Overview

SpecForge is a spec-driven AI development lifecycle for Claude Code, delivered as one installable plugin (and listable on skills.sh). It fuses three ideas that no other kit combines:

• spec-kit — executable specs: the constitution → specify → clarify → plan → tasks → analyze → implement pipeline that produces committed, PR-traveling artifacts.
• AWS AI-DLC — a governed lifecycle whose core principle is “AI produces, humans approve; approval is the bottleneck” — but it ships only as prose rules, so its gates are advisory.
• Claude Code native primitives — Skills, Commands, Subagents, and Hooks.

The differentiator

Most spec-driven tools make the “stop and let a human approve” step advisory — a well-behaved model follows it; a hurried one blows right through. SpecForge makes the gate real: a PreToolUse hook physically blocks code edits until the active feature’s spec and plan are approved. That turns “AI proposes, a human approves” from a hope into an invariant.

What’s in the box

• 15 skills — the lifecycle: init, constitution, specify, clarify, plan, tasks, analyze, implement, loop, operate, reverse-engineer, approve, quick, status, resume.
• 5 subagents — spec-analyst, architect, implementer, reviewer, reverse-engineer.
• Enforcement hooks — a gate on code edits, plus session/state context injection and artifact validation.
• A committed spec store — specforge/{constitution,state,audit,inputs,specs} that travels with your pull requests.

Trust & design

MIT-licensed · local-first · no telemetry · every gate transition is logged in specforge/audit.md · 28 tests run in CI · and the repo builds itself with itself.

Ready? Head to Installation.