Skip to main content

Review a Pull Request with AI

Experimental

This page assumes Pull Request Reviews is on and a provider is connected. See Enable Pull Request Reviews.

Run an AI evaluation of an open pull request, review its findings, and decide what to publish. Nothing is posted to the provider until you publish it.

Open a pull request​

  1. In the Pull Request Reviews view, select a pull request on the Open tab.
  2. Read the Overview, the Changed files, and the Checks. Switch the diff between Unified and Side by side.

The status shown from the provider, such as PR: Open or PR: Approved, is independent of the AI evaluation.

Run the evaluation​

  1. Under Evaluation configuration, choose the Review agent and the Model. The Review scope covers correctness, security, and maintainability.
  2. Select Evaluate PR with AI. Large pull requests are split across parallel review workers.
  3. Follow the progress. Select Stop evaluation to cancel.

The evaluation uses the pull request's head commit at the moment it starts. If the pull request changes afterwards, the result is marked Evaluation outdated; select Run again to replace it.

Read the findings​

Findings lists the issues found, marked Critical, Warning, or Suggestion, each linked to the file and lines. The AI verdict appears as AI: Approved, AI: Changes required, or AI: No verdict. It is the AI Cockpit evaluation, not a decision published to the provider.

Review each finding against the code before you act on it.

Publish the review​

  1. Select the findings to include. Each becomes a draft comment on its line.
  2. Choose the Review action: Comment, Approve, or Request changes.
  3. Edit the Review summary.
  4. Select Publish review.

After publishing, the view shows the last decision confirmed by the provider, such as Last published: Approved. If that decision was made for an older commit, the view tells you.

Reviews of a Local repository cannot be published. Select Open in provider to continue on GitHub or Bitbucket.