code-review（代码审查）

Code Review Skill

Conduct a thorough code review for quality, security, and maintainability with severity-rated feedback.

When to Use

This skill activates when:

• User requests "review this code", "code review"
• Before merging a pull request
• After implementing a major feature
• User wants quality assessment

GPT-5.4 Guidance Alignment

• Default to concise, evidence-dense progress and completion reporting unless the user or risk level requires more detail.
• Treat newer user task updates as local overrides for the active workflow branch while preserving earlier non-conflicting constraints.
• If correctness depends on additional inspection, retrieval, execution, or verification, keep using the relevant tools until the review is grounded.
• Continue through clear, low-risk, reversible next steps automatically; ask only when the next step is materially branching, destructive, or preference-dependent.

Delegates to the code-reviewer and architect agents in parallel for a two-lane review:

1. Identify Changes

- Run git diff to find changed files - Determine scope of review (specific files or entire PR)

1. Launch Parallel Review Lanes

- code-reviewer lane - owns spec compliance, security, code quality, performance, and maintainability findings - architect lane - owns the devil's-advocate / design-tradeoff perspective - Both lanes run in parallel and produce distinct outputs before final synthesis

1. Review Categories

- Security - Hardcoded secrets, injection risks, XSS, CSRF - Code Quality - Function size, complexity, nesting depth - Performance - Algorithm efficiency, N+1 queries, caching - Best Practices - Naming, documentation, error handling - Maintainability - Duplication, coupling, testability

1. Severity Rating

- CRITICAL - Security vulnerability (must fix before merge) - HIGH - Bug or major code smell (should fix before merge) - MEDIUM - Minor issue (fix when possible) - LOW - Style/suggestion (consider fixing)

1. Architectural Status Contract

- CLEAR - No unresolved architectural blocker was found - WATCH - Non-blocking design/tradeoff concern that must appear in the final synthesis - BLOCK - Unresolved design concern that prevents a merge-ready verdict

1. Specific Recommendations

- File:line locations for each issue - Concrete fix suggestions - Code examples where applicable

1. Final Synthesis

- Combine the code-reviewer recommendation and the architect status into one final verdict - Deterministic merge gating rules: - If architect status is BLOCK, final recommendation is REQUEST CHANGES - Else if code-reviewer recommendation is REQUEST CHANGES, final recommendation is REQUEST CHANGES - Else if architect status is WATCH, final recommendation is COMMENT - Else final recommendation follows the code-reviewer lane - The final report must make architect blockers impossible to miss

Agent Delegation

delegate(
  role="code-reviewer",
  tier="THOROUGH",
  prompt="CODE REVIEW TASK

Review code changes for quality, security, and maintainability.

This is the code/spec/security lane. Do not absorb architectural ownership.

Scope: [git diff or specific files]

Review Checklist:
- Security vulnerabilities (OWASP Top 10)
- Code quality (complexity, duplication)
- Performance issues (N+1, inefficient algorithms)
- Best practices (naming, documentation, error handling)
- Maintainability (coupling, testability)

Output: Code review report with:
- Files reviewed count
- Issues by severity (CRITICAL, HIGH, MEDIUM, LOW)
- Specific file:line locations
- Fix recommendations
- Approval recommendation (APPROVE / REQUEST CHANGES / COMMENT)"
)

delegate(
  role="architect",
  tier="THOROUGH",
  prompt="ARCHITECTURE / DEVIL'S-ADVOCATE REVIEW TASK

Review the same code changes from the architecture/tradeoff perspective.

Scope: [git diff or specific files]

Focus:
- System boundaries and interfaces
- Hidden coupling or long-term maintainability risks
- Tradeoff tension the main reviewer might miss
- Strongest counterargument against approving as-is

Output:
- Architectural Status: CLEAR / WATCH / BLOCK
- File:line evidence for each concern
- Concrete tradeoff or design recommendation"
)

Run both lanes in parallel, then synthesize them with the deterministic rules above.

External Model Consultation (Preferred)

The code-reviewer agent SHOULD consult Codex for cross-validation.

Protocol

1. Form your OWN review FIRST - Complete the review independently
2. Consult for validation - Cross-check findings with Codex
3. Critically evaluate - Never blindly adopt external findings
4. Graceful fallback - Never block if tools unavailable

When to Consult

• Security-sensitive code changes
• Complex architectural patterns
• Unfamiliar codebases or languages
• High-stakes production code

When to Skip

• Simple refactoring
• Well-understood patterns
• Time-critical reviews
• Small, isolated changes

Tool Usage

Before first MCP tool use, call ToolSearch("mcp") to discover deferred MCP tools. Use mcp__x__ask_codex with agent_role: "code-reviewer". If ToolSearch finds no MCP tools, fall back to the code-reviewer agent.

Note: Codex calls can take up to 1 hour. Consider the review timeline before consulting.

Output Format

CODE REVIEW REPORT
==================

Files Reviewed: 8
Total Issues: 12
Architectural Status: WATCH

CRITICAL (0)
-----------
(none)

HIGH (0)
--------
(none)

MEDIUM (7)
----------
1. src/api/auth.ts:42
   Issue: Email normalization logic is duplicated instead of reusing the shared helper
   Risk: Validation rules can drift between authentication paths
   Fix: Route both paths through the shared normalization helper

2. src/components/UserProfile.tsx:89
   Issue: Derived permissions are recalculated on every render
   Risk: Avoidable work during profile refreshes
   Fix: Memoize the derived permissions list or compute it upstream

3. src/utils/validation.ts:15
   Issue: Form-layer and server-layer validation messages are defined separately
   Risk: User-facing validation guidance can become inconsistent
   Fix: Share one validation message helper across both call sites

LOW (5)
-------
...

ARCHITECTURE WATCHLIST
----------------------
- src/review/orchestrator.ts:88
  Concern: Review result synthesis relies on implicit ordering rather than an explicit blocker contract
  Status: WATCH
  Recommendation: Define deterministic merge gating before expanding reviewers

SYNTHESIS
---------
- code-reviewer recommendation: COMMENT
- architect status: WATCH
- final recommendation: COMMENT

RECOMMENDATION: COMMENT

Address any WATCH concerns before treating the change as merge-ready.

Review Checklist

The code-reviewer lane checks:

Security

• No hardcoded secrets (API keys, passwords, tokens)
• All user inputs sanitized
• SQL/NoSQL injection prevention
• XSS prevention (escaped outputs)
• CSRF protection on state-changing operations
• Authentication/authorization properly enforced

Code Quality

• Functions < 50 lines (guideline)
• Cyclomatic complexity < 10
• No deeply nested code (> 4 levels)
• No duplicate logic (DRY principle)
• Clear, descriptive naming

Performance

• No N+1 query patterns
• Appropriate caching where applicable
• Efficient algorithms (avoid O(n²) when O(n) possible)
• No unnecessary re-renders (React/Vue)

Best Practices

• Error handling present and appropriate
• Logging at appropriate levels
• Documentation for public APIs
• Tests for critical paths
• No commented-out code

Architect Lane Checklist

The architect lane checks:

• Boundary or interface changes are explicit
• New coupling/tradeoff risks are surfaced
• Long-horizon maintainability concerns are evidence-backed
• Architectural status is one of CLEAR, WATCH, or BLOCK
• Any BLOCK concern cites the reason merge-ready status should be withheld

Approval Criteria

APPROVE - code-reviewer returns APPROVE and architect status is CLEAR REQUEST CHANGES - code-reviewer returns REQUEST CHANGES or architect status is BLOCK COMMENT - code-reviewer returns COMMENT with architect status CLEAR, architect status is WATCH, or only LOW/MEDIUM improvements remain

Scenario Examples

Good: The user says continue after the workflow already has a clear next step. Continue the current branch of work instead of restarting or re-asking the same question.

Good: The user changes only the output shape or downstream delivery step (for example make a PR). Preserve earlier non-conflicting workflow constraints and apply the update locally.

Bad: The user says continue, and the workflow restarts discovery or stops before the missing verification/evidence is gathered.

Use with Other Skills

With Team:

/team "review recent auth changes and report findings"

Includes coordinated review execution across specialized agents.

With Ralph:

/ralph code-review then fix all issues

On the explicit Ralph path, review findings should flow into automatic fix follow-up without another permission prompt. Plain code-review itself remains read-only and does not promise auto-fix.

With Ultrawork:

/ultrawork review all files in src/

Parallel code review across multiple files.

Best Practices

• Review early - Catch issues before they compound
• Review often - Small, frequent reviews better than huge ones
• Address CRITICAL/HIGH first - Fix security and bugs immediately
• Consider context - Some "issues" may be intentional trade-offs
• Learn from reviews - Use feedback to improve coding practices