Authoritative rules are in .github/agents/Developer.agent.md and SSOT docs.
- Add one failing test
- Prove failure
- Minimal implementation
- Run local target tests
- Run full suite
- Is behaviour explicit for happy/error/edge paths?
- Is async code cancellation-aware?
- Are boundaries respected and dependencies minimal?
- Are tests deterministic and behaviour-focused?
- Minimal focused diff
- No unrelated refactors
- Tests added/updated and passing
- Notes on assumptions and risk