GitHub Code Review

Coordinates parallel review agents (security, performance, architecture, style, accessibility) against a GitHub PR. Each agent focuses on one dimension; the swarm aggregates findings, posts inline comments, and enforces quality gates.

Contents

• When to Use
• Prerequisites
• Quick Start
• Workflow
• Verification
• References

When to Use

• "Run a code review on PR #123"
• "Do a security review on this PR"
• "Set up automated review on every PR"
• "Generate inline review comments for the diff"
• "Add quality gates to branch protection"

Prerequisites

• gh CLI installed and authenticated (gh auth status)
• ruv-swarm and/or claude-flow MCP server configured
• Repository access permissions for the target PR
• For automated review: GITHUB_TOKEN available in CI

Quick Start

PR=123
PR_DATA=$(gh pr view "$PR" --json files,additions,deletions,title,body)
PR_DIFF=$(gh pr diff "$PR")

npx ruv-swarm github review-init \
  --pr "$PR" \
  --pr-data "$PR_DATA" \
  --diff "$PR_DIFF" \
  --agents "security,performance,style,architecture,accessibility" \
  --depth comprehensive

gh pr comment "$PR" --body "Multi-agent code review initiated"

Workflow

1. Fetch PR context with gh pr view --json and gh pr diff
2. Pick topology based on PR size (ring < 100 LOC, mesh 100–500, hierarchical > 500) — see references/swarm-management.md
3. Dispatch agents in parallel — one per review dimension; see references/review-agents.md
4. Aggregate findings, post inline comments via the GitHub API; see references/comments-and-gates.md
5. Apply quality gates — block merge on critical security findings, warn on performance regressions, suggest on style
6. Verify (see below)

Verification

After a review run, confirm:

• Each requested agent posted at least one finding (or an explicit "no issues found")
• Inline comments resolve to actual lines in the diff (no stale commit_id)
• PR status reflects the outcome — --approve, --request-changes, or labelled
• Critical security findings blocked the merge (test by injecting a known-bad pattern, e.g. an obvious eval)
• Rate-limit budget after the run — gh api rate_limit — is not exhausted

If a gate misfires, see references/troubleshooting.md.

References

• references/review-agents.md — security / performance / architecture / style agent commands and templates
• references/swarm-management.md — swarm creation, label routing, topology, comment commands, webhook handler
• references/configuration.md — review-swarm.yml, custom triggers, GitHub Actions, PR description template, auto-merge
• references/comments-and-gates.md — inline comment generation, batch management, quality gates, metrics
• references/advanced-features.md — context-aware review, learning, cross-PR analysis, custom agents
• references/workflow-examples.md — five worked examples + Claude Code integration pattern
• references/troubleshooting.md — agents not spawning, comments not posting, slow reviews, performance tips, webhook security caveats, security checklist

Related Skills

• github-pr-manager — PR lifecycle management (open, label, merge)
• github-workflow-automation — broader GitHub Actions automation
• swarm-coordination — orchestration patterns shared across swarm-based skills

External Documentation

• GitHub CLI manual
• ruv-swarm
• claude-flow

Maintained as part of the Claude Code Flow project. MIT licensed. Version 1.1.0.