← All use cases

Review GitHub pull requests

Spot regressions and possible issues ahead of human review.

Use ZeroTwo's Code Review page in the desktop app to run a private review of a GitHub pull request, with the bugs, regressions and risks it finds, before you merge.

Easy
5m

Skills & Plugins

  • Security
    Adds security-focused scans, including a diff scan you can run on a pull request or other Git-backed change ($security-diff-scan).

Starter prompt

Review this pull request with a focus on auth and permission checks, new dependencies, and anything that logs or exposes secrets. Point out missing tests for the changed paths and any behavior change the description doesn't mention. Reference specific files and lines, and rank findings by how likely they are to hurt production.

How to use

Open Code Review in the desktop app, which needs the GitHub CLI (gh) signed in on your computer. Select a pull request and choose Review with ZeroTwo; the review happens when you ask, not automatically on every new pull request. It stays private to you: the review reports bugs and risks to you, and ZeroTwo does not change code or comment on, approve or post anything to GitHub for it. Anything the page itself writes to GitHub, such as comments, reviews, reviewers or merging, shows a confirmation dialog first. Missing tests or documentation gaps are only covered if you ask for them in the starter prompt or your review instructions. Separately, when a pull request has a failing check, a merge conflict or an unresolved comment, a Fix button appears whether or not you have run a review. Fix asks ZeroTwo to work on the pull request branch, run the relevant tests and push the result to that branch, and that push does not show a confirmation dialog. Review what it pushed before you merge. Details are in Code review.

Define review guidance

The gear beside Review with ZeroTwo opens Review instructions, where you describe what to flag and what is safe. They apply to every review you start from the page, and Save and run begins one immediately.

Experiment cohorts: don't filter treatment comparisons on post-exposure behavior.

Related use cases