Multi Code Review

A Claude Code skill that invokes external AI CLIs (Codex, opencode) to independently review your code changes and plans.

Features

• Cross-AI Review — Get a second opinion from Codex or opencode on your Claude Code work
• Pluggable Architecture — Easy to add new AI reviewers
• Code & Plan Review — Review git diffs or plan/design documents
• Single-Round Review — Quick one-shot review (multi-round coming soon)
• Hook Support — Auto-trigger reviews on git commit

Prerequisites

Install at least one reviewer CLI:

# Codex (OpenAI)
npm install -g @openai/codex

# opencode
go install github.com/opencode-ai/opencode@latest

Installation

npx skills add <your-github-username>/multi-code-review

Usage

In Claude Code, use any of these triggers:

• review code with codex
• review code with opencode
• multi code review
• /multi-code-review

CLI Usage

# Review staged changes with Codex
python3 scripts/review_engine.py --reviewer codex --staged

# Review working directory with opencode
python3 scripts/review_engine.py --reviewer opencode --working

# Review a git range
python3 scripts/review_engine.py --reviewer codex --range "main...HEAD"

# Review specific files
python3 scripts/review_engine.py --reviewer opencode --path src/main.py src/utils.py

# Review a plan document
python3 scripts/review_engine.py --reviewer codex --plan docs/design.md

Hook Configuration

Add to your Claude Code settings.json:

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Bash",
        "pattern": "git commit",
        "command": "bash <skill-path>/hooks/post-commit-review.sh"
      }
    ]
  }
}

Output

Review artifacts are saved to $TMPDIR/multi-code-review/<timestamp>/ (or a custom path via --output-dir):

File	Content
prompt.md	Full prompt sent to the reviewer
assessment.md	Reviewer's assessment
result.json	Structured result (verdict, findings, metadata)

License

MIT