Dashboard
Github Comment
Workflow Whiteboard
Dasboard - Failed
Dashboard - F - Screenshot

🐢 CodeTurtle: How We Stopped Kevin

Our intern Kevin merged a pull request Friday at 4:59 PM. He used ChatGPT to vibecode his entire PR!

It passed CI. It passed unit tests.
It also took down production in 12 seconds.

That's why we built CodeTurtle: a GitHub App powered by BrowserBase & CrewAI that tests and simulates your pull requests like real users before they hit main.

Now, Kevin still vibe codes... but CodeTurtle vibe-tests first.

Prod is safe. The vibes are strong. 🐢

We used BrowserBase's stagehand to autonomously test the PRs. BrowserBase Uses Gemini Flash, to efficiently test the Application. The Github app agent is built with CrewAI and we use to orchestrate different agents. Both the BrowserBase Agent and CrewAI, are both connected to W&B's Weave, and we use Weave to observe log the data.