Pull Request Review Process

This skill performs a comprehensive PR review by spawning multiple specialized sub-agents, each focusing on different aspects of the code. The results are then synthesized into a unified review.

How It Works

•Gather Context: Collect all changes in the PR (files modified, additions, deletions)
•Spawn Review Panel: Launch 3 parallel sub-agents with different review focuses
•Collect Reviews: Gather findings from each reviewer
•Synthesize: Combine into a single comprehensive review

Review Panel

The review panel consists of three specialized reviewers:

Reviewer 1: Correctness Focus

•Bugs and logic errors
•Edge cases not handled
•Error handling gaps
•Race conditions
•Breaking changes

Reviewer 2: Design Focus

•Architecture decisions
•Design patterns usage
•Code organization
•API design
•Scalability concerns
•Maintainability

Reviewer 3: Details Focus

•Code style consistency
•Naming conventions
•Documentation gaps
•Test coverage
•Performance optimizations
•Minor improvements

Execution Steps

Step 1: Gather PR Information

First, collect the changes:

code

1. Get the list of changed files
2. Get the diff for each file
3. Identify the base branch and PR description if available

Use these commands:

•git diff main...HEAD (or appropriate base branch)
•git log main..HEAD --oneline for commits

Step 2: Launch Review Panel

Use the task tool to spawn three parallel sub-agents. Each agent should receive:

•The full diff of changes
•Their specific review focus area
•Instructions to output findings in a structured format

Agent 1 Prompt (Correctness):

code

You are a code reviewer focused on CORRECTNESS. Review this PR for:
- Bugs and logic errors
- Unhandled edge cases  
- Missing error handling
- Race conditions or concurrency issues
- Breaking changes to existing functionality

For each issue found, provide:
- Severity (Critical/High/Medium/Low)
- File and line number
- Description of the issue
- Suggested fix

Here are the changes:
[DIFF]

Agent 2 Prompt (Design):

code

You are a code reviewer focused on DESIGN. Review this PR for:
- Architecture and structural decisions
- Design pattern usage (or missed opportunities)
- Code organization and module boundaries
- API design quality
- Scalability and maintainability concerns

For each concern, provide:
- Impact level (High/Medium/Low)
- Location in code
- Description of the concern
- Recommendation

Here are the changes:
[DIFF]

Agent 3 Prompt (Details):

code

You are a code reviewer focused on DETAILS. Review this PR for:
- Code style consistency
- Naming clarity and conventions
- Missing or outdated documentation
- Test coverage gaps
- Performance optimization opportunities
- Minor improvements and polish

For each suggestion, provide:
- Category (Style/Docs/Tests/Performance/Other)
- Location
- Suggestion

Here are the changes:
[DIFF]

Step 3: Synthesize Reviews

Combine the three reviews into a unified report:

markdown

# PR Review Summary

## Overview
[Brief summary of the PR and overall assessment]

## Verdict
[APPROVE / REQUEST_CHANGES / COMMENT]

## Critical Issues (Must Fix)
[Issues from all reviewers that block merge]

## Recommended Changes
[High-impact improvements suggested by reviewers]

## Suggestions
[Nice-to-have improvements]

## Positive Notes
[What was done well]

---
*Review conducted by panel of 3 specialized reviewers*

Review Criteria for Verdict

APPROVE when:

•No critical issues found
•No high-severity bugs
•Code meets quality standards

REQUEST_CHANGES when:

•Critical bugs found
•Security vulnerabilities present
•Breaking changes without migration path
•Missing required tests for critical paths

COMMENT when:

•Minor issues that don't block merge
•Suggestions for improvement
•Questions needing clarification

Tips for Effective Review

•Be specific: Point to exact lines and explain clearly
•Explain why: Don't just say what's wrong, explain the impact
•Offer solutions: Suggest how to fix issues
•Acknowledge good work: Positive feedback matters
•Prioritize: Focus on what matters most