claude-code-review.yml

name: Claude Code Review

on:
  pull_request:
    types: [opened, synchronize, ready_for_review, reopened]

jobs:
  claude-review:
    runs-on: ubuntu-latest
    permissions:
      contents: read
      pull-requests: write
      issues: write
      id-token: write

    steps:
      - name: Checkout repository
        uses: actions/checkout@v4
        with:
          fetch-depth: 1

      - name: Run Claude Code Review
        id: claude-review
        uses: anthropics/claude-code-action@v1
        with:
          claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
          track_progress: true
          prompt: |
            REPO: ${{ github.repository }}
            PR NUMBER: ${{ github.event.pull_request.number }}

            Review this pull request with a focus on:
            - Correctness bugs and edge cases
            - Code quality and adherence to this repo's CLAUDE.md conventions
            - Security implications
            - Performance considerations

            Provide detailed feedback using inline comments for specific issues.
          # Interactive-session tools (ScheduleWakeup/SendMessage/Monitor, background
          # Agent/Task execution) assume a persistent host that re-invokes the session
          # later. This action runs the SDK once per job and exits when the turn ends,
          # so the run silently reports success with no review ever posted. Upstream:
          # anthropics/claude-code-action#1462, #1499.
          claude_args: |
            --model claude-opus-4-8
            --allowedTools "mcp__github_inline_comment__create_inline_comment,Bash(gh pr comment:*),Bash(gh pr diff:*),Bash(gh pr view:*)"
            --disallowedTools "ScheduleWakeup,SendMessage,Monitor"
            --append-system-prompt "This session runs once inside a GitHub Actions job and cannot be resumed or notified later. Never call Agent or Task without run_in_background: false. Do not end your turn assuming a background subagent or scheduled wakeup will continue the work -- it will not. Wait for every subagent synchronously and only finish once the actual deliverable (the PR review) has been produced."
          # See https://github.com/anthropics/claude-code-action/blob/main/docs/usage.md
          # or https://code.claude.com/docs/en/cli-reference for available options