Summary
Anthropic has released two distinct ways to bring Claude Code into GitHub pull request workflows: a fully managed code review service and the `anthropics/claude-code-action@v1` GitHub Action for custom CI. This official tutorial explains when to use each and how to configure both.
The managed Code Review service activates through Claude Code’s admin settings and runs via the Claude GitHub App, which an organization admin installs and points at selected repositories. Once live, a set of review agents analyzes each PR diff against the full codebase and posts findings as inline comments on specific lines tagged by severity, with a deduplication and ranking pass so reviewers see a short list of real issues rather than a flood of nits. A summary table appears in the check run. The service never approves or blocks a PR — the judgment call stays with a human. It is currently in research preview and limited to Team and Enterprise plans. Fixes are applied locally using the `/code-review –fix` command inside a Claude Code session.
The GitHub Action handles everything beyond passive review: implementing changes requested in PR comments, generating scheduled reports, or responding to any GitHub event. Key inputs include `anthropic_api_key`, `trigger_phrase` (defaults to `@claude`), `prompt`, and `claude_args` — a passthrough string for CLI flags like `–max-turns`, `–permission-mode`, and `–allowedTools`. The video walks through two canonical workflow files: one that listens for `@claude` mentions on PR and issue comments, and a cron-based workflow that fires daily at 9 UTC.
📺 Source: Claude · Published September 03, 2026
🏷️ Format: Tutorial Demo
