feat!: rebrand to deliberation v2 core + drop 1.x compat

.claude-plugin/marketplace.json

@@ -1,21 +1,21 @@
 {
-  "name": "antonbabenko-claude-delegator",
+  "name": "antonbabenko-deliberation",
   "owner": {
     "name": "Anton Babenko",
     "url": "https://github.com/antonbabenko"
   },
   "plugins": [
     {
-      "name": "claude-delegator",
+      "name": "deliberation",
       "source": "./",
-      "description": "GPT (Codex CLI), Gemini (bundled MCP bridge), and Grok (xAI HTTP bridge) expert subagents for Claude Code. Seven specialized experts: Architect, Plan Reviewer, Scope Analyst, Code Reviewer, Security Analyst, Researcher, Debugger.",
+      "description": "Use when you want a delegated second opinion or implementation from GPT (Codex), Gemini, Grok (xAI), or OpenRouter (config-driven, 400+ models) - seven expert subagents (Architect, Plan Reviewer, Scope Analyst, Code Reviewer, Security Analyst, Researcher, Debugger) and bundled ask-gpt/ask-gemini/ask-grok/ask-openrouter/ask-all/consensus commands, advisory (read-only) or implementation (write; Grok and OpenRouter are advisory-only).",
       "version": "1.18.0",
       "author": {
         "name": "Anton Babenko",
         "url": "https://github.com/antonbabenko"
       },
-      "homepage": "https://github.com/antonbabenko/claude-delegator",
-      "repository": "https://github.com/antonbabenko/claude-delegator",
+      "homepage": "https://github.com/antonbabenko/deliberation",
+      "repository": "https://github.com/antonbabenko/deliberation",
       "license": "MIT",
       "keywords": [
         "delegation",
@@ -27,11 +27,13 @@
         "google",
         "grok",
         "xai",
-        "files",
+        "openrouter",
         "experts",
         "subagents",
-        "orchestration"
+        "orchestration",
+        "code-review",
+        "architecture"
       ]
     }
   ]
-}
+}

.claude-plugin/mcp.json

@@ -0,0 +1,29 @@
+{
+  "mcpServers": {
+    "codex": {
+      "command": "codex",
+      "args": ["mcp-server"],
+      "type": "stdio"
+    },
+    "gemini": {
+      "command": "node",
+      "args": ["${CLAUDE_PLUGIN_ROOT}/server/gemini/index.js"],
+      "type": "stdio"
+    },
+    "grok": {
+      "command": "node",
+      "args": ["${CLAUDE_PLUGIN_ROOT}/server/grok/index.js"],
+      "type": "stdio"
+    },
+    "openrouter": {
+      "command": "node",
+      "args": ["${CLAUDE_PLUGIN_ROOT}/server/openrouter/index.js"],
+      "type": "stdio"
+    },
+    "deliberation": {
+      "command": "node",
+      "args": ["${CLAUDE_PLUGIN_ROOT}/server/mcp/index.js"],
+      "type": "stdio"
+    }
+  }
+}

.claude-plugin/plugin.json

@@ -1,13 +1,13 @@
 {
-  "name": "claude-delegator",
-  "description": "GPT (Codex CLI), Gemini (bundled MCP bridge), and Grok (xAI HTTP bridge) expert subagents for Claude Code. Seven specialized experts: Architect, Plan Reviewer, Scope Analyst, Code Reviewer, Security Analyst, Researcher, Debugger.",
+  "name": "deliberation",
+  "description": "Use when you want a delegated second opinion or implementation from GPT (Codex), Gemini, Grok (xAI), or OpenRouter (config-driven, 400+ models) - seven expert subagents (Architect, Plan Reviewer, Scope Analyst, Code Reviewer, Security Analyst, Researcher, Debugger) and bundled ask-gpt/ask-gemini/ask-grok/ask-openrouter/ask-all/consensus commands, advisory (read-only) or implementation (write; Grok and OpenRouter are advisory-only).",
   "version": "1.18.0",
   "author": {
     "name": "Anton Babenko",
     "url": "https://github.com/antonbabenko"
   },
-  "homepage": "https://github.com/antonbabenko/claude-delegator",
-  "repository": "https://github.com/antonbabenko/claude-delegator",
+  "homepage": "https://github.com/antonbabenko/deliberation",
+  "repository": "https://github.com/antonbabenko/deliberation",
   "license": "MIT",
   "keywords": ["delegation", "mcp", "codex", "gpt", "openai", "gemini", "antigravity", "agy", "google", "grok", "xai", "files", "experts", "subagents", "orchestration"]
 }

.github/workflows/validate.yml

@@ -26,6 +26,13 @@ jobs:
         uses: actions/setup-node@v4
         with:
           node-version: '20'
+          cache: 'npm'
+
+      - name: Install dependencies
+        run: npm ci
+
+      - name: Typecheck
+        run: npm run typecheck
 
       - name: Run bridge tests
         run: node --test test/*.test.js

CLAUDE.md

@@ -10,13 +10,13 @@ A Claude Code plugin that provides GPT (via Codex CLI), Gemini 3 (via the Antigr
 
 ```bash
 # Test plugin locally (loads from working directory)
-claude --plugin-dir /path/to/claude-delegator
+claude --plugin-dir /path/to/deliberation
 
 # Run setup to test installation flow
-/claude-delegator:setup
+/deliberation:setup
 
 # Run uninstall to test removal flow
-/claude-delegator:uninstall
+/deliberation:uninstall
 ```
 
 No build step, no dependencies. Codex exposes a native MCP server; Gemini, Grok, and OpenRouter use bundled zero-dependency Node bridges (`server/gemini/index.js`, `server/grok/index.js`, `server/openrouter/index.js`). The Gemini bridge wraps the Antigravity CLI (`agy`) in print mode. The OpenRouter bridge calls any OpenAI-compatible `/chat/completions` endpoint.
@@ -25,7 +25,7 @@ No build step, no dependencies. Codex exposes a native MCP server; Gemini, Grok,
 
 ### Orchestration Flow
 
-Claude acts as orchestrator—delegates to specialized experts based on task type. Supports both **single-shot** (independent calls) and **multi-turn** (context preserved via `threadId`).
+Claude acts as orchestrator - delegates to specialized experts based on task type. Supports both **single-shot** (independent calls) and **multi-turn** (context preserved via `threadId`).
 
 ```
 User Request → Claude Code → [Match trigger → Select expert & provider]
@@ -44,7 +44,7 @@ User Request → Claude Code → [Match trigger → Select expert & provider]
 1. **Match trigger** - Check `rules/triggers.md` for semantic patterns
 2. **Read expert prompt** - Load from `prompts/[expert].md`
 3. **Build 7-section prompt** - Use format from `rules/delegation-format.md`
-4. **Call provider tool** - `mcp__codex__codex`, `mcp__gemini__gemini`, `mcp__grok__grok`, or `mcp__openrouter__openrouter`
+4. **Call provider tool** - `mcp__deliberation-codex__codex`, `mcp__deliberation-gemini__gemini`, `mcp__deliberation-grok__grok`, or `mcp__deliberation-openrouter__openrouter`
 5. **Synthesize response** - Never show raw output; interpret and verify
 
 ### The 7-Section Delegation Format
@@ -62,11 +62,11 @@ Retries use multi-turn (`*-reply` with `threadId`) so the expert remembers previ
 
 | Component | Purpose | Notes |
 |-----------|---------|-------|
-| `rules/*.md` | When/how to delegate | Installed to `~/.claude/rules/delegator/` |
+| `rules/*.md` | When/how to delegate | Installed to `~/.claude/rules/deliberation/` |
 | `prompts/*.md` | Expert personalities | Injected via `developer-instructions` |
 | `commands/*.md` | Slash commands | `/setup`, `/uninstall` |
 | `config/providers.json` | Provider metadata | Not used at runtime |
-| `~/.claude/claude-delegator/config.json` | OpenRouter model config | Live SSOT; stat-gated hot-reload |
+| `~/.claude/deliberation/config.json` | OpenRouter model config | Live SSOT; stat-gated hot-reload |
 
 > Expert prompts adapted from [oh-my-opencode](https://github.com/code-yeongyu/oh-my-opencode)
 
@@ -82,15 +82,15 @@ Retries use multi-turn (`*-reply` with `threadId`) so the expert remembers previ
 | **Researcher** | `prompts/researcher.md` | External libraries, docs, best practices | "how do I use X", "find examples of Y" |
 | **Debugger** | `prompts/debugger.md` | Root-cause analysis, minimal fixes | "why does this crash", "debug this failing test" |
 
-Every expert can operate in **advisory** (`sandbox: read-only`) or **implementation** (`sandbox: workspace-write`) mode based on the task. OpenRouter models are always advisory - per-model expert eligibility is controlled by the `experts` field in `~/.claude/claude-delegator/config.json`.
+Every expert can operate in **advisory** (`sandbox: read-only`) or **implementation** (`sandbox: workspace-write`) mode based on the task. OpenRouter models are always advisory - per-model expert eligibility is controlled by the `experts` field in `~/.claude/deliberation/config.json`.
 
 ## Grok file access
 
-Grok reads attached files via `files[]` and resolves them under `roots[]` (top-level array of absolute directories) or `cwd`. `path` and `dir` entries take an optional `mode: "auto" | "inline" | "upload"` — inline embeds the file as `input_text` so Grok reads it line-by-line (best for source code); upload routes through the xAI Files API and is SHA-256 dedup-cached locally. `file_id` / `file_url` entries pass through unchanged and do not accept `mode`. Directory expansion via `{dir}` entries. See **[TECHNICAL.md: Grok files and cleanup](TECHNICAL.md#grok-files-and-cleanup)** for parameters, the inline-vs-upload tradeoff, cross-repo usage, cache layout, and the `gc` cleanup subcommand.
+Grok reads attached files via `files[]` and resolves them under `roots[]` (top-level array of absolute directories) or `cwd`. `path` and `dir` entries take an optional `mode: "auto" | "inline" | "upload"` - inline embeds the file as `input_text` so Grok reads it line-by-line (best for source code); upload routes through the xAI Files API and is SHA-256 dedup-cached locally. `file_id` / `file_url` entries pass through unchanged and do not accept `mode`. Directory expansion via `{dir}` entries. See **[TECHNICAL.md: Grok files and cleanup](TECHNICAL.md#grok-files-and-cleanup)** for parameters, the inline-vs-upload tradeoff, cross-repo usage, cache layout, and the `gc` cleanup subcommand.
 
 ## Key Design Decisions
 
-1. **Native & Bridge MCP** - Codex has a native `mcp-server` command. Gemini requires a bundled bridge (`server/gemini/index.js`) that wraps the Antigravity CLI (`agy`) in print mode. Grok has no MCP or CLI server mode, so a bundled bridge (`server/grok/index.js`) wraps the xAI **Responses API** (`/v1/responses`) directly - advisory-only (no file editing), but it can READ attached files (`files:[{path|file_id|file_url|dir}]`, optional `roots[]`, per-entry `mode` for upload-vs-inline delivery); uploaded files are SHA-256 dedup-cached locally, auto-expire (7-day default, `GROK_FILE_TTL_SECONDS`), and are managed with `/grok-files` (`server/grok/files-admin.js`: `list`/`prune`/`gc`). Details in [TECHNICAL.md § Grok files and cleanup](TECHNICAL.md#grok-files-and-cleanup). OpenRouter uses a bundled bridge (`server/openrouter/index.js`) that calls any OpenAI-compatible `POST {apiBase}/chat/completions` endpoint - advisory-only, text-inline file attachment only (`{path}`/`{dir}`; no upload path), config-driven via `~/.claude/claude-delegator/config.json`. Details in [TECHNICAL.md § OpenRouter bridge](TECHNICAL.md#openrouter-bridge).
+1. **Native & Bridge MCP** - Codex has a native `mcp-server` command. Gemini requires a bundled bridge (`server/gemini/index.js`) that wraps the Antigravity CLI (`agy`) in print mode. Grok has no MCP or CLI server mode, so a bundled bridge (`server/grok/index.js`) wraps the xAI **Responses API** (`/v1/responses`) directly - advisory-only (no file editing), but it can READ attached files (`files:[{path|file_id|file_url|dir}]`, optional `roots[]`, per-entry `mode` for upload-vs-inline delivery); uploaded files are SHA-256 dedup-cached locally, auto-expire (7-day default, `GROK_FILE_TTL_SECONDS`), and are managed with `/grok-files` (`server/grok/files-admin.js`: `list`/`prune`/`gc`). Details in [TECHNICAL.md § Grok files and cleanup](TECHNICAL.md#grok-files-and-cleanup). OpenRouter uses a bundled bridge (`server/openrouter/index.js`) that calls any OpenAI-compatible `POST {apiBase}/chat/completions` endpoint - advisory-only, text-inline file attachment only (`{path}`/`{dir}`; no upload path), config-driven via `~/.claude/deliberation/config.json`. Details in [TECHNICAL.md § OpenRouter bridge](TECHNICAL.md#openrouter-bridge).
 2. **Single-shot + multi-turn** - Single-shot for advisory (full context per call), multi-turn via `threadId` for chained implementation and retries
 3. **Dual mode** - Any expert can advise or implement based on task
 4. **Synthesize, don't passthrough** - Claude interprets expert output, applies judgment

CONTRIBUTING.md

@@ -1,4 +1,4 @@
-# Contributing to claude-delegator
+# Contributing to deliberation
 
 Contributions welcome. This document covers how to contribute effectively.
 
@@ -8,15 +8,15 @@ Contributions welcome. This document covers how to contribute effectively.
 
 ```bash
 # Clone the repo
-git clone https://github.com/antonbabenko/claude-delegator
-cd claude-delegator
+git clone https://github.com/antonbabenko/deliberation
+cd deliberation
 
 # Run `claude` and install plugin in Claude Code
 claude
-/claude-delegator:setup
+/deliberation:setup
 
 # Or test your changes locally without reinstalling
-claude --plugin-dir /path/to/claude-delegator
+claude --plugin-dir /path/to/deliberation
 ```
 
 ---
@@ -36,7 +36,7 @@ claude --plugin-dir /path/to/claude-delegator
 ## Project Structure
 
 ```
-claude-delegator/
+deliberation/
 ├── .claude-plugin/         # Plugin manifest
 │   └── plugin.json
 ├── commands/               # Slash commands (/setup, /uninstall)
@@ -53,7 +53,7 @@ claude-delegator/
 
 ### Before Submitting
 
-1. **Test your changes** - Run `/claude-delegator:setup` and verify
+1. **Test your changes** - Run `/deliberation:setup` and verify
 2. **Update docs** - If you change behavior, update relevant docs
 3. **Keep commits atomic** - One logical change per commit
 
@@ -94,7 +94,7 @@ You never bump versions by hand.
 3. It opens a `chore(release): vX.Y.Z` PR that updates `version.json`, `CHANGELOG.md`, and
    the synced manifests, then auto-merges it once the `validate` check passes.
 4. On merge, a second workflow tags `vX.Y.Z` and publishes the GitHub Release.
-5. The `antonbabenko/agent-plugins` marketplace then re-pins `claude-delegator` to the new
+5. The `antonbabenko/agent-plugins` marketplace then re-pins `deliberation` to the new
    release (immediately if its dispatch token is set, otherwise within a day via cron).
 
 `version.json` is the single source of truth. `.claude-plugin/plugin.json`,
@@ -163,8 +163,8 @@ check fails if they drift.
 After changes, verify with actual MCP calls:
 
 1. Install the plugin in Claude Code
-2. Run `/claude-delegator:setup`
-3. Verify MCP tools are available (`mcp__codex__codex`)
+2. Run `/deliberation:setup`
+3. Verify MCP tools are available (`mcp__deliberation-codex__codex`)
 4. Test MCP tool calls via oracle delegation
 5. Verify responses are properly synthesized
 6. Test error cases (timeout, missing CLI)