multi-ai-code-review
Run multi-provider code reviews over git diffs/PR changes using headless AI CLIs (Claude Code `claude`, Google `gemini`, optionally GitHub `copilot`) and output structured findings + an optional synthesized report; use for security/bug/perf review of recent changes and for generating PR-ready review comments.
When & Why to Use This Skill
Multi-AI Code Review is a high-performance automation skill designed to streamline the software development lifecycle by leveraging multiple AI models—including Claude, Gemini, and GitHub Copilot—to analyze git diffs and pull requests. It provides a comprehensive, multi-perspective audit of code changes, identifying security vulnerabilities, performance bottlenecks, and logical bugs while generating structured, PR-ready reports to accelerate the peer review process.
Use Cases
- Automated Pull Request Auditing: Automatically scan and review PR changes for security risks and code quality before a human reviewer even opens the link.
- Cross-Provider Validation: Use multiple AI models (Claude, Gemini, Copilot) simultaneously to cross-reference findings and ensure high-accuracy code analysis.
- CI/CD Pipeline Integration: Incorporate the skill into automated workflows using its CI-friendly mode to enforce code quality gates and generate machine-readable JSON reports.
- Performance and Security Optimization: Specifically target and identify inefficient code patterns or potential security exploits in recent commits using specialized AI-driven insights.
| name | multi-ai-code-review |
|---|---|
| description | Run multi-provider code reviews over git diffs/PR changes using headless AI CLIs (Claude Code `claude`, Google `gemini`, optionally GitHub `copilot`) and output structured findings + an optional synthesized report; use for security/bug/perf review of recent changes and for generating PR-ready review comments. |
| version | 1.0.0 |
| allowed-tools | bash |
Multi-AI Code Review
This skill runs independent code reviews using external headless AI CLIs and produces a structured report you can paste into a PR.
Quick start (git diff)
Unstaged diff:
node codex-skills/multi-ai-code-review/scripts/review.js --providers claude,gemini
Staged diff:
node codex-skills/multi-ai-code-review/scripts/review.js --staged --providers claude,gemini
Diff range (example: typical PR branch):
node codex-skills/multi-ai-code-review/scripts/review.js --range origin/main...HEAD --providers claude,gemini
Output formats
--output json(default): machine-readable results + optional synthesis--output markdown: PR comment format
CI-friendly mode (strict JSON-only)
Use --ci to force stable CI behavior:
- implies
--no-synthesis - forces
--output json - exits non-zero if any provider fails or returns non-parseable JSON
Example:
node codex-skills/multi-ai-code-review/scripts/review.js --ci --range origin/main...HEAD --providers claude,gemini > ai-review.json
Auth behavior (session-first)
Default is --auth-mode session-first:
- Try CLI using your existing logged-in session/subscription (keys temporarily hidden)
- If that fails and env keys exist, retry using env keys
To flip:
node codex-skills/multi-ai-code-review/scripts/review.js --auth-mode env-first --providers claude,gemini
Environment variables
- Claude:
ANTHROPIC_API_KEY(optional if already logged in) - Gemini:
GEMINI_API_KEYorGOOGLE_API_KEY(optional if already logged in) - Copilot: uses its own CLI auth flow
Optional: critique the review itself
Pipe the generated review into the existing response-rater skill:
node codex-skills/multi-ai-code-review/scripts/review.js --output markdown | node codex-skills/response-rater/scripts/rate.js --providers claude,gemini
Notes / constraints
- Requires network access to contact providers.
- For very large diffs, the runner samples hunks and reports what was omitted.