Jun 14, 2026

Built the CodeRabbit for myself

Last week I tried CodeRabbit on one of my side projects and it felt pretty good. Aside from leaving feedback on a PR automatically, the premium tier delegates fixes to Claude, so Claude and CodeRabbit end up having a conversation right in the PR thread. That part is genuinely cool to watch, and the feedback it's giving is spot on. What I didn't love was the $48 a month for premium, so I decided to build a smaller version of it for myself.

My requirements were simple:

• Review the PR whenever one gets created.
• Post the feedback to a Slack channel, along with a ready-to-use prompt for an LLM.

Then I can copy that prompt straight into my coding agent and let it update the PR. No $48 subscription, and I keep control over which model does the work.

How it works

The whole thing hangs off a GitHub Action that runs on every new pull request. It grabs the diff, sends it to the model with a review prompt, and the model comes back with feedback plus a suggested fix prompt.

I used a cheaper model for the review pass instead of a frontier one. For catching the obvious stuff (naming, missing error handling, an unhandled edge case) it's more than good enough, and the cost per review is a rounding error compared to the subscription.

The Slack part

Once the review is done, the Action posts two things to a Slack channel: the human-readable feedback, and a second message formatted as a prompt I can hand to a coding agent. The second message is the real trick. It's written so I can copy it without editing and the agent has everything it needs to make the change.

What I'd change

It isn't as polished as CodeRabbit. There's no back-and-forth conversation in the thread yet, and the review quality depends a lot on how I word the prompt. But for a personal project it does exactly what I needed, and I learned more building it than I would have clicking "subscribe."