AI coding tools write code fast. But when it's time to review: what was the intent? what files changed? did tests pass? what risks remain?
ForgeDesk captures your local git changes and prepares review-ready material — summary, PR body, test evidence, risk hints, review context. You get a folder of structured files that can go straight into a PR or handoff.
It does not review code for you. It prepares the material so humans (or AI reviewers) can inspect the change faster.
npm install -g forgedeskforgedesk setup # first time: initialize and configure
forgedesk next # daily: auto-capture changes and generate evidenceThat's it. forgedesk next runs one safe step at a time — capture, generate evidence, check readiness, export. Run it again after making changes or recording tests.
forgedesk setup # one-time setup
# ... make code changes, run your tests ...
forgedesk test -- npm test # record test results
forgedesk next # auto-capture → evidence → exportAfter forgedesk next finishes, your PR body, review context, and evidence pack are ready in .forgedesk/exports/.
.forgedesk/evidence/<session>/
├── SUMMARY.md # change summary
├── PR_BODY.md # ready to paste into a PR
├── REVIEW_CONTEXT.md # structured context for reviewers
├── TEST_EVIDENCE.md # test results summary
├── PR_EVIDENCE.md # decisions, risks, manual checks
├── evidence.json # machine-readable evidence pack
└── ...
| Command | What it does |
|---|---|
forgedesk setup |
First-time setup |
forgedesk next |
Run the next safe step |
forgedesk doctor |
Check project health |
forgedesk context |
Generate AI-friendly context file |
forgedesk rules --preset security |
Install security risk rules |
forgedesk templates --init |
Generate customizable templates |
See docs/commands.md for the full command reference.
- Auto-capture: Reads your git diff and captures changed files, branch, commits
- Evidence generation: Produces structured Markdown and JSON evidence packs
- Risk detection: Flags auth, payment, config, database changes; detects hardcoded secrets and eval usage in diffs
- Evidence Score: Deterministic 0-7 quality metric for readiness
- Configurable rules: Add custom risk rules via
.forgedesk/rules.jsonor install presets - Custom templates: Override PR body, summary, and review context templates with your own
- Workspace support: Discovers test scripts across pnpm workspace packages
- CI gate:
forgedesk ci checkandforgedesk ci validatefor CI pipelines
- No AI calls — ForgeDesk does not call any AI provider
- No code review — it prepares material, not verdicts
- No auto-commit/push/PR — all git actions are explicit
- No cloud sync — everything stays local
- No background services — runs on demand
- Command Reference — all CLI commands and options
- API Contract — JSON schema stability guarantees
- Evidence Schema — evidence.json JSON Schema
- Reviewer Guide — how to inspect an evidence pack
- Troubleshooting — common errors and fixes
- Roadmap — planning reference
ForgeDesk v1.0.0 is the stable release. All --json output schemas are frozen under semver.
Requires Node.js 20+. Local development uses pnpm:
pnpm install
pnpm typecheck
pnpm test
pnpm build
pnpm smokeSee docs/publishing.md for the release checklist.