Draft PR reviews, then send them yourself

They used Claude Code to prepare GitHub review drafts for queued pull requests, while keeping final edits and submission under their control.

Done with Claude Code· Intermediate

What they did
They built Cerber to watch GitHub for pull requests that asked for their review. It ran Claude Code through their existing CLI login and created draft reviews in a local web interface. The drafts included a summary, a walkthrough of the changes, and suggested inline comments, and they could talk through the findings, rewrite them, or remove them before sending.
How it went
It helped them get through a backlog that had reached 15 pull requests. It also kept AI comments from going straight to GitHub without review. The tradeoffs were that code still went to Anthropic, the setup required a paid Claude Code subscription, and approvals only auto-send if they are opted in and pass a confidence check.
Tips
Keep the final send step manual if you want a review buffer between the agent and GitHub. Treat local drafts as a working queue, and use read-only execution unless you fully trust the repo author or organization.

Try it yourself with Claude Code

Check the open pull requests in [repository] that are waiting on my review. For each one, read the diff and prepare a review draft with a short summary, a change-by-change walkthrough, and inline comments ordered by importance. Save each draft to [location] so I can edit it later. Do not post, approve, or delete anything; stop and ask me before any action that would send, submit, or remove a review. When all drafts are ready, tell me which PRs you covered.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on reddit.com ↗

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.