"Read with intent. Question with purpose. Document with care."

1. Create REVIEW.yml
2. Identify files to review
3. Define focus areas

1. List all changed files
2. Read PR/commit description
3. Note initial impressions

1. Read the code
2. Check against criteria
3. Note findings
4. Run relevant checks

1. Run tests
2. Run linters
3. Check regressions

1. Compile findings
2. Prioritize issues
3. Generate REVIEW.md
4. State recommendation

• Security — Injection, auth, sensitive data
• Correctness — Logic errors, edge cases
• Performance — N+1 queries, memory leaks
• Maintainability — Clarity, DRY, naming
• Style — Formatting, conventions

• Input validation
• Output encoding
• Authentication/authorization
• Sensitive data handling
• Injection vulnerabilities
• Timing attacks

• Logic errors
• Edge cases handled
• Null/undefined handling
• Error handling
• Race conditions
• Resource cleanup

• Code clarity
• Appropriate comments
• Consistent naming
• DRY (no duplication)
• Single responsibility
• Testability

• Algorithmic complexity
• Memory usage
• Database queries
• Caching
• Unnecessary operations

• Summary and counts
• Issues by severity
• Verification results
• Recommendation

• rubric — Explicit scoring criteria for code quality
• evaluator — Independent assessment pattern
• adversarial-committee — Multiple reviewers debating findings