Discuss your project
Architecture IA

Vary Copilot review depth while keeping your safeguards

One pull request fixes a word and another changes sign-in. Test Copilot’s Adaptive effort and secret detection on changes that reflect the work your own repository receives.

Faire varier la profondeur d’une revue Copilot sans perdre ses garde-fous

Two pull requests are waiting. One fixes a sentence and the other changes a sign-in check. Giving both the same depth of automated review can send the team toward the wrong comments.

What changed in Copilot

On November 17, GitHub introduced Adaptive effort in public preview. Copilot chooses Lite, Balanced, or Max up to the configured limit. Reviews show the selected effort, and Balanced remains the default. GitHub also announced public preview for secret detection in review comments. Cross-repository context is in preview for public repositories. The separate general availability announcement concerns review in Azure Repos.

Test before changing repository settings

  1. Create a test repository with three focused changes: wording, logic, and a fully dummy credential matching an approved sample format. Include a similar example that should pass.
  2. Run repeated Balanced and Adaptive reviews. Record the chosen effort, duration, expected findings, and low-value comments. Use identical revisions and effort limits.
  3. Check that human approval and branch protection work as intended. Review secret alerts as findings and keep existing publication checks.

The previous Partitech article covered the merge decision. Here the pilot measures review depth and alert quality before that decision.

Share this article