Autonomous PR Reviewer

⚡ UNIQUE FEATURE: Multi-agent review system with specialized reviewers for security, performance, testing, and architecture - the first autonomous PR review skill with parallel agent coordination.

What This Skill Does

Automatically reviews pull requests with multiple specialized AI agents working in parallel:

• Security Agent: Scans for vulnerabilities, SQL injection, XSS, hardcoded secrets
• Performance Agent: Identifies bottlenecks, inefficient algorithms, memory leaks
• Testing Agent: Validates test coverage, suggests additional test cases
• Architecture Agent: Reviews design patterns, code structure, maintainability
• Style Agent: Checks code style, naming conventions, documentation

Why This Is Unique

Unlike simple code review tools, this skill:

• Runs 5 specialized agents in parallel for comprehensive analysis
• Provides actionable suggestions with code examples
• Generates review summaries for different audiences (technical/non-technical)
• Auto-suggests fixes that you can apply with one command
• Learns from your codebase patterns and conventions

Instructions

Phase 1: Setup and Discovery

1. Identify the PR:
   • Use Bash to get current branch and diff: git diff main...HEAD
   • Or accept PR number/URL from user
   • Use Bash: gh pr view <number> to get PR details
2. Gather Context:
   • Use Glob to find all changed files
   • Use Read to examine modified code
   • Use Grep to search for related code patterns
   • Identify programming languages and frameworks

Phase 2: Multi-Agent Review (Parallel Execution)

Launch 5 specialized Task agents in parallel:

Agent 1: Security Reviewer

Task: Security analysis of PR
Prompt: "Analyze these code changes for security vulnerabilities:
- SQL injection risks
- XSS vulnerabilities
- Hardcoded secrets or API keys
- Authentication/authorization issues
- Dependency vulnerabilities
- OWASP Top 10 issues

Files: [list changed files]

Provide:
1. Severity ratings (Critical/High/Medium/Low)
2. Specific line numbers
3. Exploitation scenarios
4. Remediation steps with code examples"

Agent 2: Performance Reviewer

Task: Performance analysis of PR
Prompt: "Analyze these code changes for performance issues:
- Inefficient algorithms (O(n²) vs O(n log n))
- Database N+1 queries
- Memory leaks
- Unnecessary re-renders (React/Vue)
- Blocking operations
- Resource waste

Files: [list changed files]

Provide:
1. Performance impact assessment
2. Specific bottlenecks with line numbers
3. Benchmark comparison suggestions
4. Optimized code examples"

Agent 3: Testing Reviewer

Task: Test coverage analysis of PR
Prompt: "Analyze test coverage and quality:
- Calculate test coverage for changed code
- Identify untested edge cases
- Review test quality and assertions
- Suggest additional test scenarios
- Check for test best practices

Files: [list changed files]

Provide:
1. Coverage percentage
2. Missing test cases
3. Test improvement suggestions
4. Example test code"

Agent 4: Architecture Reviewer

Task: Architecture and design analysis
Prompt: "Review architectural decisions:
- Design pattern appropriateness
- SOLID principles adherence
- Code modularity and coupling
- Separation of concerns
- Scalability considerations
- Technical debt introduced

Files: [list changed files]

Provide:
1. Architecture assessment
2. Design improvement suggestions
3. Refactoring recommendations
4. Long-term impact analysis"

Agent 5: Style & Documentation Reviewer

Task: Code style and documentation review
Prompt: "Review code style and documentation:
- Naming conventions
- Code readability
- Comment quality
- API documentation
- README updates needed
- Breaking changes documented

Files: [list changed files]

Provide:
1. Style issues with line numbers
2. Documentation gaps
3. Readability improvements
4. Suggested comments"

Phase 3: Synthesis and Reporting

1. Collect all agent results (wait for all Task agents to complete)
2. Generate comprehensive review:

   # PR Review Summary
   
   ## 📊 Overview
   - Files changed: X
   - Lines added: Y
   - Lines removed: Z
   - Overall rating: [Excellent/Good/Needs Work/Reject]
   
   ## 🔒 Security (Critical: X, High: Y, Medium: Z)
   [Agent 1 findings summary]
   
   ## ⚡ Performance (Issues: X)
   [Agent 2 findings summary]
   
   ## ✅ Testing (Coverage: X%)
   [Agent 3 findings summary]
   
   ## 🏗️ Architecture
   [Agent 4 findings summary]
   
   ## 📝 Style & Documentation
   [Agent 5 findings summary]
   
   ## 🎯 Action Items
   1. [Priority action with fix]
   2. [Priority action with fix]
   
   ## 💡 Suggested Changes
   [Code blocks with suggested improvements]
   
   ## ✨ Highlights
   [Positive aspects of the PR]

3. Generate fix suggestions:
   • Create a pr-review-fixes.md file with all suggested changes
   • Optionally create a pr-review-fixes.patch file

Phase 4: Interactive Options

Offer the user:

1. Post review as comment: Use Bash gh pr comment <number> -F pr-review.md
2. Apply suggested fixes: Use Edit to apply recommended changes
3. Re-run specific agent: Re-analyze with one agent for updated code
4. Generate test cases: Create tests based on Testing Agent suggestions
5. Export report: Save review in multiple formats (markdown, JSON, HTML)

Examples

Example 1: GitHub PR Review

User Request: "Review PR #123"

Workflow:

1. Fetch PR: gh pr view 123
2. Launch 5 agents in parallel (use Task tool 5 times in one message)
3. Wait for all agents to complete
4. Synthesize results
5. Present comprehensive review
6. Offer to post comment or apply fixes

Output:

🔍 PR #123 Review Complete

📊 Overall: Good (minor improvements needed)

🔒 Security: ✅ No issues found
⚡ Performance: ⚠️ 1 issue found
✅ Testing: ⚠️ Coverage 78% (target: 80%)
🏗️ Architecture: ✅ Well designed
📝 Style: ⚠️ 3 minor issues

📋 Action Items:
1. Add database index for user_id column (performance)
2. Add tests for error scenarios (testing)
3. Update function documentation (style)

Would you like me to:
1. Post this review as a PR comment
2. Apply the suggested fixes
3. Generate the missing tests

Example 2: Local Branch Review

User Request: "Review my current changes before I push"

Workflow:

1. Run: git diff main...HEAD
2. Analyze changes with 5 agents
3. Provide feedback before push
4. Optionally fix issues

Configuration

Customize review behavior:

# .pr-reviewer-config.yml
agents:
  security:
    enabled: true
    severity_threshold: medium
  performance:
    enabled: true
    benchmark_required: false
  testing:
    enabled: true
    min_coverage: 80
  architecture:
    enabled: true
    check_solid: true
  style:
    enabled: true
    follow_existing: true

review:
  auto_post_comment: false
  suggest_fixes: true
  blocking_issues: [critical_security, zero_tests]

Tool Requirements

• Read: Examine code changes
• Bash: Git operations, gh CLI for PR interaction
• Grep: Search codebase for patterns
• Glob: Find related files
• Task: Launch specialized review agents (KEY FEATURE)
• Write: Create review reports and fix files

Limitations

• Requires gh CLI installed for PR operations
• Best results with code <10,000 lines changed per PR
• Security agent cannot detect all vulnerabilities (not a replacement for dedicated security tools)
• Performance suggestions may need benchmarking to validate
• Works best with supported languages (Python, JavaScript, TypeScript, Go, Rust, Java)

Advanced Features

1. Incremental Review Mode

Review only new commits since last review:

git diff PR_BASE...HEAD --since="last review"

2. Custom Agent Addition

Add your own specialized agents:

• Accessibility reviewer (WCAG compliance)
• Localization reviewer (i18n/l10n)
• API contract reviewer (OpenAPI schema changes)

3. Team Learning Mode

Learns from approved/rejected reviews to adapt to team preferences.

4. Integration Ready

Can integrate with:

• GitHub Actions (automated PR reviews)
• GitLab CI
• Bitbucket Pipelines
• Custom webhooks

Best Practices

1. Run before pushing: Catch issues early
2. Review incrementally: Don't wait for huge PRs
3. Act on Critical/High issues: Always fix security and performance criticals
4. Use as learning tool: Understand why changes are suggested
5. Combine with human review: AI augments, doesn't replace human judgment

Related Skills

• unit-test-generator - Generate tests from Testing Agent suggestions
• refactor-master - Apply architecture improvements
• ci-cd-wizard - Integrate PR reviews into CI/CD

Changelog

Version 1.0.0 (2025-01-13)

• Initial release with multi-agent review system
• 5 specialized agents: Security, Performance, Testing, Architecture, Style
• GitHub PR integration
• Interactive fix application
• Custom configuration support

Contributing

This is a flagship skill for GLINCKER Marketplace. Contributions welcome:

• Add new specialized agents
• Improve detection algorithms
• Add support for new languages
• Enhance reporting formats

License

Apache License 2.0 - See LICENSE

Author

GLINCKER Team

• GitHub: @GLINCKER
• Repository: claude-code-marketplace

🌟 This is a UNIQUE skill not available in other marketplaces!