MOD-007 · Factory OrchestrationPlanned
Checker Agent Framework
The Checker Agent Framework defines how independent AI or human checkers review builder work, verify evidence, detect false-green results, block unsafe changes, and produce a clear founder-readable verdict.
What it does
- Review builder output and evidence without building new features.
- Verify task completion against the lane contract and done meaning.
- Verify protected and forbidden files were not touched.
- Verify commands, tests, blocked actions, risk, and approval review are present.
- Detect missing evidence and mark it as blocking when required.
- Detect false-green tests that do not prove the real outcome.
What it never does
- Never check the same lane it built or contributed changes to.
- Never build new features while checking.
- Never modify code, tests, or evidence to make the work pass.
- Never approve without evidence or treat passing tests as complete proof.
Roadmap
Build phases
This module is delivered in disciplined phases. Each phase is proof-gated before the next begins — the light shows the intended flow, not completion.
- Manual checker verdict template
- Read-only checker dry run
- Evidence completeness checker
- Protected file & forbidden action checker
- Risk & sensitivity checker
- False-green detection
- Gate bypass detection
- UI evidence & witness review
- Terminal-0 merge handoff
Connected in the harness
MOD-007 is a design roadmap. Pre-launch · Proof-gated · Not commercially available. Roadmap accepted does not equal product built. Internal proof does not equal outside proof.