- Review target is PR diff only.
- If a risk exists outside diff, mark it as out-of-scope risk.
- Do not claim verification for code not present in diff.
- Report all priorities: P0, P1, P2, P3.
- Each finding must include risk score (1-10).
- Security
- Correctness / Data integrity
- Error handling
- Performance
- Maintainability
- Test coverage
- Logging / observability
- Architecture consistency
- Every finding must include:
- file path
- line number
- short code snippet
- If evidence is weak, mark as possible false positive.
- Every finding must include:
- what is wrong
- impact
- minimal fix approach
- suggested fixed code snippet or pseudo patch
- Security: X/10
- Correctness: X/10
- Performance: X/10
- Reliability: X/10
- Maintainability: X/10
- Test Coverage: X/10
- Overall: X/10
- Add 1-3 deduction reasons per score.
- Findings (high to low risk)
- Open questions / assumptions
- Scorecard
- Top 5 fix priorities
- Merge duplicate findings into one root-cause item.
- For Jules/Copilot generated code, prioritize checks for:
- TODO or placeholder logic in runtime path
- weak input validation
- broad exception swallowing
- missing authorization checks
- insecure defaults
- generated code not used by runtime
- Respond in Japanese.
- Keep output concrete and technical.