Claude AI Agent for review PRs

[ebattat]

name: Claude PR Review
on:
  pull_request:
    # Trigger on open and every new push
    types: [opened, synchronize]
    # Skip review for non-code changes to save API costs
    paths-ignore:
      - '**.md'
      - 'docs/**'
      - '.gitignore'

jobs:
  claude_review:
    # Skip if it's a Draft PR (highly recommended)
    if: github.event.pull_request.draft == false
    runs-on: ubuntu-latest
    permissions:
      contents: write
      pull-requests: write
      issues: read

    steps:
      - name: Checkout Code
        uses: actions/checkout@v4
        with:
          fetch-depth: 0 # Essential: Allows Claude to see full repo context

      - name: Claude Code Agent
        uses: anthropics/claude-code-action@v1
        with:
          # Use your secret (Settings > Secrets > Actions)
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
          
          # Custom Instructions for benchmark-runner
          prompt: |
            You are a Senior Performance Engineer. Review this Python PR for:
            1. Logical bugs in benchmark timing or data collection.
            2. Pythonic best practices (type hints, PEP 8).
            3. Security (unsafe subprocess calls or shell executions).
            
            Format your response with a summary table first, then specific line comments.
          
          # 2026 Model Choice
          claude_args: >
            --model claude-3-7-sonnet-latest 
            --max-turns 5 
            --allowedTools "Read,Edit,Write"

[ebattat]

@arpsharm, wdyt ?