) + .bg(ACCENT) + .add_modifier(Modifier::BOLD) +} + +/// Backend status style by connection state. +pub fn backend_connected_style() -> Style { + Style::default().fg(STATUS_RUNNING) +} + +pub fn backend_error_style() -> Style { + Style::default().fg(STATUS_ERROR) +} + +/// Return a style for a cluster state string. +pub fn status_style(state: &str) -> Style { + match state { + "running" | "active" => Style::default().fg(STATUS_RUNNING), + "done" | "completed" | "complete" => Style::default().fg(STATUS_DONE), + "error" | "failed" => Style::default().fg(STATUS_ERROR), + "pending" | "starting" | "queued" => Style::default().fg(STATUS_PENDING), + "stopped" | "idle" => Style::default().fg(STATUS_IDLE), + _ => Style::default().fg(FG_DIM), + } +} + +/// Style for a "done" row (entire row dimmed). +pub fn done_row_style() -> Style { + Style::default().fg(FG_MUTED) +} + +/// Table header style. +pub fn table_header_style() -> Style { + Style::default() + .fg(FG_DIM) + .add_modifier(Modifier::BOLD) + .add_modifier(Modifier::UNDERLINED) +} + +/// Toast styles by level. +pub fn toast_success_style() -> Style { + Style::default().fg(ACCENT2) +} + +pub fn toast_error_style() -> Style { + Style::default().fg(STATUS_ERROR) +} + +pub fn toast_info_style() -> Style { + Style::default().fg(FG_DIM) +} diff --git a/tui-rs/crates/zeroshot-tui/src/ui/widgets/command_bar.rs b/tui-rs/crates/zeroshot-tui/src/ui/widgets/command_bar.rs new file mode 100644 index 00000000..2aa3d46a --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/ui/widgets/command_bar.rs @@ -0,0 +1,67 @@ +use ratatui::layout::{Position, Rect}; +use ratatui::text::{Line, Span}; +use ratatui::widgets::Paragraph; +use ratatui::Frame; + +use crate::app::CommandBarState; +use crate::ui::theme; + +/// Render the command bar as a single line (replaces status bar when active). +pub fn render(frame: &mut Frame<'_>, area: Rect, state: &CommandBarState, _allow_open: bool) { + if !state.active { + return; + } + + let line = Line::from(vec![ + Span::raw(" "), + Span::styled("/", theme::key_style()), + Span::styled(state.input(), theme::title_style()), + ]); + + let widget = Paragraph::new(line); + frame.render_widget(widget, area); +} + +/// Set cursor position for the command bar input. +pub fn set_cursor(frame: &mut Frame<'_>, area: Rect, state: &CommandBarState) { + if !state.active { + return; + } + if area.width <= 3 { + return; + } + + // Offset: 1 (padding) + 1 (/) + cursor position + let max_x = area.x + area.width.saturating_sub(1); + let cursor_x = area.x + 2 + state.cursor() as u16; + let cursor_x = cursor_x.min(max_x); + frame.set_cursor_position(Position::new(cursor_x, area.y)); +} + +#[cfg(test)] +mod tests { + use super::*; + use ratatui::backend::TestBackend; + use ratatui::Terminal; + + use crate::ui::widgets::test_utils::line_text; + + #[test] + fn active_command_bar_renders_input() { + let backend = TestBackend::new(40, 1); + let mut terminal = Terminal::new(backend).expect("terminal"); + let mut state = CommandBarState::default(); + state.open_with("help".to_string()); + + terminal + .draw(|frame| { + let area = frame.area(); + render(frame, area, &state, true); + }) + .expect("draw"); + + let buffer = terminal.backend().buffer(); + let content = line_text(buffer, 0); + assert!(content.contains("/help")); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/ui/widgets/mod.rs b/tui-rs/crates/zeroshot-tui/src/ui/widgets/mod.rs new file mode 100644 index 00000000..eea3df6f --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/ui/widgets/mod.rs @@ -0,0 +1,9 @@ +pub mod command_bar; +pub mod scrub_bar; +pub mod spine; +pub mod stream; +pub mod toast; +pub mod topology; + +#[cfg(test)] +pub(crate) mod test_utils; diff --git a/tui-rs/crates/zeroshot-tui/src/ui/widgets/scrub_bar.rs b/tui-rs/crates/zeroshot-tui/src/ui/widgets/scrub_bar.rs new file mode 100644 index 00000000..3607df50 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/ui/widgets/scrub_bar.rs @@ -0,0 +1,244 @@ +use ratatui::layout::Rect; +use ratatui::text::{Line, Span}; +use ratatui::widgets::Paragraph; +use ratatui::Frame; + +use crate::app::{TimeCursor, TimeCursorMode}; +use crate::protocol::ClusterLogLine; +use crate::ui::shared::TimeIndexedBuffer; +use crate::ui::theme; + +const DENSITY_LEVELS: &[u8] = b" .:-=+*#"; + +pub struct ScrubBarState<'a> { + pub time_cursor: &'a TimeCursor, + pub logs: Option<&'a TimeIndexedBuffer>, + pub agent_id: Option<&'a str>, +} + +pub fn render(frame: &mut Frame<'_>, area: Rect, state: ScrubBarState<'_>) { + if area.width == 0 || area.height == 0 { + return; + } + + let label = match state.time_cursor.mode { + TimeCursorMode::Live => "LIVE", + TimeCursorMode::Scrub => "SCRUB", + }; + let label_style = match state.time_cursor.mode { + TimeCursorMode::Live => theme::toast_success_style(), + TimeCursorMode::Scrub => theme::key_style(), + }; + + let label_len = label.chars().count() as u16; + let mut spans = Vec::new(); + spans.push(Span::styled(label, label_style)); + + let bar_width = area.width.saturating_sub(label_len.saturating_add(1)) as usize; + if bar_width > 0 { + spans.push(Span::raw(" ")); + let bar = build_bar(bar_width, &state); + spans.push(Span::styled(bar, theme::dim_style())); + } + + let widget = Paragraph::new(Line::from(spans)); + frame.render_widget(widget, area); +} + +fn build_bar(width: usize, state: &ScrubBarState<'_>) -> String { + if width == 0 { + return String::new(); + } + + let window_ms = state.time_cursor.window_ms.max(1); + let window_end = state.time_cursor.t_ms; + let window_start = window_end.saturating_sub(window_ms); + + let latest_ts = state.logs.and_then(|logs| { + logs.iter() + .filter(|line| matches_agent(line, state.agent_id)) + .map(|line| line.timestamp) + .max() + }); + + let mut bins = vec![0u32; width]; + if let Some(logs) = state.logs { + let windowed = logs.window(window_end, window_ms); + for line in windowed { + if !matches_agent(line, state.agent_id) { + continue; + } + let rel = line.timestamp.saturating_sub(window_start); + let mut pos = ((rel * width as i64) / window_ms) as usize; + if pos >= width { + pos = width - 1; + } + bins[pos] = bins[pos].saturating_add(1); + } + } + + let max = bins.iter().copied().max().unwrap_or(0); + let mut chars: Vec = bins + .into_iter() + .map(|count| { + if max == 0 { + ' ' + } else { + let idx = (count as usize * (DENSITY_LEVELS.len() - 1)) / max as usize; + DENSITY_LEVELS[idx] as char + } + }) + .collect(); + + let now_pos = latest_ts.map_or(width.saturating_sub(1), |latest| { + let rel = latest.saturating_sub(window_start); + let mut pos = ((rel * width as i64) / window_ms) as usize; + if pos >= width { + pos = width - 1; + } + pos + }); + if !chars.is_empty() { + chars[now_pos] = '|'; + } + + if matches!(state.time_cursor.mode, TimeCursorMode::Scrub) && !chars.is_empty() { + let rel = state.time_cursor.t_ms.saturating_sub(window_start); + let mut pos = ((rel * width as i64) / window_ms) as usize; + if pos >= width { + pos = width - 1; + } + if pos == now_pos { + chars[pos] = '*'; + } else { + chars[pos] = '^'; + } + } + + chars.into_iter().collect() +} + +fn matches_agent(line: &ClusterLogLine, agent_id: Option<&str>) -> bool { + let Some(agent_id) = agent_id else { + return true; + }; + line.agent.as_deref() == Some(agent_id) || line.sender.as_deref() == Some(agent_id) +} + +#[cfg(test)] +mod tests { + use super::*; + use ratatui::backend::TestBackend; + use ratatui::Terminal; + + use crate::ui::widgets::test_utils::line_text; + + fn sample_logs(timestamps: &[i64]) -> TimeIndexedBuffer { + let mut buffer = TimeIndexedBuffer::new(64); + let lines = timestamps.iter().map(|ts| ClusterLogLine { + id: format!("log-{ts}"), + timestamp: *ts, + text: "event".to_string(), + agent: None, + role: None, + sender: None, + }); + buffer.push_many(lines); + buffer + } + + #[test] + fn scrub_bar_renders_live_mode() { + let logs = sample_logs(&[100, 200, 300]); + let cursor = TimeCursor { + mode: TimeCursorMode::Live, + t_ms: 300, + window_ms: 300, + }; + + let backend = TestBackend::new(40, 1); + let mut terminal = Terminal::new(backend).expect("terminal"); + terminal + .draw(|frame| { + let area = frame.area(); + render( + frame, + area, + ScrubBarState { + time_cursor: &cursor, + logs: Some(&logs), + agent_id: None, + }, + ); + }) + .expect("draw"); + + let buffer = terminal.backend().buffer(); + let text = line_text(buffer, 0); + assert!(text.contains("LIVE")); + assert!(text.contains("|")); + } + + #[test] + fn scrub_bar_renders_scrub_mode_marker() { + let logs = sample_logs(&[100, 200, 300]); + let cursor = TimeCursor { + mode: TimeCursorMode::Scrub, + t_ms: 150, + window_ms: 300, + }; + + let backend = TestBackend::new(40, 1); + let mut terminal = Terminal::new(backend).expect("terminal"); + terminal + .draw(|frame| { + let area = frame.area(); + render( + frame, + area, + ScrubBarState { + time_cursor: &cursor, + logs: Some(&logs), + agent_id: None, + }, + ); + }) + .expect("draw"); + + let buffer = terminal.backend().buffer(); + let text = line_text(buffer, 0); + assert!(text.contains("SCRUB")); + assert!(text.contains("*")); + } + + #[test] + fn scrub_bar_handles_empty_buffers() { + let logs = sample_logs(&[]); + let cursor = TimeCursor { + mode: TimeCursorMode::Live, + t_ms: 0, + window_ms: 500, + }; + + let backend = TestBackend::new(24, 1); + let mut terminal = Terminal::new(backend).expect("terminal"); + terminal + .draw(|frame| { + let area = frame.area(); + render( + frame, + area, + ScrubBarState { + time_cursor: &cursor, + logs: Some(&logs), + agent_id: None, + }, + ); + }) + .expect("draw"); + + let buffer = terminal.backend().buffer(); + let text = line_text(buffer, 0); + assert!(text.contains("LIVE")); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/ui/widgets/spine.rs b/tui-rs/crates/zeroshot-tui/src/ui/widgets/spine.rs new file mode 100644 index 00000000..61921c32 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/ui/widgets/spine.rs @@ -0,0 +1,290 @@ +use ratatui::layout::{Alignment, Position, Rect}; +use ratatui::text::{Line, Span}; +use ratatui::widgets::{Block, Borders, Paragraph}; +use ratatui::Frame; + +use crate::app::{SpineHintTone, SpineMode, SpineState}; +use crate::ui::theme; + +const PLACEHOLDER_INTENT: &str = "Type intent..."; + +pub fn render(frame: &mut Frame<'_>, area: Rect, state: &SpineState) { + let block = spine_block(); + let inner = block.inner(area); + + let hint_text = state.hint.text.as_str(); + let hint_tone = state.hint.tone; + + let mut spans = build_spans(state); + let mut lines = Vec::new(); + let mut hint_on_first_line = false; + if hint_fits(&spans, inner.width, hint_text) { + append_hint(&mut spans, inner.width, hint_text, hint_tone); + hint_on_first_line = true; + } + lines.push(Line::from(spans)); + + if !hint_on_first_line && !hint_text.is_empty() && inner.height >= 2 { + lines.push(build_hint_line(inner.width, hint_text, hint_tone)); + } + + let widget = Paragraph::new(lines) + .block(block) + .alignment(Alignment::Left); + frame.render_widget(widget, area); +} + +pub fn set_cursor(frame: &mut Frame<'_>, area: Rect, state: &SpineState) { + let block = spine_block(); + let inner = block.inner(area); + if inner.width == 0 || inner.height == 0 { + return; + } + + let label_len = mode_label(state.mode).len() as u16; + let prefix_len = command_prefix(state.mode).len() as u16; + let base_x = inner.x + 1 + label_len + 1 + prefix_len; + let cursor_x = base_x.saturating_add(state.input.cursor as u16); + let max_x = inner.x + inner.width.saturating_sub(1); + let cursor_x = cursor_x.min(max_x); + + frame.set_cursor_position(Position::new(cursor_x, inner.y)); +} + +fn spine_block<'a>() -> Block<'a> { + Block::default() + .borders(Borders::TOP) + .border_style(theme::spine_border_style()) +} + +fn mode_label(mode: SpineMode) -> &'static str { + match mode { + SpineMode::Intent => "Intent", + SpineMode::Command => "Command", + SpineMode::WhisperCluster => "Whisper Cluster", + SpineMode::WhisperAgent => "Whisper Agent", + } +} + +fn command_prefix(mode: SpineMode) -> &'static str { + match mode { + SpineMode::Command => "/", + _ => "", + } +} + +fn build_spans<'a>(state: &'a SpineState) -> Vec> { + let mut spans = Vec::new(); + push_mode_label(&mut spans, state.mode); + push_prefix(&mut spans, state.mode); + push_input_or_placeholder(&mut spans, state); + push_completion(&mut spans, state); + spans +} + +fn push_mode_label<'a>(spans: &mut Vec>, mode: SpineMode) { + spans.push(Span::raw(" ")); + spans.push(Span::styled(mode_label(mode), theme::spine_mode_style())); + spans.push(Span::raw(" ")); +} + +fn push_prefix<'a>(spans: &mut Vec>, mode: SpineMode) { + let prefix = command_prefix(mode); + if !prefix.is_empty() { + spans.push(Span::styled(prefix, theme::spine_prefix_style())); + } +} + +fn push_input_or_placeholder<'a>(spans: &mut Vec>, state: &'a SpineState) { + if state.input.input.is_empty() { + if matches!(state.mode, SpineMode::Intent) { + spans.push(Span::styled( + PLACEHOLDER_INTENT, + theme::spine_placeholder_style(), + )); + } + return; + } + + spans.push(Span::styled( + state.input.input.as_str(), + theme::spine_input_style(), + )); +} + +fn push_completion<'a>(spans: &mut Vec>, state: &'a SpineState) { + let Some(completion) = &state.completion else { + return; + }; + if completion.ghost.is_empty() || state.input.input.is_empty() { + return; + } + spans.push(Span::styled( + completion.ghost.as_str(), + theme::spine_completion_style(), + )); +} + +fn hint_fits(spans: &[Span<'_>], width: u16, hint: &str) -> bool { + if hint.is_empty() || width == 0 { + return false; + } + let used_len: usize = spans.iter().map(|span| span.content.len()).sum(); + let hint_len = hint.len(); + let width = width as usize; + width > used_len + hint_len + 1 +} + +fn append_hint<'a>(spans: &mut Vec>, width: u16, hint: &'a str, tone: SpineHintTone) { + if hint.is_empty() || width == 0 { + return; + } + let used_len: usize = spans.iter().map(|span| span.content.len()).sum(); + let hint_len = hint.len(); + let width = width as usize; + if width <= used_len + hint_len + 1 { + return; + } + let gap = width - used_len - hint_len; + spans.push(Span::raw(" ".repeat(gap))); + spans.push(Span::styled(hint, theme::spine_hint_style_for(tone))); +} + +fn build_hint_line<'a>(width: u16, hint: &'a str, tone: SpineHintTone) -> Line<'a> { + if width == 0 { + return Line::from(Span::raw("")); + } + let width = width as usize; + let hint_len = hint.len(); + if hint_len >= width { + return Line::from(Span::styled(hint, theme::spine_hint_style_for(tone))); + } + let gap = width - hint_len; + Line::from(vec![ + Span::raw(" ".repeat(gap)), + Span::styled(hint, theme::spine_hint_style_for(tone)), + ]) +} + +#[cfg(test)] +mod tests { + use super::*; + use ratatui::backend::TestBackend; + use ratatui::layout::Position; + use ratatui::Terminal; + + use crate::ui::widgets::test_utils::line_text; + + #[test] + fn intent_mode_shows_placeholder() { + let backend = TestBackend::new(40, 3); + let mut terminal = Terminal::new(backend).expect("terminal"); + let state = SpineState::default(); + + terminal + .draw(|frame| { + let area = frame.area(); + render(frame, area, &state); + }) + .expect("draw"); + + let buffer = terminal.backend().buffer(); + let content = line_text(buffer, 1); + assert!(content.contains(PLACEHOLDER_INTENT)); + } + + #[test] + fn command_mode_prefix_is_rendered() { + let backend = TestBackend::new(40, 3); + let mut terminal = Terminal::new(backend).expect("terminal"); + let mut state = SpineState::default(); + state.mode = SpineMode::Command; + state.input.input = "help".to_string(); + + terminal + .draw(|frame| { + let area = frame.area(); + render(frame, area, &state); + }) + .expect("draw"); + + let buffer = terminal.backend().buffer(); + let content = line_text(buffer, 1); + assert!(content.contains("/help")); + } + + #[test] + fn spine_cursor_follows_input() { + let backend = TestBackend::new(40, 3); + let mut terminal = Terminal::new(backend).expect("terminal"); + let mut state = SpineState::default(); + state.input.input = "abcd".to_string(); + state.input.cursor = 2; + + terminal + .draw(|frame| { + let area = frame.area(); + render(frame, area, &state); + set_cursor(frame, area, &state); + }) + .expect("draw"); + + let label_len = mode_label(state.mode).len() as u16; + let base_x = 1 + label_len + 1; + let expected_x = base_x + state.input.cursor as u16; + terminal + .backend_mut() + .assert_cursor_position(Position::new(expected_x, 1)); + } + + #[test] + fn hint_moves_to_second_line_when_no_space() { + let backend = TestBackend::new(22, 4); + let mut terminal = Terminal::new(backend).expect("terminal"); + let mut state = SpineState::default(); + state.input.input = "extremelylonginput".to_string(); + state.hint = crate::app::SpineHint::new("Second line hint", SpineHintTone::Info); + + terminal + .draw(|frame| { + let area = frame.area(); + render(frame, area, &state); + }) + .expect("draw"); + + let buffer = terminal.backend().buffer(); + let line1 = line_text(buffer, 1); + let line2 = line_text(buffer, 2); + assert!(!line1.contains("Second line hint")); + assert!(line2.contains("Second line hint")); + } + + #[test] + fn completion_renders_dimmed_after_input() { + let backend = TestBackend::new(40, 3); + let mut terminal = Terminal::new(backend).expect("terminal"); + let mut state = SpineState::default(); + state.mode = SpineMode::Command; + state.input.input = "pro".to_string(); + state.input.cursor = 3; + state.completion = Some(crate::app::SpineCompletion { + candidates: vec!["provider".to_string()], + selected: 0, + ghost: "vider".to_string(), + }); + + terminal + .draw(|frame| { + let area = frame.area(); + render(frame, area, &state); + }) + .expect("draw"); + + let buffer = terminal.backend().buffer(); + let line = line_text(buffer, 1); + let ghost_start = line.find("vider").expect("ghost text"); + let cell = buffer.cell((ghost_start as u16, 1)).expect("ghost cell"); + let expected_fg = theme::spine_completion_style().fg.expect("fg"); + assert_eq!(cell.fg, expected_fg); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/ui/widgets/stream.rs b/tui-rs/crates/zeroshot-tui/src/ui/widgets/stream.rs new file mode 100644 index 00000000..9527eb4e --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/ui/widgets/stream.rs @@ -0,0 +1,479 @@ +use ratatui::buffer::Buffer; +use ratatui::layout::Rect; +use ratatui::style::Style; +use ratatui::text::{Line, Span}; +use ratatui::widgets::{Block, Borders, Paragraph, Widget, Wrap}; + +use crate::app::{TimeCursor, TimeCursorMode}; +use crate::protocol::{ClusterLogLine, TimelineEvent}; +use crate::ui::shared::{HasTimestamp, TimeIndexedBuffer}; +use crate::ui::theme; + +pub const PHASE_MARKER_LIMIT: usize = 50; + +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub enum LogPlaceholderContext { + Cluster, + Agent, + Overlay, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct PhaseMarker { + pub timestamp_ms: i64, + pub label: String, + pub topic: String, +} + +pub struct StreamOverlay<'a> { + title: Line<'a>, + lines: Vec>, + placeholder: Vec>, + border_style: Style, +} + +impl<'a> StreamOverlay<'a> { + pub fn new(title: impl Into>, lines: Vec>) -> Self { + Self { + title: title.into(), + lines, + placeholder: Vec::new(), + border_style: theme::unfocus_border_style(), + } + } + + pub fn placeholder_lines(mut self, lines: Vec>) -> Self { + self.placeholder = lines; + self + } + + pub fn border_style(mut self, style: Style) -> Self { + self.border_style = style; + self + } +} + +impl<'a> Widget for StreamOverlay<'a> { + fn render(self, area: Rect, buf: &mut Buffer) { + if area.width == 0 || area.height == 0 { + return; + } + + let block = Block::default() + .title(self.title) + .borders(Borders::ALL) + .border_style(self.border_style); + let inner = block.inner(area); + block.render(area, buf); + + if inner.width == 0 || inner.height == 0 { + return; + } + + let mut lines = if self.lines.is_empty() { + self.placeholder + } else { + self.lines + }; + let max_lines = inner.height as usize; + if lines.len() > max_lines { + lines.truncate(max_lines); + } + + let paragraph = Paragraph::new(lines).wrap(Wrap { trim: false }); + paragraph.render(inner, buf); + } +} + +pub fn log_placeholder_lines<'a>(context: LogPlaceholderContext) -> Vec> { + let detail = match context { + LogPlaceholderContext::Cluster => "Waiting for cluster output.", + LogPlaceholderContext::Agent => "Waiting for agent output.", + LogPlaceholderContext::Overlay => "Waiting for stream output.", + }; + vec![ + Line::from(Span::styled("No logs yet.", theme::muted_style())), + Line::from(Span::styled(detail, theme::muted_style())), + ] +} + +pub fn timeline_placeholder_lines<'a>() -> Vec> { + vec![ + Line::from(Span::styled( + "No timeline events yet.", + theme::muted_style(), + )), + Line::from(Span::styled( + "New activity will appear here.", + theme::muted_style(), + )), + ] +} + +pub fn mode_tag_span(time_cursor: &TimeCursor) -> Span<'static> { + let (label, style) = match time_cursor.mode { + TimeCursorMode::Live => ("LIVE", theme::toast_success_style()), + TimeCursorMode::Scrub => ("SCRUB", theme::key_style()), + }; + Span::styled(format!("[{label}]"), style) +} + +pub fn derive_phase_markers( + timeline: &TimeIndexedBuffer, + time_cursor: &TimeCursor, + max_markers: usize, +) -> Vec { + if max_markers == 0 || timeline.is_empty() { + return Vec::new(); + } + + let max_items = timeline.len(); + let events = select_time_window(timeline, time_cursor, max_items, |_| true); + if events.is_empty() { + return Vec::new(); + } + + let mut markers = Vec::new(); + let mut last_topic = String::new(); + let mut last_label = String::new(); + let mut has_last = false; + for event in events { + if has_last && last_topic == event.topic && last_label == event.label { + continue; + } + markers.push(PhaseMarker { + timestamp_ms: event.timestamp, + label: event.label.clone(), + topic: event.topic.clone(), + }); + last_topic = event.topic.clone(); + last_label = event.label.clone(); + has_last = true; + } + + if markers.len() > max_markers { + let start = markers.len().saturating_sub(max_markers); + markers = markers.split_off(start); + } + markers +} + +pub fn format_phase_marker_label(topic: &str, label: &str) -> String { + if label.is_empty() { + topic.to_string() + } else { + format!("{topic}: {label}") + } +} + +pub fn truncate_marker_label(label: &str, max_chars: usize) -> String { + if max_chars == 0 { + return String::new(); + } + let label_len = label.chars().count(); + if label_len <= max_chars { + return label.to_string(); + } + if max_chars <= 3 { + return label.chars().take(max_chars).collect(); + } + let mut out: String = label.chars().take(max_chars - 3).collect(); + out.push_str("..."); + out +} + +pub fn overlay_title(base: impl Into, time_cursor: &TimeCursor) -> Line<'static> { + Line::from(vec![ + Span::raw(base.into()), + Span::raw(" "), + mode_tag_span(time_cursor), + ]) +} + +pub fn select_time_window<'a, T, F>( + buffer: &'a TimeIndexedBuffer, + time_cursor: &TimeCursor, + max_items: usize, + filter: F, +) -> Vec<&'a T> +where + T: HasTimestamp, + F: Fn(&T) -> bool, +{ + if max_items == 0 || buffer.is_empty() { + return Vec::new(); + } + + match time_cursor.mode { + TimeCursorMode::Live => select_live_tail(buffer, max_items, filter), + TimeCursorMode::Scrub => select_scrub_window(buffer, time_cursor, max_items, filter), + } +} + +fn select_live_tail(buffer: &TimeIndexedBuffer, max_items: usize, filter: F) -> Vec<&T> +where + T: HasTimestamp, + F: Fn(&T) -> bool, +{ + let mut collected = Vec::with_capacity(max_items); + for item in buffer.iter_rev() { + if filter(item) { + collected.push(item); + if collected.len() >= max_items { + break; + } + } + } + collected.reverse(); + collected +} + +fn select_scrub_window<'a, T, F>( + buffer: &'a TimeIndexedBuffer, + time_cursor: &TimeCursor, + max_items: usize, + filter: F, +) -> Vec<&'a T> +where + T: HasTimestamp, + F: Fn(&T) -> bool, +{ + let windowed = buffer.window(time_cursor.t_ms, time_cursor.window_ms); + let mut collected: Vec<&T> = windowed.into_iter().filter(|item| filter(item)).collect(); + if collected.len() > max_items { + let start = collected.len().saturating_sub(max_items); + collected = collected.split_off(start); + } + collected +} + +pub fn format_log_line_styled(line: &ClusterLogLine) -> Line<'_> { + if let Some(agent) = line.agent.as_deref().or(line.sender.as_deref()) { + let color = theme::agent_color(agent); + Line::from(vec![ + Span::styled(format!("[{agent}]"), Style::default().fg(color)), + Span::raw(" "), + Span::raw(line.text.as_str()), + ]) + } else { + Line::from(line.text.as_str()) + } +} + +pub fn format_timeline_event_styled(event: &TimelineEvent) -> Line<'_> { + let icon = timeline_icon(&event.topic); + let label_style = timeline_label_style(&event.label); + let mut spans = vec![ + Span::styled(icon, theme::dim_style()), + Span::raw(" "), + Span::styled(event.topic.as_str(), theme::dim_style()), + Span::raw(" "), + Span::styled(event.label.as_str(), label_style), + ]; + if let Some(sender) = event.sender.as_deref() { + spans.push(Span::raw(" ")); + spans.push(Span::styled(format!("({sender})"), theme::muted_style())); + } + Line::from(spans) +} + +fn timeline_icon(topic: &str) -> &'static str { + let topic_lower = topic.to_lowercase(); + if topic_lower.contains("issue") { + "\u{25b6}" // ā–¶ + } else if topic_lower.contains("implementation") || topic_lower.contains("impl") { + "\u{25cf}" // ā— + } else if topic_lower.contains("validation") || topic_lower.contains("review") { + "\u{25c6}" // ā—† + } else if topic_lower.contains("consensus") || topic_lower.contains("complete") { + "\u{2605}" // ā˜… + } else { + "\u{00b7}" // Ā· + } +} + +fn timeline_label_style(label: &str) -> Style { + let label_lower = label.to_lowercase(); + if label_lower.contains("approved") + || label_lower.contains("done") + || label_lower.contains("complete") + { + theme::status_style("done") + } else if label_lower.contains("rejected") + || label_lower.contains("failed") + || label_lower.contains("error") + { + theme::status_style("error") + } else if label_lower.contains("pending") || label_lower.contains("waiting") { + theme::status_style("pending") + } else { + Style::default() + } +} + +#[cfg(test)] +mod tests { + use super::*; + use ratatui::backend::TestBackend; + use ratatui::Terminal; + + use crate::ui::widgets::test_utils::line_text; + + fn buffer_contains(terminal: &Terminal, needle: &str) -> bool { + let buffer = terminal.backend().buffer(); + for y in 0..buffer.area.height { + if line_text(buffer, y).contains(needle) { + return true; + } + } + false + } + + #[test] + fn stream_overlay_renders_title_and_lines() { + let backend = TestBackend::new(32, 8); + let mut terminal = Terminal::new(backend).expect("terminal"); + + terminal + .draw(|frame| { + let area = frame.area(); + let overlay = StreamOverlay::new( + Line::from("Logs - agent alpha"), + vec![Line::from("hello"), Line::from("world")], + ); + frame.render_widget(overlay, area); + }) + .expect("draw"); + + assert!(buffer_contains(&terminal, "Logs - agent alpha")); + assert!(buffer_contains(&terminal, "hello")); + assert!(buffer_contains(&terminal, "world")); + } + + #[test] + fn stream_overlay_renders_empty_placeholder() { + let backend = TestBackend::new(30, 7); + let mut terminal = Terminal::new(backend).expect("terminal"); + + terminal + .draw(|frame| { + let area = frame.area(); + let overlay = StreamOverlay::new(Line::from("Logs"), Vec::new()) + .placeholder_lines(log_placeholder_lines(LogPlaceholderContext::Overlay)); + frame.render_widget(overlay, area); + }) + .expect("draw"); + + assert!(buffer_contains(&terminal, "No logs yet.")); + assert!(buffer_contains(&terminal, "Waiting for stream output.")); + } + + fn sample_log(id: &str, timestamp: i64, agent: Option<&str>) -> ClusterLogLine { + ClusterLogLine { + id: id.to_string(), + timestamp, + text: format!("log-{id}"), + agent: agent.map(|value| value.to_string()), + role: None, + sender: None, + } + } + + fn sample_event(id: &str, timestamp: i64, topic: &str, label: &str) -> TimelineEvent { + TimelineEvent { + id: id.to_string(), + timestamp, + topic: topic.to_string(), + label: label.to_string(), + approved: None, + sender: None, + } + } + + #[test] + fn stream_window_live_uses_tail() { + let mut buffer = TimeIndexedBuffer::new(16); + buffer.push_many(vec![ + sample_log("one", 100, Some("alpha")), + sample_log("two", 200, Some("alpha")), + sample_log("three", 300, Some("alpha")), + ]); + + let cursor = TimeCursor::default(); + let selected = select_time_window(&buffer, &cursor, 2, |_| true); + let ids: Vec<&str> = selected.iter().map(|line| line.id.as_str()).collect(); + assert_eq!(ids, vec!["two", "three"]); + } + + #[test] + fn stream_window_scrub_uses_window() { + let mut buffer = TimeIndexedBuffer::new(16); + buffer.push_many(vec![ + sample_log("one", 100, Some("alpha")), + sample_log("two", 200, Some("alpha")), + sample_log("three", 300, Some("alpha")), + ]); + + let cursor = TimeCursor { + mode: TimeCursorMode::Scrub, + t_ms: 250, + window_ms: 120, + }; + let selected = select_time_window(&buffer, &cursor, 10, |_| true); + let ids: Vec<&str> = selected.iter().map(|line| line.id.as_str()).collect(); + assert_eq!(ids, vec!["two"]); + } + + #[test] + fn stream_overlay_renders_mode_tag() { + let backend = TestBackend::new(40, 6); + let mut terminal = Terminal::new(backend).expect("terminal"); + let cursor = TimeCursor::default(); + + terminal + .draw(|frame| { + let area = frame.area(); + let overlay = StreamOverlay::new(overlay_title("Logs", &cursor), Vec::new()) + .placeholder_lines(log_placeholder_lines(LogPlaceholderContext::Overlay)); + frame.render_widget(overlay, area); + }) + .expect("draw"); + + assert!(buffer_contains(&terminal, "LIVE")); + } + + #[test] + fn derive_phase_markers_caps_and_dedup() { + let mut buffer = TimeIndexedBuffer::new(128); + buffer.push_many(vec![ + sample_event("e1", 100, "topic-a", "phase-1"), + sample_event("e2", 110, "topic-a", "phase-1"), + sample_event("e3", 120, "topic-b", "phase-2"), + sample_event("e4", 130, "topic-b", "phase-2"), + sample_event("e5", 140, "topic-b", "phase-3"), + ]); + + let cursor = TimeCursor::default(); + let markers = derive_phase_markers(&buffer, &cursor, PHASE_MARKER_LIMIT); + assert_eq!(markers.len(), 3); + assert_eq!(markers[0].topic, "topic-a"); + assert_eq!(markers[1].label, "phase-2"); + assert_eq!(markers[2].label, "phase-3"); + + let mut buffer = TimeIndexedBuffer::new(128); + let mut events = Vec::new(); + for idx in 0..60 { + events.push(sample_event( + &format!("cap-{idx}"), + 1000 + idx as i64, + &format!("topic-{idx}"), + "phase", + )); + } + buffer.push_many(events); + let markers = derive_phase_markers(&buffer, &cursor, 50); + assert_eq!(markers.len(), 50); + assert_eq!(markers.first().unwrap().topic, "topic-10"); + assert_eq!(markers.last().unwrap().topic, "topic-59"); + } +} diff --git a/tui-rs/crates/zeroshot-tui/src/ui/widgets/test_utils.rs b/tui-rs/crates/zeroshot-tui/src/ui/widgets/test_utils.rs new file mode 100644 index 00000000..21599062 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/ui/widgets/test_utils.rs @@ -0,0 +1,10 @@ +use ratatui::buffer::Buffer; + +pub fn line_text(buffer: &Buffer, y: u16) -> String { + let area = buffer.area; + let mut line = String::new(); + for x in area.left()..area.right() { + line.push_str(buffer.cell((x, y)).map_or("", |c| c.symbol())); + } + line +} diff --git a/tui-rs/crates/zeroshot-tui/src/ui/widgets/toast.rs b/tui-rs/crates/zeroshot-tui/src/ui/widgets/toast.rs new file mode 100644 index 00000000..93697bca --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/ui/widgets/toast.rs @@ -0,0 +1,32 @@ +use ratatui::style::Style; + +use crate::app::{ToastLevel, ToastState}; +use crate::ui::theme; + +/// Format toast for inline display in the status bar. +/// Returns (text, style) or None if no toast. +pub fn format_inline(toast: Option<&ToastState>) -> Option<(String, Style)> { + let toast = toast?; + let (prefix, style) = match toast.level { + ToastLevel::Info => ("\u{2139}", theme::toast_info_style()), // ā„¹ + ToastLevel::Success => ("\u{2713}", theme::toast_success_style()), // āœ“ + ToastLevel::Error => ("\u{2717}", theme::toast_error_style()), // āœ— + }; + let first_line = toast.message.lines().next().unwrap_or(""); + let msg = format!("{prefix} {}", truncate_toast_line(first_line)); + Some((msg, style)) +} + +fn truncate_toast_line(line: &str) -> String { + const MAX_LEN: usize = 40; + const TRUNC_LEN: usize = 37; + if line.chars().count() <= MAX_LEN { + return line.to_string(); + } + let mut out = String::new(); + for ch in line.chars().take(TRUNC_LEN) { + out.push(ch); + } + out.push_str("..."); + out +} diff --git a/tui-rs/crates/zeroshot-tui/src/ui/widgets/topology.rs b/tui-rs/crates/zeroshot-tui/src/ui/widgets/topology.rs new file mode 100644 index 00000000..7eb20719 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/src/ui/widgets/topology.rs @@ -0,0 +1,161 @@ +use ratatui::layout::Rect; +use ratatui::text::Line; +use ratatui::widgets::{Block, Paragraph, Wrap}; +use ratatui::Frame; + +use crate::protocol::{ + ClusterSummary, ClusterTopology, TopologyAgent, TopologyEdge, TopologyEdgeKind, +}; + +pub fn render( + frame: &mut Frame<'_>, + area: Rect, + block: Block<'_>, + summary: Option<&ClusterSummary>, + topology: Option<&ClusterTopology>, + error: Option<&str>, +) { + let lines = build_lines(summary, topology, error) + .into_iter() + .map(Line::from) + .collect::>(); + let widget = Paragraph::new(lines) + .block(block) + .wrap(Wrap { trim: false }); + frame.render_widget(widget, area); +} + +pub fn build_lines( + summary: Option<&ClusterSummary>, + topology: Option<&ClusterTopology>, + error: Option<&str>, +) -> Vec { + let mut lines = Vec::new(); + lines.push(summary_line(summary)); + + if let Some(message) = error { + append_error(&mut lines, message); + return lines; + } + + let Some(topology) = topology else { + append_pending(&mut lines); + return lines; + }; + + let (agents, topics, edges) = sorted_topology(topology); + append_counts(&mut lines, &agents, &topics, &edges); + + if edges.is_empty() { + append_no_edges(&mut lines, &agents, &topics); + return lines; + } + + append_edges(&mut lines, edges); + append_focus_hint(&mut lines); + lines +} + +fn append_error(lines: &mut Vec, message: &str) { + lines.push(format!("Topology unavailable: {message}")); + append_focus_hint(lines); +} + +fn append_pending(lines: &mut Vec) { + lines.push("Topology pending. Waiting for backend.".to_string()); + append_focus_hint(lines); +} + +fn append_counts( + lines: &mut Vec, + agents: &[TopologyAgent], + topics: &[String], + edges: &[TopologyEdge], +) { + lines.push(format!( + "Agents: {} | Topics: {} | Edges: {}", + agents.len(), + topics.len(), + edges.len() + )); +} + +fn append_no_edges(lines: &mut Vec, agents: &[TopologyAgent], topics: &[String]) { + if !agents.is_empty() { + let list = agents + .iter() + .map(|agent| agent.id.as_str()) + .collect::>() + .join(", "); + lines.push(format!("Agents: {list}")); + } + if !topics.is_empty() { + lines.push(format!("Topics: {}", topics.join(", "))); + } + lines.push("No edges yet.".to_string()); + append_focus_hint(lines); +} + +fn append_edges(lines: &mut Vec, edges: Vec) { + let mut current_from: Option = None; + for edge in edges { + if current_from.as_deref() != Some(edge.from.as_str()) { + current_from = Some(edge.from.clone()); + lines.push(format!("{}:", edge.from)); + } + lines.push(format!(" -> {}", edge_details(&edge))); + } +} + +fn append_focus_hint(lines: &mut Vec) { + lines.push("Tab/Shift+Tab or h/l (Left/Right) to switch panes".to_string()); +} + +fn sorted_topology( + topology: &ClusterTopology, +) -> (Vec, Vec, Vec) { + let mut agents = topology.agents.clone(); + agents.sort_by(|a, b| a.id.cmp(&b.id)); + + let mut topics = topology.topics.clone(); + topics.sort(); + + let mut edges = topology.edges.clone(); + edges.sort_by(|a, b| { + let kind_a = kind_label(&a.kind); + let kind_b = kind_label(&b.kind); + (a.from.as_str(), a.to.as_str(), kind_a, a.topic.as_str()).cmp(&( + b.from.as_str(), + b.to.as_str(), + kind_b, + b.topic.as_str(), + )) + }); + + (agents, topics, edges) +} + +fn summary_line(summary: Option<&ClusterSummary>) -> String { + summary + .map(|summary| { + let provider = summary.provider.as_deref().unwrap_or("default"); + format!("State: {} | Provider: {}", summary.state, provider) + }) + .unwrap_or_else(|| "Summary pending.".to_string()) +} + +fn edge_details(edge: &TopologyEdge) -> String { + let mut suffix = format!("{}:{}", kind_label(&edge.kind), edge.topic); + if edge.dynamic.unwrap_or(false) { + suffix.push_str(" dynamic"); + } + format!("{} ({suffix})", edge.to) +} + +fn kind_label(kind: &TopologyEdgeKind) -> &'static str { + match kind { + TopologyEdgeKind::Trigger => "trigger", + TopologyEdgeKind::Publish => "publish", + TopologyEdgeKind::Source => "source", + } +} diff --git a/tui-rs/crates/zeroshot-tui/tests/animation_smoothing.rs b/tui-rs/crates/zeroshot-tui/tests/animation_smoothing.rs new file mode 100644 index 00000000..113b52c1 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/animation_smoothing.rs @@ -0,0 +1,40 @@ +use zeroshot_tui::app::animation::{pulse_factor, step_spring_f32, AnimClock, PHASE_TICKS}; + +#[test] +fn anim_clock_advances_and_wraps() { + let mut clock = AnimClock::default(); + clock.advance(100); + let first_phase = clock.phase; + assert_eq!(clock.tick, 1); + assert_eq!(clock.now_ms, 100); + + for _ in 0..PHASE_TICKS { + clock.advance(200); + } + + assert_eq!(clock.tick, 1 + PHASE_TICKS); + assert!((clock.phase - first_phase).abs() < 1e-6); +} + +#[test] +fn camera_smoothing_moves_toward_target() { + let position = (0.0_f32, 0.0_f32); + let velocity = (0.0_f32, 0.0_f32); + let target = (10.0_f32, 0.0_f32); + + let (pos1, vel1) = step_spring_f32(position, velocity, target, 250, 0.16, 0.82); + assert!(pos1.0 > 0.0); + assert!(vel1.0 > 0.0); + + let (pos2, _vel2) = step_spring_f32(pos1, vel1, target, 250, 0.16, 0.82); + assert!(pos2.0 > pos1.0); +} + +#[test] +fn error_pulse_varies_with_phase() { + let start = pulse_factor(0.0); + let mid = pulse_factor(0.25); + assert!((start - mid).abs() > 0.1); + assert!((0.0..=1.0).contains(&start)); + assert!((0.0..=1.0).contains(&mid)); +} diff --git a/tui-rs/crates/zeroshot-tui/tests/app_navigation.rs b/tui-rs/crates/zeroshot-tui/tests/app_navigation.rs new file mode 100644 index 00000000..171ee76b --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/app_navigation.rs @@ -0,0 +1,143 @@ +use zeroshot_tui::app::{ + self, Action, AppState, BackendRequest, Effect, NavigationAction, ScreenAction, ScreenId, +}; +use zeroshot_tui::screens::cluster; + +#[test] +fn esc_pops_until_launcher_root() { + let state = AppState::default(); + let (state, _) = app::update( + state, + Action::Navigate(NavigationAction::Push(ScreenId::Monitor)), + ); + let (state, _) = app::update( + state, + Action::Navigate(NavigationAction::Push(ScreenId::Cluster { + id: "cluster-1".to_string(), + })), + ); + + let (state, _) = app::update(state, Action::Navigate(NavigationAction::Pop)); + assert!(matches!(state.active_screen(), ScreenId::Monitor)); + + let (state, _) = app::update(state, Action::Navigate(NavigationAction::Pop)); + assert!(matches!(state.active_screen(), ScreenId::Launcher)); + + let (state, _) = app::update(state, Action::Navigate(NavigationAction::Pop)); + assert_eq!(state.screen_stack, vec![ScreenId::Launcher]); +} + +#[test] +fn push_replace_pop_behave_correctly() { + let state = AppState::default(); + let (state, _) = app::update( + state, + Action::Navigate(NavigationAction::Push(ScreenId::Monitor)), + ); + assert_eq!(state.screen_stack.len(), 2); + + let (state, _) = app::update( + state, + Action::Navigate(NavigationAction::ReplaceTop(ScreenId::Cluster { + id: "cluster-2".to_string(), + })), + ); + assert!(matches!(state.active_screen(), ScreenId::Cluster { .. })); + + let (state, _) = app::update(state, Action::Navigate(NavigationAction::Pop)); + assert!(matches!(state.active_screen(), ScreenId::Launcher)); +} + +#[test] +fn cluster_entry_requests_topology() { + let state = AppState::default(); + let (_state, effects) = app::update( + state, + Action::Navigate(NavigationAction::Push(ScreenId::Cluster { + id: "cluster-1".to_string(), + })), + ); + + assert!( + effects.contains(&Effect::Backend(BackendRequest::GetClusterTopology { + cluster_id: "cluster-1".to_string(), + })) + ); +} + +#[test] +fn pop_from_cluster_unsubscribes_active_streams() { + let state = AppState::default(); + let (state, _) = app::update( + state, + Action::Navigate(NavigationAction::Push(ScreenId::Cluster { + id: "cluster-1".to_string(), + })), + ); + + let mut state = state; + let entry = state.clusters.get_mut("cluster-1").expect("cluster state"); + entry.log_subscription = Some("log-sub".to_string()); + entry.timeline_subscription = Some("timeline-sub".to_string()); + + let (state, effects) = app::update(state, Action::Navigate(NavigationAction::Pop)); + assert!( + effects.contains(&Effect::Backend(BackendRequest::Unsubscribe { + subscription_id: "log-sub".to_string(), + })) + ); + assert!( + effects.contains(&Effect::Backend(BackendRequest::Unsubscribe { + subscription_id: "timeline-sub".to_string(), + })) + ); + + let entry = state.clusters.get("cluster-1").expect("cluster state"); + assert!(entry.log_subscription.is_none()); + assert!(entry.timeline_subscription.is_none()); +} + +#[test] +fn push_to_agent_unsubscribes_cluster_streams() { + let state = AppState::default(); + let (state, _) = app::update( + state, + Action::Navigate(NavigationAction::Push(ScreenId::Cluster { + id: "cluster-1".to_string(), + })), + ); + + let mut state = state; + let entry = state.clusters.get_mut("cluster-1").expect("cluster state"); + entry.log_subscription = Some("log-sub".to_string()); + entry.timeline_subscription = Some("timeline-sub".to_string()); + + let (state, effects) = app::update( + state, + Action::Screen(ScreenAction::Cluster { + id: "cluster-1".to_string(), + action: cluster::Action::OpenAgent("agent-1".to_string()), + }), + ); + + assert!( + effects.contains(&Effect::Backend(BackendRequest::Unsubscribe { + subscription_id: "log-sub".to_string(), + })) + ); + assert!( + effects.contains(&Effect::Backend(BackendRequest::Unsubscribe { + subscription_id: "timeline-sub".to_string(), + })) + ); + + let entry = state.clusters.get("cluster-1").expect("cluster state"); + assert!(entry.log_subscription.is_none()); + assert!(entry.timeline_subscription.is_none()); + + assert!(matches!( + state.active_screen(), + ScreenId::Agent { cluster_id, agent_id } + if cluster_id == "cluster-1" && agent_id == "agent-1" + )); +} diff --git a/tui-rs/crates/zeroshot-tui/tests/backend_framing.rs b/tui-rs/crates/zeroshot-tui/tests/backend_framing.rs new file mode 100644 index 00000000..ead91e61 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/backend_framing.rs @@ -0,0 +1,57 @@ +use zeroshot_tui::backend::framing::{FrameDecoder, FrameEncoder, FrameError, MAX_FRAME_SIZE}; + +#[test] +fn single_frame_roundtrip() { + let payload = br#"{"jsonrpc":"2.0","id":1,"method":"ping"}"#; + let framed = FrameEncoder::encode(payload).expect("encode"); + let mut decoder = FrameDecoder::new(); + let frames = decoder.push(&framed).expect("decode"); + assert_eq!(frames.len(), 1); + assert_eq!(frames[0], payload); +} + +#[test] +fn multiple_frames_in_one_buffer() { + let payload_a = br#"{"jsonrpc":"2.0","id":1,"method":"one"}"#; + let payload_b = br#"{"jsonrpc":"2.0","id":2,"method":"two"}"#; + let mut combined = Vec::new(); + combined.extend(FrameEncoder::encode(payload_a).unwrap()); + combined.extend(FrameEncoder::encode(payload_b).unwrap()); + + let mut decoder = FrameDecoder::new(); + let frames = decoder.push(&combined).expect("decode"); + assert_eq!(frames.len(), 2); + assert_eq!(frames[0], payload_a); + assert_eq!(frames[1], payload_b); +} + +#[test] +fn split_frame_across_chunks() { + let payload = br#"{"jsonrpc":"2.0","id":3,"method":"split"}"#; + let framed = FrameEncoder::encode(payload).unwrap(); + let mid = framed.len() / 2; + + let mut decoder = FrameDecoder::new(); + let frames = decoder.push(&framed[..mid]).expect("decode"); + assert!(frames.is_empty()); + + let frames = decoder.push(&framed[mid..]).expect("decode"); + assert_eq!(frames.len(), 1); + assert_eq!(frames[0], payload); +} + +#[test] +fn oversized_frame_rejected() { + let oversized = MAX_FRAME_SIZE + 1; + let header = format!("Content-Length: {}\r\n\r\n", oversized); + let mut decoder = FrameDecoder::new(); + let error = decoder.push(header.as_bytes()).expect_err("oversized"); + assert!(matches!(error, FrameError::FrameTooLarge(_))); +} + +#[test] +fn encoder_rejects_oversized_payload() { + let payload = vec![0u8; MAX_FRAME_SIZE + 1]; + let error = FrameEncoder::encode(&payload).expect_err("oversized"); + assert!(matches!(error, FrameError::FrameTooLarge(_))); +} diff --git a/tui-rs/crates/zeroshot-tui/tests/backend_integration.rs b/tui-rs/crates/zeroshot-tui/tests/backend_integration.rs new file mode 100644 index 00000000..505d8019 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/backend_integration.rs @@ -0,0 +1,203 @@ +use std::path::{Path, PathBuf}; +use std::process::Command; +use std::sync::{Mutex, OnceLock}; +use std::time::Duration; + +use zeroshot_tui::backend::stdio::StdioBackendClient; +use zeroshot_tui::backend::{BackendClient, BackendConfig, BackendEvent}; +use zeroshot_tui::protocol::{ + GetClusterSummaryParams, SubscribeClusterLogsParams, SubscribeClusterTimelineParams, + UnsubscribeParams, +}; + +static ENV_LOCK: OnceLock> = OnceLock::new(); +static BACKEND_PATH: OnceLock> = OnceLock::new(); + +fn is_ci() -> bool { + match std::env::var("CI") { + Ok(value) => matches!(value.to_lowercase().as_str(), "1" | "true" | "yes"), + Err(_) => false, + } +} + +struct EnvGuard { + keys: Vec<&'static str>, +} + +impl EnvGuard { + fn set(pairs: &[(&'static str, &'static str)]) -> Self { + let mut keys = Vec::with_capacity(pairs.len()); + for (key, value) in pairs { + std::env::set_var(key, value); + keys.push(*key); + } + Self { keys } + } +} + +impl Drop for EnvGuard { + fn drop(&mut self) { + for key in &self.keys { + std::env::remove_var(key); + } + } +} + +fn env_lock() -> std::sync::MutexGuard<'static, ()> { + ENV_LOCK + .get_or_init(|| Mutex::new(())) + .lock() + .unwrap_or_else(|err| err.into_inner()) +} + +fn repo_root() -> Option { + let cwd = std::env::current_dir().ok()?; + for ancestor in cwd.ancestors() { + if ancestor.join("package.json").is_file() { + return Some(ancestor.to_path_buf()); + } + } + None +} + +fn find_backend_path(start: &Path) -> Option { + for ancestor in start.ancestors() { + let candidate = ancestor.join("lib/tui-backend/server.js"); + if candidate.is_file() { + return Some(candidate); + } + } + None +} + +fn resolve_backend_path() -> Option { + let cwd = std::env::current_dir().ok()?; + if let Some(path) = find_backend_path(&cwd) { + return Some(path); + } + + let root = repo_root()?; + let status = Command::new("npm") + .args(["run", "build:tui-backend"]) + .current_dir(&root) + .status() + .ok()?; + if !status.success() { + eprintln!("npm run build:tui-backend failed"); + return None; + } + + find_backend_path(&root) +} + +fn build_client() -> Option { + let backend_path = BACKEND_PATH.get_or_init(resolve_backend_path).clone(); + let Some(backend_path) = backend_path else { + if is_ci() { + panic!("TUI backend not available. Run `npm ci` and `npm run build:tui-backend` before cargo test."); + } + return None; + }; + let mut config = BackendConfig::with_backend_path(backend_path); + config.request_timeout = Some(Duration::from_secs(10)); + match StdioBackendClient::connect(config) { + Ok(client) => Some(client), + Err(err) => { + if is_ci() { + panic!("Failed to connect to TUI backend: {err}"); + } + eprintln!("Skipping backend integration: {err}"); + None + } + } +} + +#[test] +fn initialize_and_list_clusters() { + let _guard = env_lock(); + let _env = EnvGuard::set(&[ + ("ZEROSHOT_TUI_BACKEND_MOCK_LAUNCH", "1"), + ("ZEROSHOT_TUI_BACKEND_MOCK_GUIDANCE", "1"), + ]); + + let Some(client) = build_client() else { + eprintln!("Skipping backend integration: backend build unavailable"); + return; + }; + assert_eq!(client.protocol_version(), 1); + assert!(client.server_capabilities().is_some()); + let result = client.list_clusters().expect("listClusters"); + let _ = result.clusters.len(); +} + +#[test] +fn interleaved_notifications_do_not_break_requests() { + let _guard = env_lock(); + let _env = EnvGuard::set(&[ + ("ZEROSHOT_TUI_BACKEND_MOCK_LAUNCH", "1"), + ("ZEROSHOT_TUI_BACKEND_MOCK_GUIDANCE", "1"), + ]); + + let Some(mut client) = build_client() else { + eprintln!("Skipping backend integration: backend build unavailable"); + return; + }; + let events = client.take_event_receiver().expect("event receiver"); + + let subscription = client + .subscribe_cluster_logs(SubscribeClusterLogsParams { + cluster_id: "unknown-cluster".to_string(), + agent_id: None, + }) + .expect("subscribe logs"); + + let _timeline = client + .subscribe_cluster_timeline(SubscribeClusterTimelineParams { + cluster_id: "unknown-cluster".to_string(), + }) + .expect("subscribe timeline"); + + let list_result = client.list_clusters().expect("listClusters"); + let summary = list_result + .clusters + .get(0) + .map(|cluster| cluster.id.clone()) + .unwrap_or_else(|| "unknown-cluster".to_string()); + let _ = client + .get_cluster_summary(GetClusterSummaryParams { + cluster_id: summary, + }) + .err(); + + let _ = events.recv_timeout(Duration::from_millis(200)); + + let _ = client + .unsubscribe(UnsubscribeParams { + subscription_id: subscription.subscription_id, + }) + .expect("unsubscribe"); +} + +#[test] +fn backend_exit_and_drop() { + let _guard = env_lock(); + let _env = EnvGuard::set(&[ + ("ZEROSHOT_TUI_BACKEND_MOCK_LAUNCH", "1"), + ("ZEROSHOT_TUI_BACKEND_MOCK_GUIDANCE", "1"), + ]); + + let Some(mut client) = build_client() else { + eprintln!("Skipping backend integration: backend build unavailable"); + return; + }; + let events = client.take_event_receiver().expect("event receiver"); + + client.shutdown().expect("shutdown"); + let event = events + .recv_timeout(Duration::from_secs(2)) + .expect("backend exit event"); + match event { + BackendEvent::BackendExited(_) => {} + BackendEvent::Notification(_) => panic!("expected BackendExited"), + } +} diff --git a/tui-rs/crates/zeroshot-tui/tests/cluster_canvas_snapshots.rs b/tui-rs/crates/zeroshot-tui/tests/cluster_canvas_snapshots.rs new file mode 100644 index 00000000..2664a5cd --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/cluster_canvas_snapshots.rs @@ -0,0 +1,115 @@ +mod ui_snapshot_helpers; + +use ui_snapshot_helpers::render_to_text; + +use zeroshot_tui::app::{AppState, ScreenId, TimeCursor, TimeCursorMode, UiVariant}; +use zeroshot_tui::protocol::{ + ClusterLogLine, ClusterTopology, TimelineEvent, TopologyAgent, TopologyEdge, TopologyEdgeKind, +}; +use zeroshot_tui::screens::{cluster, cluster_canvas}; +use zeroshot_tui::ui; + +fn sample_topology() -> ClusterTopology { + ClusterTopology { + agents: vec![ + TopologyAgent { + id: "worker".to_string(), + role: Some("implementation".to_string()), + }, + TopologyAgent { + id: "validator".to_string(), + role: Some("validator".to_string()), + }, + ], + edges: vec![ + TopologyEdge { + from: "ISSUE_OPENED".to_string(), + to: "worker".to_string(), + topic: "ISSUE_OPENED".to_string(), + kind: TopologyEdgeKind::Trigger, + dynamic: Some(true), + }, + TopologyEdge { + from: "worker".to_string(), + to: "IMPLEMENTATION_READY".to_string(), + topic: "IMPLEMENTATION_READY".to_string(), + kind: TopologyEdgeKind::Publish, + dynamic: None, + }, + ], + topics: vec![ + "ISSUE_OPENED".to_string(), + "IMPLEMENTATION_READY".to_string(), + ], + } +} + +fn sample_cluster_state() -> cluster::State { + let mut state = cluster::State::default(); + let topology = sample_topology(); + state.topology = Some(topology); + + state.logs_time.push_many(vec![ + ClusterLogLine { + id: "log-1".to_string(), + timestamp: 900, + text: "agent started".to_string(), + agent: Some("worker".to_string()), + role: Some("implementation".to_string()), + sender: Some("worker".to_string()), + }, + ClusterLogLine { + id: "log-2".to_string(), + timestamp: 950, + text: "cluster event".to_string(), + agent: None, + role: None, + sender: Some("system".to_string()), + }, + ]); + + state.timeline_time.push_many(vec![TimelineEvent { + id: "evt-1".to_string(), + timestamp: 920, + topic: "ISSUE_OPENED".to_string(), + label: "opened".to_string(), + approved: None, + sender: Some("system".to_string()), + }]); + + state +} + +#[test] +fn cluster_canvas_snapshot_with_focus_and_overlay() { + let cluster_id = "cluster-1".to_string(); + let mut state = AppState::default(); + state.ui_variant = UiVariant::Disruptive; + state.screen_stack = vec![ScreenId::ClusterCanvas { + id: cluster_id.clone(), + }]; + state.time_cursor = TimeCursor { + mode: TimeCursorMode::Live, + t_ms: 1000, + window_ms: 1000, + }; + + let cluster_state = sample_cluster_state(); + let topology = cluster_state.topology.clone().expect("topology"); + + let mut canvas_state = cluster_canvas::State::default(); + canvas_state.focused_id = Some("worker".to_string()); + canvas_state.update_layout(&topology); + + state.clusters.insert(cluster_id.clone(), cluster_state); + state + .cluster_canvases + .insert(cluster_id.clone(), canvas_state); + + let content = render_to_text(100, 26, |frame| ui::render(frame, &state)); + assert!(content.contains("Cluster Canvas cluster-1")); + assert!(content.contains("worker")); + assert!(content.contains("Logs - agent worker")); + assert!(content.contains("[LIVE]")); + assert!(content.contains("agent started")); +} diff --git a/tui-rs/crates/zeroshot-tui/tests/cluster_reducer.rs b/tui-rs/crates/zeroshot-tui/tests/cluster_reducer.rs new file mode 100644 index 00000000..be9a2b16 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/cluster_reducer.rs @@ -0,0 +1,88 @@ +use zeroshot_tui::protocol::{ClusterLogLine, TimelineEvent}; +use zeroshot_tui::screens::cluster::{self, ClusterPane, FocusDirection}; + +fn log_line(id: usize, agent: Option<&str>, role: Option<&str>) -> ClusterLogLine { + ClusterLogLine { + id: format!("log-{id}"), + timestamp: id as i64, + text: format!("line-{id}"), + agent: agent.map(|value| value.to_string()), + role: role.map(|value| value.to_string()), + sender: None, + } +} + +fn timeline_event(id: usize) -> TimelineEvent { + TimelineEvent { + id: format!("event-{id}"), + timestamp: id as i64, + topic: format!("topic-{id}"), + label: format!("label-{id}"), + approved: None, + sender: None, + } +} + +#[test] +fn log_buffer_bounds_and_dropped_count() { + let mut state = cluster::State::default(); + state.push_log_lines(vec![log_line(0, None, None)], Some(3)); + assert_eq!(state.logs.len(), 2); + let first = state.logs.items.front().expect("expected synthetic line"); + assert!(first.text.contains("dropped 3")); + + let mut state = cluster::State::default(); + let lines: Vec<_> = (0..(cluster::MAX_LOG_LINES + 5)) + .map(|id| log_line(id, None, None)) + .collect(); + state.push_log_lines(lines, None); + assert_eq!(state.logs.len(), cluster::MAX_LOG_LINES); + let first = state.logs.items.front().expect("expected log line"); + assert_eq!(first.id, "log-5"); +} + +#[test] +fn timeline_buffer_bounds() { + let mut state = cluster::State::default(); + let events: Vec<_> = (0..(cluster::MAX_TIMELINE_EVENTS + 3)) + .map(timeline_event) + .collect(); + state.push_timeline_events(events); + assert_eq!(state.timeline.len(), cluster::MAX_TIMELINE_EVENTS); + let first = state.timeline.items.front().expect("expected event"); + assert_eq!(first.id, "event-3"); +} + +#[test] +fn focus_cycles_and_activate_uses_selected_agent() { + let mut state = cluster::State::default(); + assert_eq!(state.focus, ClusterPane::Topology); + state.cycle_focus(FocusDirection::Next); + assert_eq!(state.focus, ClusterPane::Logs); + state.cycle_focus(FocusDirection::Next); + assert_eq!(state.focus, ClusterPane::Timeline); + state.cycle_focus(FocusDirection::Next); + assert_eq!(state.focus, ClusterPane::Agents); + state.cycle_focus(FocusDirection::Prev); + assert_eq!(state.focus, ClusterPane::Timeline); + + state.focus = ClusterPane::Agents; + state.push_log_lines(vec![log_line(1, Some("agent-a"), Some("role-a"))], None); + state.push_log_lines(vec![log_line(2, Some("agent-b"), Some("role-b"))], None); + state.move_focused(1); + assert_eq!(state.activate_focused(), Some("agent-b".to_string())); + + state.push_log_lines(vec![log_line(3, Some("agent-c"), Some("role-c"))], None); + assert_eq!(state.activate_focused(), Some("agent-b".to_string())); +} + +#[test] +fn scroll_offset_grows_when_new_lines_arrive() { + let mut state = cluster::State::default(); + state.focus = ClusterPane::Logs; + state.push_log_lines(vec![log_line(0, None, None), log_line(1, None, None)], None); + state.move_focused(-1); + assert_eq!(state.logs.scroll_offset, 1); + state.push_log_lines(vec![log_line(2, None, None)], None); + assert_eq!(state.logs.scroll_offset, 2); +} diff --git a/tui-rs/crates/zeroshot-tui/tests/disruptive_render.rs b/tui-rs/crates/zeroshot-tui/tests/disruptive_render.rs new file mode 100644 index 00000000..e738aeb1 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/disruptive_render.rs @@ -0,0 +1,80 @@ +use ratatui::backend::TestBackend; +use ratatui::buffer::Buffer; +use ratatui::Terminal; + +use zeroshot_tui::app::{AppState, UiVariant}; +use zeroshot_tui::protocol::ClusterSummary; +use zeroshot_tui::ui; + +fn buffer_text(buffer: &Buffer) -> String { + let area = buffer.area; + let mut lines = Vec::new(); + for y in area.top()..area.bottom() { + let mut line = String::new(); + for x in area.left()..area.right() { + line.push_str(buffer.cell((x, y)).map_or("", |c| c.symbol())); + } + lines.push(line); + } + lines.join("\n") +} + +fn cluster_summary(id: &str, state: &str) -> ClusterSummary { + ClusterSummary { + id: id.to_string(), + state: state.to_string(), + provider: None, + created_at: 0, + agent_count: 0, + message_count: 0, + cwd: None, + } +} + +#[test] +fn disruptive_render_empty_radar() { + let backend = TestBackend::new(60, 12); + let mut terminal = Terminal::new(backend).expect("terminal"); + let mut state = AppState::default(); + state.ui_variant = UiVariant::Disruptive; + + terminal + .draw(|frame| ui::render(frame, &state)) + .expect("draw"); + + let content = buffer_text(terminal.backend().buffer()); + assert!(content.contains("Fleet Radar")); + assert!(content.contains("No clusters yet.")); + assert!(content.contains("Type an intent in the spine to start a cluster.")); + assert!(!content.contains("ID")); + assert!(!content.contains("STATE")); + assert!(!content.contains("ZEROSHOT")); +} + +#[test] +fn disruptive_render_radar_three_states() { + let backend = TestBackend::new(80, 16); + let mut terminal = Terminal::new(backend).expect("terminal"); + let mut state = AppState::default(); + state.ui_variant = UiVariant::Disruptive; + state.now_ms = 10_000; + state.fleet_radar.set_clusters( + vec![ + cluster_summary("run-1", "running"), + cluster_summary("done-1", "done"), + cluster_summary("err-1", "error"), + ], + state.now_ms, + ); + + terminal + .draw(|frame| ui::render(frame, &state)) + .expect("draw"); + + let content = buffer_text(terminal.backend().buffer()); + assert!(content.contains("run-1")); + assert!(content.contains("done-1")); + assert!(content.contains("err-1")); + assert!(!content.contains("ID")); + assert!(!content.contains("STATE")); +} diff --git a/tui-rs/crates/zeroshot-tui/tests/input_routing.rs b/tui-rs/crates/zeroshot-tui/tests/input_routing.rs new file mode 100644 index 00000000..e9043a55 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/input_routing.rs @@ -0,0 +1,312 @@ +use crossterm::event::{KeyCode, KeyEvent, KeyModifiers}; + +use zeroshot_tui::app::{ + self, Action, NavigationAction, ScreenAction, ScreenId, SpineAction, SpineMode, UiVariant, +}; +use zeroshot_tui::input; +use zeroshot_tui::protocol::ClusterSummary; +use zeroshot_tui::screens::{cluster, launcher, monitor}; + +fn state_for(screen: ScreenId) -> app::AppState { + let mut state = app::AppState::default(); + state.screen_stack = vec![screen]; + state +} + +#[test] +fn global_keys_apply_everywhere() { + let screens = vec![ + ScreenId::Launcher, + ScreenId::Monitor, + ScreenId::Cluster { + id: "c1".to_string(), + }, + ScreenId::Agent { + cluster_id: "c1".to_string(), + agent_id: "a1".to_string(), + }, + ]; + + for screen in screens { + let state = state_for(screen); + let esc = KeyEvent::new(KeyCode::Esc, KeyModifiers::NONE); + let action = input::route_key(&state, esc); + assert!(matches!( + action, + Some(Action::Navigate(NavigationAction::Pop)) + )); + + let ctrl_c = KeyEvent::new(KeyCode::Char('c'), KeyModifiers::CONTROL); + let action = input::route_key(&state, ctrl_c); + assert!(matches!(action, Some(Action::Quit))); + } +} + +#[test] +fn screen_specific_keys_only_apply_to_focused_screen() { + let launcher = ScreenId::Launcher; + let monitor_screen = ScreenId::Monitor; + let cluster_screen = ScreenId::Cluster { + id: "c1".to_string(), + }; + + let down = KeyEvent::new(KeyCode::Down, KeyModifiers::NONE); + let state = state_for(launcher); + assert!(input::route_key(&state, down).is_none()); + + let down = KeyEvent::new(KeyCode::Down, KeyModifiers::NONE); + let state = state_for(monitor_screen); + let action = input::route_key(&state, down); + assert!(matches!( + action, + Some(Action::Screen(ScreenAction::Monitor( + monitor::Action::MoveSelection(1) + ))) + )); + + let tab = KeyEvent::new(KeyCode::Tab, KeyModifiers::NONE); + let state = state_for(cluster_screen.clone()); + let action = input::route_key(&state, tab); + match action { + Some(Action::Screen(ScreenAction::Cluster { id, action })) => { + assert_eq!(id, "c1"); + assert!(matches!( + action, + cluster::Action::CycleFocus(cluster::FocusDirection::Next) + )); + } + _ => panic!("expected cluster focus action"), + } + + let up = KeyEvent::new(KeyCode::Up, KeyModifiers::NONE); + let state = state_for(cluster_screen.clone()); + let action = input::route_key(&state, up); + assert!(matches!( + action, + Some(Action::Screen(ScreenAction::Cluster { + action: cluster::Action::MoveFocused(-1), + .. + })) + )); + + let down = KeyEvent::new(KeyCode::Down, KeyModifiers::NONE); + let state = state_for(cluster_screen.clone()); + let action = input::route_key(&state, down); + assert!(matches!( + action, + Some(Action::Screen(ScreenAction::Cluster { + action: cluster::Action::MoveFocused(1), + .. + })) + )); + + let k = KeyEvent::new(KeyCode::Char('k'), KeyModifiers::NONE); + let state = state_for(cluster_screen.clone()); + let action = input::route_key(&state, k); + assert!(matches!( + action, + Some(Action::Screen(ScreenAction::Cluster { + action: cluster::Action::MoveFocused(-1), + .. + })) + )); + + let j = KeyEvent::new(KeyCode::Char('j'), KeyModifiers::NONE); + let state = state_for(cluster_screen.clone()); + let action = input::route_key(&state, j); + assert!(matches!( + action, + Some(Action::Screen(ScreenAction::Cluster { + action: cluster::Action::MoveFocused(1), + .. + })) + )); + + let enter = KeyEvent::new(KeyCode::Enter, KeyModifiers::NONE); + let state = state_for(cluster_screen); + let action = input::route_key(&state, enter); + assert!(matches!( + action, + Some(Action::Screen(ScreenAction::Cluster { + action: cluster::Action::ActivateFocused, + .. + })) + )); + + let tab = KeyEvent::new(KeyCode::Tab, KeyModifiers::NONE); + let state = state_for(ScreenId::Monitor); + assert!(input::route_key(&state, tab).is_none()); +} + +#[test] +fn launcher_keys_edit_input_state() { + let mut state = app::AppState::default(); + state.screen_stack = vec![ScreenId::Launcher]; + + let action = input::route_key( + &state, + KeyEvent::new(KeyCode::Char('a'), KeyModifiers::NONE), + ) + .expect("expected insert char"); + let (next_state, _) = app::update(state, action); + state = next_state; + assert_eq!(state.launcher.input, "a"); + assert_eq!(state.launcher.cursor, 1); + + let action = input::route_key( + &state, + KeyEvent::new(KeyCode::Char('b'), KeyModifiers::NONE), + ) + .expect("expected insert char"); + let (next_state, _) = app::update(state, action); + state = next_state; + assert_eq!(state.launcher.input, "ab"); + assert_eq!(state.launcher.cursor, 2); + + let action = input::route_key(&state, KeyEvent::new(KeyCode::Left, KeyModifiers::NONE)) + .expect("expected move left"); + let (next_state, _) = app::update(state, action); + state = next_state; + assert_eq!(state.launcher.cursor, 1); + + let action = input::route_key( + &state, + KeyEvent::new(KeyCode::Backspace, KeyModifiers::NONE), + ) + .expect("expected backspace"); + let (next_state, _) = app::update(state, action); + state = next_state; + assert_eq!(state.launcher.input, "b"); + assert_eq!(state.launcher.cursor, 0); + + let action = input::route_key(&state, KeyEvent::new(KeyCode::Delete, KeyModifiers::NONE)) + .expect("expected delete"); + let (next_state, _) = app::update(state, action); + state = next_state; + assert_eq!(state.launcher.input, ""); + assert_eq!(state.launcher.cursor, 0); +} + +#[test] +fn q_quits_except_in_launcher_input() { + let key = KeyEvent::new(KeyCode::Char('q'), KeyModifiers::NONE); + let state = state_for(ScreenId::Launcher); + let action = input::route_key(&state, key); + assert!(matches!( + action, + Some(Action::Screen(ScreenAction::Launcher( + launcher::Action::InsertChar('q') + ))) + )); + + let screens = vec![ + ScreenId::Monitor, + ScreenId::Cluster { + id: "c1".to_string(), + }, + ScreenId::Agent { + cluster_id: "c1".to_string(), + agent_id: "a1".to_string(), + }, + ]; + + for screen in screens { + let state = state_for(screen); + let action = input::route_key( + &state, + KeyEvent::new(KeyCode::Char('q'), KeyModifiers::NONE), + ); + assert!(matches!(action, Some(Action::Quit))); + } +} + +#[test] +fn disruptive_routes_spine_shortcuts() { + let mut state = state_for(ScreenId::FleetRadar); + state.ui_variant = UiVariant::Disruptive; + + let action = input::route_key( + &state, + KeyEvent::new(KeyCode::Char('/'), KeyModifiers::NONE), + ); + match action { + Some(Action::Spine(SpineAction::EnterMode { mode, prefill })) => { + assert_eq!(mode, SpineMode::Command); + assert_eq!(prefill, ""); + } + _ => panic!("expected command mode for /"), + } + + let action = input::route_key( + &state, + KeyEvent::new(KeyCode::Char('?'), KeyModifiers::NONE), + ); + match action { + Some(Action::Spine(SpineAction::EnterMode { mode, prefill })) => { + assert_eq!(mode, SpineMode::Command); + assert_eq!(prefill, "help "); + } + _ => panic!("expected help prefill for ?"), + } + + let action = input::route_key( + &state, + KeyEvent::new(KeyCode::Char('u'), KeyModifiers::CONTROL), + ); + assert!(matches!(action, Some(Action::Spine(SpineAction::Clear)))); +} + +#[test] +fn disruptive_esc_cancels_or_pops() { + let mut state = state_for(ScreenId::FleetRadar); + state.ui_variant = UiVariant::Disruptive; + + let action = input::route_key(&state, KeyEvent::new(KeyCode::Esc, KeyModifiers::NONE)); + assert!(matches!( + action, + Some(Action::Navigate(NavigationAction::Pop)) + )); + + state.spine.mode = SpineMode::Command; + let action = input::route_key(&state, KeyEvent::new(KeyCode::Esc, KeyModifiers::NONE)); + assert!(matches!(action, Some(Action::Spine(SpineAction::Cancel)))); +} + +#[test] +fn disruptive_enter_submits_or_zooms() { + let mut state = state_for(ScreenId::FleetRadar); + state.ui_variant = UiVariant::Disruptive; + state.fleet_radar.set_clusters( + vec![ClusterSummary { + id: "c1".to_string(), + state: "running".to_string(), + provider: None, + created_at: 0, + agent_count: 0, + message_count: 0, + cwd: None, + }], + 1000, + ); + + let action = input::route_key(&state, KeyEvent::new(KeyCode::Enter, KeyModifiers::NONE)); + assert!(matches!( + action, + Some(Action::Navigate(NavigationAction::Push(ScreenId::ClusterCanvas { id }))) + if id == "c1" + )); + + state.spine.input.input = "hello".to_string(); + let action = input::route_key(&state, KeyEvent::new(KeyCode::Enter, KeyModifiers::NONE)); + assert!(matches!(action, Some(Action::Spine(SpineAction::Submit)))); +} + +#[test] +fn disruptive_ctrl_c_quits() { + let mut state = state_for(ScreenId::FleetRadar); + state.ui_variant = UiVariant::Disruptive; + + let ctrl_c = KeyEvent::new(KeyCode::Char('c'), KeyModifiers::CONTROL); + let action = input::route_key(&state, ctrl_c); + assert!(matches!(action, Some(Action::Quit))); +} diff --git a/tui-rs/crates/zeroshot-tui/tests/launcher_reducer.rs b/tui-rs/crates/zeroshot-tui/tests/launcher_reducer.rs new file mode 100644 index 00000000..6d2cb5ef --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/launcher_reducer.rs @@ -0,0 +1,61 @@ +use zeroshot_tui::app::{ + self, Action, AppState, BackendAction, BackendRequest, CommandContext, CommandRequest, Effect, + ScreenAction, ScreenId, +}; +use zeroshot_tui::screens::launcher; + +#[test] +fn submit_text_routes_to_start_cluster_from_text() { + let mut state = AppState::default(); + state.launcher.input = "123".to_string(); + state.provider_override = Some("claude".to_string()); + + let (state, effects) = app::update( + state, + Action::Screen(ScreenAction::Launcher(launcher::Action::Submit)), + ); + + assert_eq!(state.last_error, None); + assert_eq!( + effects, + vec![Effect::Backend(BackendRequest::StartClusterFromText { + text: "123".to_string(), + provider_override: Some("claude".to_string()), + })] + ); +} + +#[test] +fn submit_command_routes_to_command_effect() { + let mut state = AppState::default(); + state.launcher.input = "/help".to_string(); + + let (state, effects) = app::update( + state, + Action::Screen(ScreenAction::Launcher(launcher::Action::Submit)), + ); + + assert_eq!(state.last_error, None); + assert_eq!( + effects, + vec![Effect::Command(CommandRequest::SubmitRaw { + raw: "/help".to_string(), + context: CommandContext { + provider_override: None, + active_screen: ScreenId::Launcher, + ui_variant: state.ui_variant, + }, + })] + ); +} + +#[test] +fn backend_error_sets_last_error() { + let state = AppState::default(); + let (state, _) = app::update( + state, + Action::Backend(BackendAction::Error("boom".to_string())), + ); + + assert_eq!(state.last_error.as_deref(), Some("boom")); +} diff --git a/tui-rs/crates/zeroshot-tui/tests/microscope_snapshots.rs b/tui-rs/crates/zeroshot-tui/tests/microscope_snapshots.rs new file mode 100644 index 00000000..adafd1a6 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/microscope_snapshots.rs @@ -0,0 +1,114 @@ +mod ui_snapshot_helpers; + +use ui_snapshot_helpers::render_to_text; + +use zeroshot_tui::app::{agent_microscope, TimeCursor, TimeCursorMode}; +use zeroshot_tui::protocol::{ClusterLogLine, TimelineEvent}; +use zeroshot_tui::screens::agent_microscope as microscope_screen; +use zeroshot_tui::ui::shared::TimeIndexedBuffer; + +fn sample_logs() -> TimeIndexedBuffer { + let mut buffer = TimeIndexedBuffer::new(32); + buffer.push_many(vec![ + ClusterLogLine { + id: "log-1".to_string(), + timestamp: 100, + text: "started task".to_string(), + agent: Some("agent-9".to_string()), + role: Some("implementation".to_string()), + sender: Some("agent-9".to_string()), + }, + ClusterLogLine { + id: "log-2".to_string(), + timestamp: 200, + text: "finished task".to_string(), + agent: Some("agent-9".to_string()), + role: Some("implementation".to_string()), + sender: Some("agent-9".to_string()), + }, + ]); + buffer +} + +fn sample_timeline() -> TimeIndexedBuffer { + let mut buffer = TimeIndexedBuffer::new(16); + buffer.push_many(vec![TimelineEvent { + id: "evt-1".to_string(), + timestamp: 150, + topic: "IMPLEMENTATION_READY".to_string(), + label: "ready".to_string(), + approved: None, + sender: Some("agent-9".to_string()), + }]); + buffer +} + +fn sample_state() -> agent_microscope::State { + let mut state = agent_microscope::State::default(); + state.logs_time = sample_logs(); + state.role = Some("implementation".to_string()); + state.status = Some("running".to_string()); + state +} + +#[test] +fn microscope_snapshot_live_mode() { + let microscope_state = sample_state(); + let timeline = sample_timeline(); + let cursor = TimeCursor { + mode: TimeCursorMode::Live, + t_ms: 200, + window_ms: 400, + }; + + let content = render_to_text(80, 18, |frame| { + let area = frame.area(); + microscope_screen::render( + frame, + area, + "cluster-7", + "agent-9", + Some(&timeline), + Some(µscope_state), + &cursor, + ); + }); + + assert!(content.contains("Stream")); + assert!(content.contains("[LIVE]")); + assert!(content.contains("Agent: agent-9")); + assert!(content.contains("Role: implementation")); + assert!(content.contains("Status: running")); + assert!(content.contains("Cluster: cluster-7")); + assert!(content.contains("started task")); +} + +#[test] +fn microscope_snapshot_scrub_mode() { + let microscope_state = sample_state(); + let timeline = sample_timeline(); + let cursor = TimeCursor { + mode: TimeCursorMode::Scrub, + t_ms: 200, + window_ms: 200, + }; + + let content = render_to_text(80, 18, |frame| { + let area = frame.area(); + microscope_screen::render( + frame, + area, + "cluster-7", + "agent-9", + Some(&timeline), + Some(µscope_state), + &cursor, + ); + }); + + assert!(content.contains("Stream")); + assert!(content.contains("[SCRUB]")); + assert!(content.contains("Agent: agent-9")); + assert!(content.contains("Cluster: cluster-7")); + assert!(content.contains("finished task")); +} diff --git a/tui-rs/crates/zeroshot-tui/tests/monitor_reducer.rs b/tui-rs/crates/zeroshot-tui/tests/monitor_reducer.rs new file mode 100644 index 00000000..e9d520c0 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/monitor_reducer.rs @@ -0,0 +1,107 @@ +use zeroshot_tui::app::{self, Action, AppState, BackendRequest, Effect, ScreenAction, ScreenId}; +use zeroshot_tui::protocol::ClusterSummary; +use zeroshot_tui::screens::monitor; + +fn cluster_summary(id: &str, message_count: i64) -> ClusterSummary { + ClusterSummary { + id: id.to_string(), + state: "running".to_string(), + provider: Some("codex".to_string()), + created_at: 0, + agent_count: 0, + message_count, + cwd: None, + } +} + +#[test] +fn monitor_selection_clamps_on_shrink() { + let mut state = monitor::State::default(); + state.set_clusters( + vec![ + cluster_summary("c1", 0), + cluster_summary("c2", 0), + cluster_summary("c3", 0), + ], + 1000, + ); + state.selected = 2; + + state.set_clusters(vec![cluster_summary("c1", 0)], 2000); + + assert_eq!(state.selected, 0); +} + +#[test] +fn monitor_tick_triggers_polling_interval() { + let mut state = AppState::default(); + state.screen_stack = vec![ScreenId::Monitor]; + state.monitor.last_poll_at = Some(1000); + + let (state, effects) = app::update(state, Action::Tick { now_ms: 1500 }); + assert!(effects.is_empty()); + assert_eq!(state.monitor.last_poll_at, Some(1000)); + + let (_state, effects) = app::update(state, Action::Tick { now_ms: 2001 }); + assert_eq!(effects, vec![Effect::Backend(BackendRequest::ListClusters)]); +} + +#[test] +fn monitor_open_selected_pushes_cluster() { + let mut state = AppState::default(); + state + .monitor + .set_clusters(vec![cluster_summary("c1", 0)], 1000); + + let (_state, effects) = app::update( + state, + Action::Screen(ScreenAction::Monitor(monitor::Action::OpenSelected)), + ); + + assert_eq!( + effects, + vec![ + Effect::Backend(BackendRequest::GetClusterSummary { + cluster_id: "c1".to_string() + }), + Effect::Backend(BackendRequest::GetClusterTopology { + cluster_id: "c1".to_string() + }), + Effect::Backend(BackendRequest::SubscribeClusterLogs { + cluster_id: "c1".to_string(), + agent_id: None, + }), + Effect::Backend(BackendRequest::SubscribeClusterTimeline { + cluster_id: "c1".to_string() + }), + ] + ); +} + +#[test] +fn monitor_last_activity_updates_only_on_increase() { + let mut state = monitor::State::default(); + state.set_clusters( + vec![cluster_summary("c1", 1), cluster_summary("c2", 2)], + 1000, + ); + + assert_eq!(state.last_activity_at.get("c1"), Some(&1000)); + assert_eq!(state.last_activity_at.get("c2"), Some(&1000)); + + state.set_clusters( + vec![cluster_summary("c1", 1), cluster_summary("c2", 2)], + 2000, + ); + + assert_eq!(state.last_activity_at.get("c1"), Some(&1000)); + assert_eq!(state.last_activity_at.get("c2"), Some(&1000)); + + state.set_clusters( + vec![cluster_summary("c1", 3), cluster_summary("c2", 2)], + 3000, + ); + + assert_eq!(state.last_activity_at.get("c1"), Some(&3000)); + assert_eq!(state.last_activity_at.get("c2"), Some(&1000)); +} diff --git a/tui-rs/crates/zeroshot-tui/tests/protocol_fixtures.rs b/tui-rs/crates/zeroshot-tui/tests/protocol_fixtures.rs new file mode 100644 index 00000000..5d8a1a95 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/protocol_fixtures.rs @@ -0,0 +1,85 @@ +use std::fs; +use std::path::PathBuf; + +use serde::de::DeserializeOwned; +use serde::Serialize; + +use zeroshot_tui::protocol::*; + +fn fixtures_dir() -> PathBuf { + PathBuf::from(env!("CARGO_MANIFEST_DIR")) + .join("../../..") + .join("tests/fixtures/tui-v2/protocol") +} + +fn round_trip(path: &PathBuf) { + let raw = fs::read_to_string(path).expect("read fixture"); + let parsed: T = serde_json::from_str(&raw).expect("deserialize fixture"); + let serialized = serde_json::to_string(&parsed).expect("serialize fixture"); + let reparsed: T = serde_json::from_str(&serialized).expect("re-deserialize fixture"); + assert_eq!(parsed, reparsed, "round-trip mismatch for {:?}", path); +} + +#[test] +fn protocol_fixtures_round_trip() { + let dir = fixtures_dir(); + let entries = fs::read_dir(&dir).expect("read fixtures dir"); + + for entry in entries { + let entry = entry.expect("read entry"); + let path = entry.path(); + let file_name = entry.file_name(); + let file_name = file_name.to_string_lossy(); + + if !file_name.ends_with(".json") { + continue; + } + if file_name.starts_with("invalid.") { + continue; + } + + let parts: Vec<&str> = file_name.split('.').collect(); + if parts.len() < 3 { + continue; + } + + match parts[0] { + "request" => match parts[1] { + "initialize" => round_trip::(&path), + "listClusters" => round_trip::(&path), + "getClusterSummary" => round_trip::(&path), + "listClusterMetrics" => round_trip::(&path), + "startClusterFromText" => round_trip::(&path), + "startClusterFromIssue" => round_trip::(&path), + "sendGuidanceToAgent" => round_trip::(&path), + "sendGuidanceToCluster" => round_trip::(&path), + "subscribeClusterLogs" => round_trip::(&path), + "subscribeClusterTimeline" => round_trip::(&path), + "unsubscribe" => round_trip::(&path), + "getClusterTopology" => round_trip::(&path), + other => panic!("unknown request fixture: {other}"), + }, + "response" => match parts[1] { + "initialize" => round_trip::(&path), + "listClusters" => round_trip::(&path), + "getClusterSummary" => round_trip::(&path), + "listClusterMetrics" => round_trip::(&path), + "startClusterFromText" => round_trip::(&path), + "startClusterFromIssue" => round_trip::(&path), + "sendGuidanceToAgent" => round_trip::(&path), + "sendGuidanceToCluster" => round_trip::(&path), + "subscribeClusterLogs" => round_trip::(&path), + "subscribeClusterTimeline" => round_trip::(&path), + "unsubscribe" => round_trip::(&path), + "getClusterTopology" => round_trip::(&path), + other => panic!("unknown response fixture: {other}"), + }, + "notification" => match parts[1] { + "clusterLogLines" => round_trip::(&path), + "clusterTimelineEvents" => round_trip::(&path), + other => panic!("unknown notification fixture: {other}"), + }, + other => panic!("unknown fixture prefix: {other}"), + } + } +} diff --git a/tui-rs/crates/zeroshot-tui/tests/radar_layout.rs b/tui-rs/crates/zeroshot-tui/tests/radar_layout.rs new file mode 100644 index 00000000..e6c192c9 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/radar_layout.rs @@ -0,0 +1,39 @@ +use zeroshot_tui::protocol::ClusterSummary; +use zeroshot_tui::screens::radar::{layout_position, FleetRadarState}; + +fn cluster_summary(id: &str, message_count: i64) -> ClusterSummary { + ClusterSummary { + id: id.to_string(), + state: "running".to_string(), + provider: None, + created_at: 0, + agent_count: 0, + message_count, + cwd: None, + } +} + +#[test] +fn radar_layout_is_deterministic() { + let first = layout_position("cluster-1", 10_000); + let second = layout_position("cluster-1", 10_000); + assert_eq!(first, second); +} + +#[test] +fn radar_selects_first_cluster() { + let mut state = FleetRadarState::default(); + state.set_clusters( + vec![ + cluster_summary("c2", 0), + cluster_summary("c1", 0), + cluster_summary("c3", 0), + ], + 1000, + ); + assert_eq!(state.selected_cluster_id(), Some("c2".to_string())); + + state.selected = 2; + state.set_clusters(vec![cluster_summary("c1", 0)], 2000); + assert_eq!(state.selected_cluster_id(), Some("c1".to_string())); +} diff --git a/tui-rs/crates/zeroshot-tui/tests/radar_snapshots.rs b/tui-rs/crates/zeroshot-tui/tests/radar_snapshots.rs new file mode 100644 index 00000000..35d31529 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/radar_snapshots.rs @@ -0,0 +1,54 @@ +mod ui_snapshot_helpers; + +use ui_snapshot_helpers::render_to_text; + +use zeroshot_tui::app::{AppState, ScreenId, UiVariant}; +use zeroshot_tui::protocol::ClusterSummary; +use zeroshot_tui::ui; + +fn cluster_summary(id: &str, state: &str) -> ClusterSummary { + ClusterSummary { + id: id.to_string(), + state: state.to_string(), + provider: None, + created_at: 0, + agent_count: 0, + message_count: 0, + cwd: None, + } +} + +#[test] +fn radar_snapshot_empty_state() { + let mut state = AppState::default(); + state.ui_variant = UiVariant::Disruptive; + state.screen_stack = vec![ScreenId::FleetRadar]; + + let content = render_to_text(60, 12, |frame| ui::render(frame, &state)); + assert!(content.contains("Fleet Radar")); + assert!(content.contains("No clusters yet.")); + assert!(content.contains("Type an intent in the spine")); +} + +#[test] +fn radar_snapshot_three_cluster_states() { + let mut state = AppState::default(); + state.ui_variant = UiVariant::Disruptive; + state.screen_stack = vec![ScreenId::FleetRadar]; + state.now_ms = 10_000; + state.fleet_radar.set_clusters( + vec![ + cluster_summary("run-1", "running"), + cluster_summary("done-1", "done"), + cluster_summary("err-1", "error"), + ], + state.now_ms, + ); + + let content = render_to_text(80, 16, |frame| ui::render(frame, &state)); + assert!(content.contains("Fleet Radar")); + assert!(content.contains("run-1")); + assert!(content.contains("done-1")); + assert!(content.contains("err-1")); + assert!(!content.contains("No clusters yet.")); +} diff --git a/tui-rs/crates/zeroshot-tui/tests/scrub_bar_snapshots.rs b/tui-rs/crates/zeroshot-tui/tests/scrub_bar_snapshots.rs new file mode 100644 index 00000000..aa2c2147 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/scrub_bar_snapshots.rs @@ -0,0 +1,74 @@ +mod ui_snapshot_helpers; + +use ui_snapshot_helpers::render_to_text; + +use zeroshot_tui::app::{TimeCursor, TimeCursorMode}; +use zeroshot_tui::protocol::ClusterLogLine; +use zeroshot_tui::ui::shared::TimeIndexedBuffer; +use zeroshot_tui::ui::widgets::scrub_bar::{self, ScrubBarState}; + +fn sample_logs(timestamps: &[i64]) -> TimeIndexedBuffer { + let mut buffer = TimeIndexedBuffer::new(64); + let lines = timestamps.iter().map(|ts| ClusterLogLine { + id: format!("log-{ts}"), + timestamp: *ts, + text: "event".to_string(), + agent: None, + role: None, + sender: None, + }); + buffer.push_many(lines); + buffer +} + +#[test] +fn scrub_bar_snapshot_live_mode() { + let logs = sample_logs(&[100, 200, 300]); + let cursor = TimeCursor { + mode: TimeCursorMode::Live, + t_ms: 300, + window_ms: 300, + }; + + let content = render_to_text(40, 1, |frame| { + let area = frame.area(); + scrub_bar::render( + frame, + area, + ScrubBarState { + time_cursor: &cursor, + logs: Some(&logs), + agent_id: None, + }, + ); + }); + + assert!(content.contains("LIVE")); + assert!(content.contains("|")); +} + +#[test] +fn scrub_bar_snapshot_scrub_mode() { + let logs = sample_logs(&[100, 200, 300]); + let cursor = TimeCursor { + mode: TimeCursorMode::Scrub, + t_ms: 150, + window_ms: 300, + }; + + let content = render_to_text(40, 1, |frame| { + let area = frame.area(); + scrub_bar::render( + frame, + area, + ScrubBarState { + time_cursor: &cursor, + logs: Some(&logs), + agent_id: None, + }, + ); + }); + + assert!(content.contains("SCRUB")); + assert!(content.contains("*") || content.contains("^")); +} diff --git a/tui-rs/crates/zeroshot-tui/tests/spine_snapshots.rs b/tui-rs/crates/zeroshot-tui/tests/spine_snapshots.rs new file mode 100644 index 00000000..1e264909 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/spine_snapshots.rs @@ -0,0 +1,74 @@ +mod ui_snapshot_helpers; + +use ui_snapshot_helpers::render_to_text; + +use zeroshot_tui::app::{SpineCompletion, SpineHint, SpineHintTone, SpineMode, SpineState}; +use zeroshot_tui::ui::widgets::spine; + +const WIDTH: u16 = 72; +const HEIGHT: u16 = 3; + +fn render_spine(state: &SpineState) -> String { + render_to_text(WIDTH, HEIGHT, |frame| { + let area = frame.area(); + spine::render(frame, area, state); + }) +} + +#[test] +fn spine_intent_mode_with_hint() { + let mut state = SpineState::default(); + state.hint = SpineHint::new("Ready", SpineHintTone::Info); + + let content = render_spine(&state); + assert!(content.contains("Intent")); + assert!(content.contains("Type intent...")); + assert!(content.contains("Ready")); +} + +#[test] +fn spine_command_mode_with_completion() { + let mut state = SpineState::default(); + state.mode = SpineMode::Command; + state.input.input = "pin".to_string(); + state.input.cursor = state.input.input.len(); + state.completion = Some(SpineCompletion { + candidates: vec!["cluster-1".to_string()], + selected: 0, + ghost: " cluster-1".to_string(), + }); + state.hint = SpineHint::new("Pin focus", SpineHintTone::Muted); + + let content = render_spine(&state); + assert!(content.contains("Command")); + assert!(content.contains("/pin cluster-1")); + assert!(content.contains("Pin focus")); +} + +#[test] +fn spine_whisper_cluster_mode() { + let mut state = SpineState::default(); + state.mode = SpineMode::WhisperCluster; + state.input.input = "cluster-7".to_string(); + state.input.cursor = state.input.input.len(); + state.hint = SpineHint::new("Whisper to cluster", SpineHintTone::Info); + + let content = render_spine(&state); + assert!(content.contains("Whisper Cluster")); + assert!(content.contains("cluster-7")); + assert!(content.contains("Whisper to cluster")); +} + +#[test] +fn spine_whisper_agent_mode() { + let mut state = SpineState::default(); + state.mode = SpineMode::WhisperAgent; + state.input.input = "agent-2".to_string(); + state.input.cursor = state.input.input.len(); + state.hint = SpineHint::new("Whisper to agent", SpineHintTone::Info); + + let content = render_spine(&state); + assert!(content.contains("Whisper Agent")); + assert!(content.contains("agent-2")); + assert!(content.contains("Whisper to agent")); +} diff --git a/tui-rs/crates/zeroshot-tui/tests/spine_submit.rs b/tui-rs/crates/zeroshot-tui/tests/spine_submit.rs new file mode 100644 index 00000000..f2cd3f7f --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/spine_submit.rs @@ -0,0 +1,87 @@ +use zeroshot_tui::app::{ + self, Action, AppState, BackendRequest, Effect, ScreenId, SpineAction, SpineMode, +}; + +#[test] +fn intent_submit_text_starts_cluster_from_text() { + let mut state = AppState::default(); + state.spine.mode = SpineMode::Intent; + state.spine.input.input = "build".to_string(); + state.spine.input.cursor = 5; + + let (state, effects) = app::update(state, Action::Spine(SpineAction::Submit)); + + assert!( + effects.contains(&Effect::Backend(BackendRequest::StartClusterFromText { + text: "build".to_string(), + provider_override: None, + })) + ); + assert_eq!(state.spine.input.input, ""); + assert_eq!(state.spine.mode, SpineMode::Intent); +} + +#[test] +fn intent_submit_issue_starts_cluster_from_issue() { + let mut state = AppState::default(); + state.spine.mode = SpineMode::Intent; + state.spine.input.input = "org/repo#42".to_string(); + state.spine.input.cursor = 11; + + let (state, effects) = app::update(state, Action::Spine(SpineAction::Submit)); + + assert!( + effects.contains(&Effect::Backend(BackendRequest::StartClusterFromIssue { + reference: "org/repo#42".to_string(), + provider_override: None, + })) + ); + assert_eq!(state.spine.input.input, ""); + assert_eq!(state.spine.mode, SpineMode::Intent); +} + +#[test] +fn whisper_cluster_submit_sends_guidance_to_cluster() { + let mut state = AppState::default(); + state.screen_stack = vec![ScreenId::ClusterCanvas { + id: "cluster-1".to_string(), + }]; + state.spine.mode = SpineMode::WhisperCluster; + state.spine.input.input = "ping".to_string(); + state.spine.input.cursor = 4; + + let (state, effects) = app::update(state, Action::Spine(SpineAction::Submit)); + + assert!( + effects.contains(&Effect::Backend(BackendRequest::SendGuidanceToCluster { + cluster_id: "cluster-1".to_string(), + message: "ping".to_string(), + })) + ); + assert_eq!(state.spine.input.input, ""); + assert_eq!(state.spine.mode, SpineMode::Intent); +} + +#[test] +fn whisper_agent_submit_sends_guidance_to_agent() { + let mut state = AppState::default(); + state.screen_stack = vec![ScreenId::AgentMicroscope { + cluster_id: "cluster-1".to_string(), + agent_id: "agent-1".to_string(), + }]; + state.spine.mode = SpineMode::WhisperAgent; + state.spine.input.input = "ping".to_string(); + state.spine.input.cursor = 4; + + let (state, effects) = app::update(state, Action::Spine(SpineAction::Submit)); + + assert!( + effects.contains(&Effect::Backend(BackendRequest::SendGuidanceToAgent { + cluster_id: "cluster-1".to_string(), + agent_id: "agent-1".to_string(), + message: "ping".to_string(), + })) + ); + assert_eq!(state.spine.input.input, ""); + assert_eq!(state.spine.mode, SpineMode::WhisperAgent); +} diff --git a/tui-rs/crates/zeroshot-tui/tests/startup_options.rs b/tui-rs/crates/zeroshot-tui/tests/startup_options.rs new file mode 100644 index 00000000..7dc2ced7 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/startup_options.rs @@ -0,0 +1,20 @@ +use zeroshot_tui::app::{AppState, InitialScreen, ScreenId, StartupOptions, UiVariant}; + +#[test] +fn startup_options_apply_monitor_and_provider_override() { + let mut state = AppState::default(); + let options = StartupOptions { + initial_screen: Some(InitialScreen::Monitor), + provider_override: Some("codex".to_string()), + ui_variant: Some(UiVariant::Disruptive), + }; + + state.apply_startup_options(options); + + assert_eq!( + state.screen_stack, + vec![ScreenId::IntentConsole, ScreenId::FleetRadar] + ); + assert_eq!(state.provider_override, Some("codex".to_string())); + assert_eq!(state.ui_variant, UiVariant::Disruptive); +} diff --git a/tui-rs/crates/zeroshot-tui/tests/topology_widget.rs b/tui-rs/crates/zeroshot-tui/tests/topology_widget.rs new file mode 100644 index 00000000..f5e3100a --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/topology_widget.rs @@ -0,0 +1,82 @@ +use zeroshot_tui::protocol::{ClusterTopology, TopologyAgent, TopologyEdge, TopologyEdgeKind}; +use zeroshot_tui::ui::widgets::topology; + +fn sample_topology() -> ClusterTopology { + ClusterTopology { + agents: vec![ + TopologyAgent { + id: "worker".to_string(), + role: Some("implementation".to_string()), + }, + TopologyAgent { + id: "system".to_string(), + role: None, + }, + TopologyAgent { + id: "validator".to_string(), + role: Some("validator".to_string()), + }, + ], + edges: vec![ + TopologyEdge { + from: "worker".to_string(), + to: "IMPLEMENTATION_READY".to_string(), + topic: "IMPLEMENTATION_READY".to_string(), + kind: TopologyEdgeKind::Publish, + dynamic: None, + }, + TopologyEdge { + from: "ISSUE_OPENED".to_string(), + to: "worker".to_string(), + topic: "ISSUE_OPENED".to_string(), + kind: TopologyEdgeKind::Trigger, + dynamic: Some(true), + }, + TopologyEdge { + from: "system".to_string(), + to: "ISSUE_OPENED".to_string(), + topic: "ISSUE_OPENED".to_string(), + kind: TopologyEdgeKind::Source, + dynamic: None, + }, + ], + topics: vec![ + "IMPLEMENTATION_READY".to_string(), + "ISSUE_OPENED".to_string(), + ], + } +} + +#[test] +fn renders_sorted_edges() { + let topology = sample_topology(); + let lines = topology::build_lines(None, Some(&topology), None); + + assert_eq!( + lines, + vec![ + "Summary pending.", + "Agents: 3 | Topics: 2 | Edges: 3", + "ISSUE_OPENED:", + " -> worker (trigger:ISSUE_OPENED dynamic)", + "system:", + " -> ISSUE_OPENED (source:ISSUE_OPENED)", + "worker:", + " -> IMPLEMENTATION_READY (publish:IMPLEMENTATION_READY)", + "Tab/Shift+Tab or h/l (Left/Right) to switch panes", + ] + ); +} + +#[test] +fn renders_placeholder_on_error() { + let lines = topology::build_lines(None, None, Some("backend unavailable")); + assert_eq!( + lines, + vec![ + "Summary pending.", + "Topology unavailable: backend unavailable", + "Tab/Shift+Tab or h/l (Left/Right) to switch panes", + ] + ); +} diff --git a/tui-rs/crates/zeroshot-tui/tests/ui_snapshot_helpers.rs b/tui-rs/crates/zeroshot-tui/tests/ui_snapshot_helpers.rs new file mode 100644 index 00000000..693ceeb8 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/ui_snapshot_helpers.rs @@ -0,0 +1,39 @@ +use ratatui::backend::TestBackend; +use ratatui::buffer::Buffer; +use ratatui::Frame; +use ratatui::Terminal; + +pub fn buffer_lines(buffer: &Buffer) -> Vec { + let area = buffer.area; + let mut lines = Vec::new(); + for y in area.top()..area.bottom() { + let mut line = String::new(); + for x in area.left()..area.right() { + line.push_str(buffer.cell((x, y)).map_or("", |c| c.symbol())); + } + lines.push(line); + } + lines +} + +pub fn buffer_text(buffer: &Buffer) -> String { + buffer_lines(buffer).join("\n") +} + +pub fn render_to_buffer(width: u16, height: u16, draw: F) -> Buffer +where + F: FnOnce(&mut Frame<'_>), +{ + let backend = TestBackend::new(width, height); + let mut terminal = Terminal::new(backend).expect("terminal"); + terminal.draw(draw).expect("draw"); + terminal.backend().buffer().clone() +} + +pub fn render_to_text(width: u16, height: u16, draw: F) -> String +where + F: FnOnce(&mut Frame<'_>), +{ + let buffer = render_to_buffer(width, height, draw); + buffer_text(&buffer) +} diff --git a/tui-rs/crates/zeroshot-tui/tests/ui_variant_parsing.rs b/tui-rs/crates/zeroshot-tui/tests/ui_variant_parsing.rs new file mode 100644 index 00000000..dcef1524 --- /dev/null +++ b/tui-rs/crates/zeroshot-tui/tests/ui_variant_parsing.rs @@ -0,0 +1,25 @@ +use zeroshot_tui::app::{resolve_ui_variant, UiVariant}; + +#[test] +fn ui_variant_defaults_to_none() { + let result = resolve_ui_variant(None, None).expect("resolve"); + assert_eq!(result, None); +} + +#[test] +fn ui_variant_parses_case_insensitive() { + let result = resolve_ui_variant(None, Some("Disruptive")).expect("resolve"); + assert_eq!(result, Some(UiVariant::Disruptive)); +} + +#[test] +fn ui_variant_cli_overrides_env() { + let result = resolve_ui_variant(Some("classic"), Some("disruptive")).expect("resolve"); + assert_eq!(result, Some(UiVariant::Classic)); +} + +#[test] +fn ui_variant_rejects_unknown() { + let err = resolve_ui_variant(Some("weird"), None).expect_err("expected error"); + assert!(err.contains("Unknown UI variant")); +}