Run a two-pass bug check before opening a PR

Use an agent to flag possible bugs first, then have it verify each one before any pull request is opened.

Done with Custom agentยท Intermediateยท Nov 2026

What they did
They used a tool-using loop called roast on code or prose before a PR was created. The agent made one pass to surface possible bugs in a structured list, then a second pass to challenge each item by checking the code and, when useful, running tests or small scripts. Only after that did it send the results to a person or a harness.
How it went
The post was just an announcement, so it did not report any bugs found or any counts. It also gave no false-positive rate or other measured outcome.
Tips
A setup like this works best when the verifier is stricter than the finder and every claim needs evidence. Keep the agent in read-only mode until you approve any fix, submission, or other irreversible step.

Try it yourself with Claude Code

In [repo or path], review [branch, commit, or PR] before I open a pull request. Do this in two stages: first, scan for every plausible bug or flaw and list them with a short reason; second, examine each item like a skeptic and try to confirm or rule it out using the code, tests, or small scripts. Return only issues that you can support with evidence, and for each one include the file, line, proof, and a fix suggestion. Pause and ask for my approval before making any change, submitting anything, or opening a PR.

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

Read the original on bsky.app โ†—

Get 3 agent uses like this every morning

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