Use built-in Claude /code-review for auto review

[AnnatarHe]

Updates .github/workflows/claude-code-review.yml so automatic Claude reviews follow the working rawback-app/web setup. The runner stays ubuntu-latest. No other files change.

What changed

• Prompt is now the built-in /code-review high <PR#> --comment, instead of a free-form prompt that asked Claude to skim the diff and post with gh pr comment. That free-form path was posting "No issues found" after a few seconds.
• Checkout uses actions/checkout@v7 with fetch-depth: 0 so the review can diff, log, and blame against the base branch.
• Permissions are pull-requests: write and issues: write, plus checks, statuses, and actions: read.
• show_full_output: true prints Claude's messages in the job log.
• Allowed tools are Bash(gh:*), Bash(git:*), and mcp__github_inline_comment__create_inline_comment, with additional_permissions for actions: read.
• Concurrency group claude-review-<PR number> cancels an in-progress review when a new push arrives.
• Drafts and bot PRs (release-please) are skipped. Reviews also run on ready_for_review and reopened.
• An "Assert the review actually ran" step fails the job when Claude spent 0 turns and $0, and warns when no subagents were launched.
• CLAUDE_CODE_DISABLE_BACKGROUND_TASKS was not set here, so nothing was removed.
• Comments that pointed at another repository's pull request were rewritten for this repo. The note that the repository is private was dropped, because this repository is public.

The review run on this pull request may be skipped, because the workflow file differs from the default branch. That is expected.

  

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.