fab-kit
fab-kit is a development toolkit for AI-assisted coding. It includes a 7-stage pipeline (intake → spec → apply → review → hydrate → ship → review-PR), standalone CLI tools, and batch orchestration for running multiple AI agents in parallel.
Plain markdown prompts. No SDK. No vendor lock-in.
The 7 stages
Section titled “The 7 stages”| # | Stage | Purpose |
|---|---|---|
| 1 | Intake | Capture intent, scope, approach |
| 2 | Spec | Define requirements and acceptance criteria |
| 3 | Apply | Generate plan, execute the tasks |
| 4 | Review | Sub-agent validates against spec and constitution |
| 5 | Hydrate | Save learnings into project memory |
| 6 | Ship | Commit, push, create PR |
| 7 | Review-PR | Triage and fix PR review comments |
Key ideas
Section titled “Key ideas”- Self-contained change folders — each change has its own spec, plan, and status under
fab/changes/. - Project constitution —
fab/project/constitution.mddefines MUST/SHOULD/MUST NOT rules; every spec, plan, and review checks against it. - Shared memory that grows — design decisions hydrate into
docs/memory/, loaded as context for future changes. - SRAD autonomy framework — confidence-scored decisions: proceed silently, proceed with marker, or block and ask.
- Works with any agent — Claude Code, Codex, Cursor, Windsurf. The contract is markdown files, not an SDK.
When to use it
Section titled “When to use it”- You want AI to do real implementation work, not just autocomplete
- You’re tired of agents producing plausible-looking-but-wrong code
- You need a paper trail for AI-driven changes (specs, plans, review reports — all in git)
Next: install fab-kit, or read the commands reference.