⚖️ GitCourt — AI Code Review Tribunal

Three AI agents debate your Pull Request. The Judge delivers the verdict.

What is this?

GitCourt takes any GitHub Pull Request and runs it through three AI agents:

• 🔴 The Prosecutor — aggressively finds every bug, anti-pattern, and security issue
• 🔵 The Defender — justifies the code, provides context, and counters the prosecution
• 🟡 The Judge — weighs both sides and delivers a verdict with a score (1-10)

It's code review, but make it a courtroom drama.

---

⚡ GitHub Action (one-liner)

Add GitCourt to any repo — every PR gets an AI courtroom verdict automatically:

# .github/workflows/gitcourt.yml
name: GitCourt

on:
  pull_request: [opened, synchronize]

jobs:
  review:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: mamoor123/gitcourt@main
        with:
          openai-api-key: ${{ secrets.OPENAI_API_KEY }}
          # anthropic-api-key: ${{ secrets.ANTHROPIC_API_KEY }}  # if using anthropic

That's it. Every PR now gets a sticky comment with the verdict:

---

🖥️ CLI Usage

# Install
pip install "git+https://github.com/mamoor123/gitcourt.git#egg=gitcourt[all]"

# Run with OpenAI
export OPENAI_API_KEY="sk-..."
gitcourt https://github.com/user/repo/pull/123

# Run with Anthropic
export ANTHROPIC_API_KEY="sk-ant-..."
gitcourt https://github.com/user/repo/pull/123 --provider anthropic

# Run 100% locally (no API keys!)
ollama pull llama3.1
gitcourt https://github.com/user/repo/pull/123 --provider ollama

# JSON output (for CI/CD)
gitcourt https://github.com/user/repo/pull/123 --json

📋 Options

Flag	Description	Default
--provider	openai, anthropic, or ollama	openai
--model	Model name	Provider default
--api-key	API key (or use env vars)	—
--ollama-url	Ollama server URL	http://localhost:11434
--temperature	LLM temperature	0.7
--json	JSON output (for CI)	off

🔑 Environment Variables

Variable	Description
OPENAI_API_KEY	OpenAI API key
ANTHROPIC_API_KEY	Anthropic API key
GITHUB_TOKEN	GitHub token (optional, higher rate limits)
OLLAMA_URL	Ollama server URL
GITCOURT_PROVIDER	Default provider

🔧 CI/CD Integration

GitCourt posts a sticky comment on every PR with the verdict. You can also gate on the score:

gitcourt $PR_URL --json > verdict.json
score=$(jq '.verdict.score' verdict.json)
if [ "$score" -lt 5 ]; then
  echo "GitCourt says: CHANGES REQUESTED"
  exit 1
fi

Why?

Because regular code review is boring, and every dev secretly wants an AI to roast their code in a courtroom setting.

Also because:

• Multi-perspective review catches more issues than single-agent review
• The debate format surfaces nuanced trade-offs, not just "fix this"
• The score gives a quick signal for PR health
• It's hilarious and your team will actually use it

🏗️ How It Works

PR URL → GitHub API → Diff
                          ↓
              ┌───────────┼───────────┐
              ↓           ↓           ↓
         Prosecutor    Defender     (waiting)
         (find bugs)   (defend code)
              ↓           ↓
              └─────┬─────┘
                    ↓
              Judge (weigh arguments)
                    ↓
              Verdict + Score → PR Comment

Each agent gets the full PR diff plus the previous agent's arguments. The Prosecutor goes first, the Defender responds, and the Judge sees everything before ruling.

📜 License

MIT — fork it, ship it, go viral.