GitHub adds central controls for Copilot code review
GitHub has added programmatic management of Copilot code review settings, letting administrators control policies across enterprise, organization and repository levels. New comment metadata identifies the tools and instructions used in a review. The update also reorganizes the review overview and states why particular files were left out of an inspection.
Artificial Intelligence··Midday
Enterprise policies bind lower-level settings
GitHub’s November 6 update gives administrators programmatic control over Copilot code review, its automated examination of proposed code changes. Enterprise administrators can enable or disable reviews across organizations or let organization administrators decide. Organization settings cover default effort and runner configuration. At repository level, administrators choose effort and decide whether custom instructions apply. Policies set by an enterprise remain binding at lower levels.[1]
Comments identify their reported inputs
New attribution fields distinguish Copilot reviews and comments, with effort metadata where available. Comments can also identify skills, external tool servers and custom instructions that Copilot reports using. Authentication support differs by management level: enterprise and organization settings endpoints exclude GitHub Apps and fine-grained personal access tokens. Repository endpoints accept both when the appropriate Administration permission is present; classic tokens require repository administrator access.[1]
Review overviews list excluded files
The revised overview brings findings to the top and moves effort information to a footer. It groups unreviewed files with their names and reasons for exclusion, while feedback moves to a dropdown. Balanced reviews use three reviewing models and another model to combine results. Failed or timed-out reviews receive one automatic retry with another eligible model. Other fixes preserve billing settings when automatic review is switched off and back on.[1]
