code-review（代码审查）

Code Review

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

Systematic code analysis with evidence collection. Code review IS an adventure — the codebase is the dungeon, findings are clues.

Review Process

READ → NOTE ISSUES → CLASSIFY → REPORT

Step 1: Setup

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

Step 2: Overview

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

Step 3: Deep Review

For each file:

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

Step 4: Verification

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

Step 5: Synthesize

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

Finding Severity

Finding Types

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

Review Checklist

Security

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

Correctness

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

Maintainability

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

Performance

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

Core Files

REVIEW.yml

review:
  name: "PR #123: Add user authentication"
  status: "in_progress"

findings:
  blocking:
    - id: "B1"
      file: "src/auth/login.ts"
      line: 45
      type: "security"
      summary: "Timing attack vulnerability"

  important: []
  minor: []
  praise: []

verification:
  tests: { ran: true, passed: true }
  linter: { ran: true, passed: false, issues: 3 }

REVIEW.md

Formatted document with:

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

Verification Commands

tests:
  - "npm test"
  - "pytest"
  - "go test ./..."

linters:
  - "npm run lint"
  - "flake8"
  - "golangci-lint run"

Recommendation Output

See Also

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