The harness for your AI coding agent.
✗ Agent rewrites files outside the goal
✓ Scope is locked to the blueprint. Nothing leaks.
✗ Agent deletes failing tests instead of fixing bugs
✓ Every test must pass. Failures trigger repair, not evasion.
✗ Agent hallucinates steps, loses context mid-task
✓ Structured phases keep execution on track, start to finish.
✗ No way to know if the output actually works
✓ Build → test → verify. Automated repair loop on every run.
✗ Changes pile up with no history or rollback
✓ Every step is versioned. Any point is rewindable.