code-review（代码审查）

Code Review Checklist

Thorough, structured approach to reviewing code. Work through each dimension systematically rather than scanning randomly.

Installation

OpenClaw / Moltbot / Clawbot

npx clawhub@latest install code-review

Review Dimensions

Security Checklist

Review every change for these vulnerabilities:

• [ ] SQL Injection — All queries use parameterized statements or an ORM; no string concatenation with user input
• [ ] XSS — User-provided content is escaped/sanitized before rendering; dangerouslySetInnerHTML or equivalent is justified and safe
• [ ] CSRF Protection — State-changing requests require valid CSRF tokens; SameSite cookie attributes are set
• [ ] Authentication — Every protected endpoint verifies the user is authenticated before processing
• [ ] Authorization — Resource access is scoped to the requesting user's permissions; no IDOR vulnerabilities
• [ ] Input Validation — All external input (params, headers, body, files) is validated for type, length, format, and range on the server side
• [ ] Secrets Management — No API keys, passwords, tokens, or credentials in source code; secrets come from environment variables or a vault
• [ ] Dependency Safety — New dependencies are from trusted sources, actively maintained, and free of known CVEs
• [ ] Sensitive Data — PII, tokens, and secrets are never logged, included in error messages, or returned in API responses
• [ ] Rate Limiting — Public and auth endpoints have rate limits to prevent brute-force and abuse
• [ ] File Upload Safety — Uploaded files are validated for type and size, stored outside the webroot, and served with safe Content-Type headers
• [ ] HTTP Security Headers — Content-Security-Policy, X-Content-Type-Options, Strict-Transport-Security are set

Performance Checklist

• [ ] N+1 Queries — Database access patterns are batched or joined; no loops issuing individual queries
• [ ] Unnecessary Re-renders — Components only re-render when their relevant state/props change; memoization is applied where measurable
• [ ] Memory Leaks — Event listeners, subscriptions, timers, and intervals are cleaned up on unmount/disposal
• [ ] Bundle Size — New dependencies are tree-shakeable; large libraries are loaded dynamically; no full-library imports for a single function
• [ ] Lazy Loading — Heavy components, routes, and below-the-fold content use lazy loading / code splitting
• [ ] Caching Strategy — Expensive computations and API responses use appropriate caching (memoization, HTTP cache headers, Redis)
• [ ] Database Indexing — Queries filter/sort on indexed columns; new queries have been checked with EXPLAIN
• [ ] Pagination — List endpoints and queries use pagination or cursor-based fetching; no unbounded SELECT *
• [ ] Async Operations — Long-running tasks are offloaded to background jobs or queues rather than blocking request threads
• [ ] Image & Asset Optimization — Images are properly sized, use modern formats (WebP/AVIF), and leverage CDN delivery

Correctness Checklist

• [ ] Edge Cases — Empty arrays, empty strings, zero values, negative numbers, and maximum values are handled
• [ ] Null/Undefined Handling — Nullable values are checked before access; optional chaining or guards prevent runtime errors
• [ ] Off-by-One Errors — Loop bounds, array slicing, pagination offsets, and range calculations are verified
• [ ] Race Conditions — Concurrent access to shared state uses locks, transactions, or atomic operations
• [ ] Timezone Handling — Dates are stored in UTC; display conversion happens at the presentation layer
• [ ] Unicode & Encoding — String operations handle multi-byte characters; text encoding is explicit (UTF-8)
• [ ] Integer Overflow / Precision — Arithmetic on large numbers or currency uses appropriate types (BigInt, Decimal)
• [ ] Error Propagation — Errors from async calls and external services are caught and handled; promises are never silently swallowed
• [ ] State Consistency — Multi-step mutations are transactional; partial failures leave the system in a valid state
• [ ] Boundary Validation — Values at the boundaries of valid ranges (min, max, exactly-at-limit) are tested

Maintainability Checklist

• [ ] Naming Clarity — Variables, functions, and classes have descriptive names that reveal intent
• [ ] Single Responsibility — Each function/class/module does one thing; changes to one concern don't ripple through unrelated code
• [ ] DRY — Duplicated logic is extracted into shared utilities; copy-pasted blocks are consolidated
• [ ] Cyclomatic Complexity — Functions have low branching complexity; deeply nested chains are refactored
• [ ] Error Handling — Errors are caught at appropriate boundaries, logged with context, and surfaced meaningfully
• [ ] Dead Code Removal — Commented-out code, unused imports, unreachable branches, and obsolete feature flags are removed
• [ ] Magic Numbers & Strings — Literal values are extracted into named constants with clear semantics
• [ ] Consistent Patterns — New code follows the conventions already established in the codebase
• [ ] Function Length — Functions are short enough to understand at a glance; long functions are decomposed
• [ ] Dependency Direction — Dependencies point inward (infrastructure to domain); core logic does not import from UI or framework layers

Testing Checklist

• [ ] Test Coverage — New logic paths have corresponding tests; critical paths have both happy-path and failure-case tests
• [ ] Edge Case Tests — Tests cover boundary values, empty inputs, nulls, and error conditions
• [ ] No Flaky Tests — Tests are deterministic; no reliance on timing, external services, or shared mutable state
• [ ] Test Independence — Each test sets up its own state and tears it down; test order does not affect results
• [ ] Meaningful Assertions — Tests assert on behavior and outcomes, not implementation details
• [ ] Test Readability — Tests follow Arrange-Act-Assert; test names describe the scenario and expected outcome
• [ ] Mocking Discipline — Only external boundaries (network, DB, filesystem) are mocked
• [ ] Regression Tests — Bug fixes include a test that reproduces the original bug and proves it is resolved

Review Process

Work through the code in three passes. Do not try to catch everything in one read.

First Pass (2-5 minutes)

1. Read the PR description and linked issue
2. Scan the file list — does the change scope make sense?
3. Check the overall approach — is this the right solution to the problem?
4. Verify the change does not introduce architectural drift

Second Pass (bulk of review time)

1. Read each file diff top to bottom
2. Check every function change against the checklists above
3. Verify error handling at every I/O boundary
4. Flag anything that makes you pause — trust your instincts

Third Pass (5 minutes)

1. Think about what could go wrong in production
2. Check for missing tests on the code paths you flagged
3. Verify rollback safety — can this change be reverted without data loss?
4. Confirm documentation and changelog are updated if needed

Severity Levels

Classify every comment by severity so the author knows what blocks merge.

Always prefix your review comment with the severity label. This removes ambiguity about what matters.

Giving Feedback

Principles

• Be specific — Point to the exact line and explain the issue, not just "this is wrong"
• Explain why — State the risk or consequence, not just the rule
• Suggest a fix — Offer a concrete alternative or code snippet when possible
• Ask, don't demand — Use questions for subjective points: "What do you think about...?"
• Acknowledge good work — Call out clean solutions, clever optimizations, or thorough tests
• Separate blocking from non-blocking — Use severity labels so the author knows what matters

Example Comments

Bad:

This is wrong. Fix it.

Good:

[MAJOR] This query interpolates user input directly into the SQL string (line 42), which is vulnerable to SQL injection. Consider using a parameterized query: ``sql SELECT * FROM users WHERE id = $1 ``

Bad:

Why didn't you add tests?

Good:

[MINOR] The new calculateDiscount() function has a few branching paths — could we add tests for the zero-quantity and negative-price edge cases to prevent regressions?

Bad:

I would have done this differently.

Good:

[NIT] This works well. An alternative approach could be extracting the retry logic into a shared withRetry() wrapper — but that's optional and could be a follow-up.

Review Anti-Patterns

Avoid these common traps that waste time and damage team trust:

NEVER Do

1. NEVER approve without reading every changed line — rubber-stamping is worse than no review
2. NEVER block a PR solely for style preferences — use a linter; humans review logic
3. NEVER leave feedback without a severity level — ambiguity causes wasted cycles
4. NEVER request changes without explaining why — "fix this" teaches nothing
5. NEVER review more than 400 lines in one sitting — comprehension drops sharply; break large PRs into sessions
6. NEVER skip the security checklist — one missed vulnerability outweighs a hundred style nits
7. NEVER make it personal — review the code, never the coder; assume good intent