Code Review Playbook

Overview

This skill provides a comprehensive framework for effective code reviews that improve code quality, share knowledge, and foster collaboration. Whether you're a reviewer giving feedback or an author preparing code for review, this playbook ensures reviews are thorough, consistent, and constructive.

When to use this skill:

• Reviewing pull requests or merge requests
• Preparing code for review (self-review)
• Establishing code review standards for teams
• Training new developers on review best practices
• Resolving disagreements about code quality
• Improving review processes and efficiency

Code Review Philosophy

Purpose of Code Reviews

Code reviews serve multiple purposes:

1. Quality Assurance: Catch bugs, logic errors, and edge cases
2. Knowledge Sharing: Spread domain knowledge across the team
3. Consistency: Ensure codebase follows conventions and patterns
4. Mentorship: Help developers improve their skills
5. Collective Ownership: Build shared responsibility for code
6. Documentation: Create discussion history for future reference

Principles

Be Kind and Respectful:

• Review the code, not the person
• Assume positive intent
• Praise good solutions
• Frame feedback constructively

Be Specific and Actionable:

• Point to specific lines of code
• Explain why something should change
• Suggest concrete improvements
• Provide examples when helpful

Balance Speed with Thoroughness:

• Aim for timely feedback (< 24 hours)
• Don't rush critical reviews
• Use automation for routine checks
• Focus human review on logic and design

Distinguish Must-Fix from Nice-to-Have:

• Use conventional comments to indicate severity
• Block merges only for critical issues
• Allow authors to defer minor improvements
• Capture deferred work in follow-up tickets

---

Conventional Comments

A standardized format for review comments that makes intent clear.

Format

<label> [decorations]: <subject>

[discussion]

Labels

Label	Meaning	Blocks Merge?
praise	Highlight something positive	No
nitpick	Minor, optional suggestion	No
suggestion	Propose an improvement	No
issue	Problem that should be addressed	Usually
question	Request clarification	No
thought	Idea to consider	No
chore	Routine task (formatting, deps)	No
note	Informational comment	No
todo	Follow-up work needed	Maybe
security	Security concern	Yes
bug	Potential bug	Yes
breaking	Breaking change	Yes

Decorations

Optional modifiers in square brackets:

Decoration	Meaning
[blocking]	Must be addressed before merge
[non-blocking]	Optional, can be deferred
[if-minor]	Only if it's a quick fix

Examples

// ✅ Good: Clear, specific, actionable

praise: Excellent use of TypeScript generics here!

This makes the function much more reusable while maintaining type safety.

---

nitpick [non-blocking]: Consider using const instead of let

This variable is never reassigned, so `const` would be more appropriate:
```typescript
const MAX_RETRIES = 3;

---

issue: Missing error handling for API call

If the API returns a 500 error, this will crash the application. Add a try/catch block:

try {
  const data = await fetchUser(userId);
  // ...
} catch (error) {
  logger.error('Failed to fetch user', { userId, error });
  throw new UserNotFoundError(userId);
}

---

question: Why use a Map instead of an object here?

Is there a specific reason for this data structure choice? If it's for performance, could you add a comment explaining?

---

security [blocking]: API endpoint is not authenticated

The /api/admin/users endpoint is missing authentication middleware. This allows unauthenticated access to sensitive user data.

Add the auth middleware:

router.get('/api/admin/users', requireAdmin, getUsers);

---

suggestion [if-minor]: Extract magic number to named constant

Consider extracting this value:

const CACHE_TTL_SECONDS = 3600;
cache.set(key, value, CACHE_TTL_SECONDS);

---

## Review Process

### 1. Before Reviewing

**Check Context:**
- Read the PR/MR description
- Understand the purpose and scope
- Review linked tickets or issues
- Check CI/CD pipeline status

**Verify Automated Checks:**
- [ ] Tests are passing
- [ ] Linting has no errors
- [ ] Type checking passes
- [ ] Code coverage meets targets
- [ ] No merge conflicts

**Set Aside Time:**
- Small PR (< 200 lines): 15-30 minutes
- Medium PR (200-500 lines): 30-60 minutes
- Large PR (> 500 lines): 1-2 hours (or ask to split)

### 2. During Review

**Follow a Pattern:**

1. **High-Level Review** (5-10 minutes)
   - Read PR description and understand intent
   - Skim all changed files to get overview
   - Verify approach makes sense architecturally
   - Check that changes align with stated purpose

2. **Detailed Review** (20-45 minutes)
   - Line-by-line code review
   - Check logic, edge cases, error handling
   - Verify tests cover new code
   - Look for security vulnerabilities
   - Ensure code follows team conventions

3. **Testing Considerations** (5-10 minutes)
   - Are tests comprehensive?
   - Do tests test the right things?
   - Are edge cases covered?
   - Is test data realistic?

4. **Documentation Check** (5 minutes)
   - Are complex sections commented?
   - Is public API documented?
   - Are breaking changes noted?
   - Is README updated if needed?

### 3. After Reviewing

**Provide Clear Decision:**
- ✅ **Approve**: Code is ready to merge
- 💬 **Comment**: Feedback provided, no action required
- 🔄 **Request Changes**: Issues must be addressed before merge

**Respond to Author:**
- Answer questions promptly
- Re-review after changes made
- Approve when issues resolved
- Thank author for addressing feedback

---

## Review Checklists

### General Code Quality

- [ ] **Readability**: Code is easy to understand
- [ ] **Naming**: Variables and functions have clear, descriptive names
- [ ] **Comments**: Complex logic is explained
- [ ] **Formatting**: Code follows team style guide
- [ ] **DRY**: No unnecessary duplication
- [ ] **SOLID Principles**: Code follows SOLID where applicable
- [ ] **Function Size**: Functions are focused and < 50 lines
- [ ] **Cyclomatic Complexity**: Functions have complexity < 10

### Functionality

- [ ] **Correctness**: Code does what it's supposed to do
- [ ] **Edge Cases**: Boundary conditions handled (null, empty, min/max)
- [ ] **Error Handling**: Errors caught and handled appropriately
- [ ] **Logging**: Appropriate log levels and messages
- [ ] **Input Validation**: User input is validated and sanitized
- [ ] **Output Validation**: Responses match expected schema

### Testing

- [ ] **Test Coverage**: New code has tests
- [ ] **Test Quality**: Tests actually test the right things
- [ ] **Edge Cases Tested**: Tests cover boundary conditions
- [ ] **Error Paths Tested**: Error handling is tested
- [ ] **Test Isolation**: Tests don't depend on each other
- [ ] **Test Naming**: Test names describe what's being tested

### Performance

- [ ] **Database Queries**: N+1 queries avoided
- [ ] **Caching**: Appropriate caching used
- [ ] **Algorithm Efficiency**: No unnecessarily slow algorithms (O(n²) when O(n) possible)
- [ ] **Resource Cleanup**: Files, connections, memory released
- [ ] **Lazy Loading**: Heavy operations deferred when possible

### Security

- [ ] **Authentication**: Protected endpoints require auth
- [ ] **Authorization**: Users can only access their own data
- [ ] **Input Sanitization**: SQL injection, XSS prevented
- [ ] **Secrets Management**: No hardcoded credentials or API keys
- [ ] **Encryption**: Sensitive data encrypted at rest and in transit
- [ ] **HTTPS Only**: Production traffic uses HTTPS
- [ ] **Rate Limiting**: Endpoints protected from abuse

### Language-Specific (JavaScript/TypeScript)

- [ ] **Async/Await**: Promises handled correctly
- [ ] **Type Safety**: TypeScript types are specific (not `any`)
- [ ] **Nullability**: Null checks where needed (`?.` operator)
- [ ] **Array Methods**: Using map/filter/reduce appropriately
- [ ] **Const vs Let**: Using const for immutable values
- [ ] **Arrow Functions**: Appropriate use of arrow vs regular functions

### Language-Specific (Python)

- [ ] **PEP 8**: Follows Python style guide
- [ ] **Type Hints**: Functions have type annotations
- [ ] **List Comprehensions**: Used where appropriate (not overused)
- [ ] **Context Managers**: Using `with` for file/connection handling
- [ ] **Exception Handling**: Specific exceptions caught (not bare `except:`)
- [ ] **F-Strings**: Modern string formatting used

### API Design

- [ ] **RESTful**: Follows REST conventions (if REST API)
- [ ] **Consistent Naming**: Endpoints follow naming patterns
- [ ] **HTTP Methods**: Correct methods used (GET, POST, PUT, DELETE)
- [ ] **Status Codes**: Appropriate HTTP status codes returned
- [ ] **Error Responses**: Consistent error format
- [ ] **Pagination**: Large lists are paginated
- [ ] **Versioning**: API version strategy followed

### Database

- [ ] **Migrations**: Database changes have migrations
- [ ] **Indexes**: Appropriate indexes created
- [ ] **Transactions**: ACID properties maintained
- [ ] **Cascades**: Delete cascades handled correctly
- [ ] **Constraints**: Foreign keys, unique constraints defined
- [ ] **N+1 Queries**: Eager loading used where needed

### Documentation

- [ ] **PR Description**: Clear explanation of changes
- [ ] **Code Comments**: Complex logic explained
- [ ] **API Docs**: Public API documented (JSDoc, docstrings)
- [ ] **README**: Updated if functionality changed
- [ ] **CHANGELOG**: Breaking changes documented

---

## Review Feedback Patterns

### How to Give Constructive Feedback

#### ✅ Good Feedback

issue: This function doesn't handle the case where the user is null

If getUserById() returns null (user not found), this will throw: Cannot read property 'email' of null

Add a null check:

const user = await getUserById(userId);
if (!user) {
  throw new UserNotFoundError(userId);
}
return user.email;

**Why it's good:**
- Specific (points to exact problem)
- Explains the impact (what will happen)
- Suggests a concrete solution
- Provides example code

#### ❌ Bad Feedback

This is wrong. Fix it.

**Why it's bad:**
- Not specific (what's wrong?)
- Not helpful (how to fix?)
- Sounds harsh (not constructive)

---

### How to Receive Feedback

**As an Author:**

1. **Assume Good Intent**: Reviewers are trying to help
2. **Ask Questions**: If feedback is unclear, ask for clarification
3. **Acknowledge Valid Points**: Accept feedback graciously
4. **Explain Your Reasoning**: If you disagree, explain why
5. **Make Changes Promptly**: Address feedback quickly
6. **Say Thank You**: Appreciate the reviewer's time

**Responding to Feedback:**

```markdown
✅ Good Response:
> Good catch! I didn't consider the null case.
> I've added the null check in commit abc123.

✅ Good Response (Disagreement):
> I chose a Map here because we need O(1) lookups on a large dataset.
> An object would work but performs worse at scale (n > 10,000).
> I added a comment explaining this tradeoff.

❌ Bad Response:
> This works fine. Not changing it.

---

Code Review Anti-Patterns

For Reviewers

❌ Nitpicking Without Clear Value

nitpick: Add a space here
nitpick: Rename this variable to `userInfo` instead of `userData`
nitpick: Move this import to the top

Better: Use automated tools (Prettier, linters) for formatting.

❌ Reviewing Line-by-Line Without Understanding Context Better: Read PR description first, understand the big picture.

❌ Blocking on Personal Preferences

This should be a class instead of functions.

Better: Only block on objective issues (bugs, security). Discuss preferences separately.

❌ Rewriting Code in Comments Better: Suggest improvements, don't provide full implementations (unless very helpful).

❌ Review Fatigue (Approving Without Reading) Better: If you don't have time, say so. Don't rubber-stamp.

For Authors

❌ Giant Pull Requests (> 1000 lines) Better: Break into smaller, focused PRs.

❌ No Description Better: Write clear PR description with context and testing notes.

❌ Submitting Failing Tests Better: Ensure all automated checks pass before requesting review.

❌ Getting Defensive About Feedback Better: Accept feedback gracefully, discuss respectfully.

❌ Force-Pushing After Review Better: Add new commits so reviewers can see what changed.

---

Review Templates

Standard PR Template

## Description

Brief summary of what changed and why.

Fixes #[issue number]

## Type of Change

- [ ] Bug fix (non-breaking change which fixes an issue)
- [ ] New feature (non-breaking change which adds functionality)
- [ ] Breaking change (fix or feature that would cause existing functionality to not work as expected)
- [ ] Refactoring (no functional changes)
- [ ] Documentation update

## How Has This Been Tested?

- [ ] Unit tests added/updated
- [ ] Integration tests added/updated
- [ ] Tested manually in local environment
- [ ] Tested in staging environment

## Checklist

- [ ] My code follows the style guidelines of this project
- [ ] I have performed a self-review of my own code
- [ ] I have commented my code, particularly in hard-to-understand areas
- [ ] I have made corresponding changes to the documentation
- [ ] My changes generate no new warnings
- [ ] I have added tests that prove my fix is effective or that my feature works
- [ ] New and existing unit tests pass locally with my changes
- [ ] Any dependent changes have been merged and published

## Screenshots (if applicable)

[Add screenshots for UI changes]

## Additional Notes

[Any additional context or notes for reviewers]

Security Review Template

## Security Review Checklist

### Authentication & Authorization
- [ ] Endpoints require appropriate authentication
- [ ] User permissions are checked before actions
- [ ] JWT tokens are validated and not expired
- [ ] Session management is secure

### Input Validation
- [ ] All user inputs are validated
- [ ] SQL injection prevented (parameterized queries)
- [ ] XSS prevented (input sanitization, CSP headers)
- [ ] CSRF protection in place for state-changing operations

### Data Protection
- [ ] Sensitive data is encrypted at rest
- [ ] TLS/HTTPS used for data in transit
- [ ] API keys and secrets not hardcoded
- [ ] PII is properly handled (GDPR/CCPA compliance)

### Dependencies
- [ ] No known vulnerabilities in dependencies (npm audit / pip-audit)
- [ ] Dependencies from trusted sources
- [ ] Dependency versions pinned

### Logging & Monitoring
- [ ] Sensitive data not logged (passwords, tokens)
- [ ] Security events logged (failed auth attempts)
- [ ] Rate limiting in place for public endpoints

**Security Concerns Identified:**
[List any security issues found]

**Reviewer Signature:**
[Name, Date]

---

Review Metrics and Goals

Healthy Review Metrics

Metric	Target	Purpose
Review Time	< 24 hours	Fast feedback loop
PR Size	< 400 lines	Manageable reviews
Approval Rate (first review)	20-30%	Balance speed vs quality
Comments per PR	3-10	Engaged but not overwhelming
Back-and-Forth Rounds	1-2	Efficient communication
Time to Merge (after approval)	< 2 hours	Avoid stale branches

Warning Signs

• ⚠️ PRs sitting unreviewed for > 3 days (review capacity issue)
• ⚠️ 90%+ approval rate on first review (rubber-stamping)
• ⚠️ Average PR size > 800 lines (PRs too large)
• ⚠️ 15+ comments per PR (overly nitpicky or unclear requirements)
• ⚠️ 5+ review rounds (poor communication or unclear standards)

---

Integration with Agents

Code Quality Reviewer Agent

• Uses this playbook when reviewing code
• Applies conventional comments format
• Follows language-specific checklists
• Provides structured, actionable feedback

Backend System Architect

• Reviews for architectural soundness
• Checks design patterns and scalability
• Validates API design against best practices

Frontend UI Developer

• Reviews component structure and patterns
• Checks accessibility and responsive design
• Validates UI/UX implementation

Security Reviewer (Future Agent)

• Focuses on security checklist
• Identifies vulnerabilities
• Validates compliance requirements

---

Advanced Pattern Detection (Opus 4.5)

Extended Thinking for Complex Reviews

For large PRs or systemic analysis, leverage Opus 4.5's extended thinking to perform deep pattern detection:

When to Use Extended Thinking:

• PR touches > 10 files across multiple modules
• Reviewing security-critical code paths
• Analyzing architectural consistency
• Detecting cross-file code smells
• Evaluating breaking change impact

Extended Thinking Review Pattern:

Based on Anthropic's Extended Thinking API:

import Anthropic from '@anthropic-ai/sdk';

interface ReviewResult {
  issues: ReviewIssue[];
  patterns: DetectedPattern[];
  recommendations: string[];
  riskScore: number;
}

async function performDeepCodeReview(
  prDiff: string,
  codebaseContext: string
): Promise<ReviewResult> {
  const anthropic = new Anthropic();

  // Extended thinking requires budget_tokens < max_tokens
  // Minimum budget: 1,024 tokens
  const response = await anthropic.messages.create({
    model: 'claude-opus-4-5-20251101',
    max_tokens: 16000,
    thinking: {
      type: 'enabled',
      budget_tokens: 8000 // Deep analysis for complex reviews
    },
    messages: [{
      role: 'user',
      content: `
        Perform a comprehensive code review analyzing:

        ## PR Changes
        ${prDiff}

        ## Codebase Context
        ${codebaseContext}

        ## Analysis Requirements
        1. **Systemic Pattern Detection**: Identify recurring patterns (good and bad)
        2. **Cross-File Impact**: Trace how changes affect other modules
        3. **Security Analysis**: Deep scan for vulnerabilities (OWASP Top 10)
        4. **Architectural Consistency**: Verify alignment with existing patterns
        5. **Technical Debt Assessment**: Identify debt being added or resolved

        Provide structured output with:
        - Critical issues (must fix)
        - Warnings (should fix)
        - Suggestions (nice to have)
        - Detected patterns (positive and negative)
        - Risk score (1-10)
      `
    }]
  });

  // Response contains thinking blocks followed by text blocks
  // content: [{ type: 'thinking', thinking: '...' }, { type: 'text', text: '...' }]
  return parseReviewResponse(response);
}

Systemic Code Smell Detection

Detect patterns that span multiple files:

interface CodeSmellPattern {
  type: 'duplication' | 'coupling' | 'complexity' | 'naming' | 'architecture';
  severity: 'critical' | 'warning' | 'info';
  locations: FileLocation[];
  description: string;
  suggestion: string;
}

const SMELL_PATTERNS = {
  // Cross-file duplication
  duplication: {
    triggers: ['similar logic in 3+ files', 'copy-paste patterns', 'repeated error handling'],
    action: 'Extract to shared utility or base class'
  },

  // Tight coupling
  coupling: {
    triggers: ['circular imports', 'god objects', 'feature envy'],
    action: 'Apply dependency injection or interface segregation'
  },

  // Complexity creep
  complexity: {
    triggers: ['nested callbacks > 3 levels', 'functions > 50 lines', 'cyclomatic complexity > 10'],
    action: 'Decompose into smaller, focused functions'
  },

  // Inconsistent naming
  naming: {
    triggers: ['mixed conventions', 'unclear abbreviations', 'inconsistent pluralization'],
    action: 'Align with codebase naming conventions'
  },

  // Architectural drift
  architecture: {
    triggers: ['layer violations', 'missing abstractions', 'inappropriate intimacy'],
    action: 'Refactor to follow established architectural patterns'
  }
};

Security Deep Analysis

Extended thinking enables comprehensive security review:

interface SecurityFinding {
  category: 'injection' | 'auth' | 'crypto' | 'exposure' | 'config';
  severity: 'critical' | 'high' | 'medium' | 'low';
  cwe: string; // CWE ID
  location: FileLocation;
  description: string;
  remediation: string;
}

async function performSecurityReview(
  code: string,
  context: SecurityContext
): Promise<SecurityFinding[]> {
  const anthropic = new Anthropic();

  const response = await anthropic.messages.create({
    model: 'claude-opus-4-5-20251101',
    max_tokens: 12000,
    thinking: {
      type: 'enabled',
      budget_tokens: 6000 // Security analysis needs deep reasoning
    },
    messages: [{
      role: 'user',
      content: `
        Perform security analysis on this code:

        ${code}

        Context:
        - Language: ${context.language}
        - Framework: ${context.framework}
        - Exposure: ${context.isPublicFacing ? 'Public' : 'Internal'}

        Check for:
        1. Injection vulnerabilities (SQL, XSS, command)
        2. Authentication/Authorization flaws
        3. Cryptographic weaknesses
        4. Sensitive data exposure
        5. Security misconfiguration

        For each finding, provide CWE ID and specific remediation.
      `
    }]
  });

  return parseSecurityFindings(response);
}

Cross-File Impact Analysis

Trace how changes ripple through the codebase:

interface ImpactAnalysis {
  directImpact: FileImpact[];
  indirectImpact: FileImpact[];
  breakingChanges: BreakingChange[];
  testCoverage: TestCoverageGap[];
}

async function analyzeChangeImpact(
  changedFiles: string[],
  dependencyGraph: DependencyGraph
): Promise<ImpactAnalysis> {
  // Build impact graph
  const directlyAffected = new Set<string>();
  const indirectlyAffected = new Set<string>();

  for (const file of changedFiles) {
    // Files that import this file
    const dependents = dependencyGraph.getDependents(file);
    dependents.forEach(d => directlyAffected.add(d));

    // Second-level dependents
    for (const dependent of dependents) {
      const secondLevel = dependencyGraph.getDependents(dependent);
      secondLevel.forEach(d => {
        if (!directlyAffected.has(d)) {
          indirectlyAffected.add(d);
        }
      });
    }
  }

  // Use extended thinking for breaking change analysis
  const breakingAnalysis = await analyzeBreakingChanges(
    changedFiles,
    Array.from(directlyAffected)
  );

  return {
    directImpact: Array.from(directlyAffected).map(f => ({ file: f, type: 'direct' })),
    indirectImpact: Array.from(indirectlyAffected).map(f => ({ file: f, type: 'indirect' })),
    breakingChanges: breakingAnalysis.breaking,
    testCoverage: breakingAnalysis.gaps
  };
}

Review Automation with Model Tiering

Optimize review costs with intelligent model selection:

type ModelTier = 'opus' | 'sonnet' | 'haiku';

interface ReviewConfig {
  model: ModelTier;
  thinkingBudget?: number;
  focus: string[];
}

function selectReviewModel(pr: PullRequest): ReviewConfig {
  // Critical path - use Opus with extended thinking
  if (pr.touchesCriticalPath || pr.isSecurityRelated) {
    return {
      model: 'opus',
      thinkingBudget: 8000,
      focus: ['security', 'architecture', 'breaking-changes']
    };
  }

  // Large PRs - use Opus for systemic analysis
  if (pr.filesChanged > 10 || pr.linesChanged > 500) {
    return {
      model: 'opus',
      thinkingBudget: 5000,
      focus: ['patterns', 'impact', 'consistency']
    };
  }

  // Standard PRs - Sonnet for thorough review
  if (pr.linesChanged > 100) {
    return {
      model: 'sonnet',
      focus: ['correctness', 'style', 'tests']
    };
  }

  // Small changes - Haiku for quick review
  return {
    model: 'haiku',
    focus: ['correctness', 'style']
  };
}

Review Comment Generation

Generate conventional comments automatically:

interface GeneratedComment {
  label: 'praise' | 'nitpick' | 'suggestion' | 'issue' | 'question' | 'security' | 'bug';
  decoration?: 'blocking' | 'non-blocking' | 'if-minor';
  subject: string;
  discussion: string;
  location: { file: string; line: number };
}

function formatConventionalComment(comment: GeneratedComment): string {
  const decoration = comment.decoration ? ` [${comment.decoration}]` : '';
  return `${comment.label}${decoration}: ${comment.subject}\n\n${comment.discussion}`;
}

// Example output:
// security [blocking]: SQL injection vulnerability in user search
//
// The search query is constructed using string concatenation:
// ```typescript
// const query = `SELECT * FROM users WHERE name = '${searchTerm}'`;
// ```
//
// Use parameterized queries instead:
// ```typescript
// const query = 'SELECT * FROM users WHERE name = $1';
// const result = await db.query(query, [searchTerm]);
// ```

---

Quick Start Guide

For Reviewers:

1. Read PR description and understand intent
2. Check that automated checks pass
3. Do high-level review (architecture, approach)
4. Do detailed review (logic, edge cases, tests)
5. Use conventional comments for clear communication
6. Provide decision: Approve, Comment, or Request Changes

For Authors:

1. Write clear PR description
2. Perform self-review before requesting review
3. Ensure all automated checks pass
4. Keep PR focused and reasonably sized (< 400 lines)
5. Respond to feedback promptly and respectfully
6. Make requested changes or explain reasoning

---

Skill Version: 2.0.0 Last Updated: 2025-11-27 Maintained by: AI Agent Hub Team