GitHub lets Copilot choose how deeply to review each code change
The public preview chooses among three effort levels without exceeding configured limits. Related updates add public-repository context, credential checks and repository-level outcome tracking.
GitHub’s November 17, 2026 Copilot update adds controls for allocating review effort, checking for leaked credentials and measuring what teams do with review feedback. Adaptive can select Lite, Balanced or Max for each pull request, but cannot exceed an administrator-set limit; Balanced stays the default unless changed. The package also extends reviews with context from related public repositories, while making Copilot code review generally available for Azure Repos. These features give teams more options to tailor and assess AI review, with several capabilities still in preview.
01
Adaptive, cross-repository context and secret detection are in public preview for Copilot Pro, Pro+, Max, Business and Enterprise; GitHub warns preview features may change.
02
Secret detection combines pattern-based checks covering more than 500 secret types with language-model detection for credentials without recognizable formats.
03
Repository insights reports comment actions, later human change requests, unreviewed merged pull requests and trends; the Copilot usage metrics policy must be enabled.
Developers can now let GitHub Copilot decide how closely to examine a proposed code change, rather than choosing the review depth themselves. GitHub’s November 17, 2026 update puts that decision in a public preview called Adaptive, alongside new checks for exposed credentials, context from related public repositories and tools for tracking review outcomes.
Copilot chooses the effort, administrators set the ceiling
Until this preview, someone had to select a review effort level in settings or when requesting a review. Adaptive instead assesses each pull request—a proposal to merge code changes—for complexity and risk, then chooses Lite, Balanced or Max. Each review displays the level it selected.
GitHub illustrates the distinction with two kinds of work: a documentation update may need only Lite, while a change to authentication logic benefits from Max. That is the intended allocation of effort, rather than a requirement to run every proposed change through the deepest available review.
Adaptive can be selected for one pull request from the Reviewers bar, or made the default in personal, repository, organization or enterprise settings. Balanced remains the default unless that setting changes. Adaptive cannot choose an effort level above the maximum configured for the repository, organization or enterprise.
Looking beyond the changed repository
The cross-repository preview gives the reviewer more context than the project containing the proposed change. Copilot can read related public repositories to find code that depends on that change. It can also draw on documentation, custom instructions and skills stored elsewhere. The preview is limited to public repositories.
Secret detection adds a different check inside the review itself. Copilot flags exposed credentials in comments on the pull request, so developers can remove them before merging. Teams turn it on through the Copilot code review controls in repository or organization settings.
GitHub says the credential checks combine two approaches. For known formats, it uses rule-based detectors, including text-pattern matching, covering more than 500 secret types developed through partnerships with more than 150 technology companies. For credentials without recognizable formats, it uses large language models and GitHub’s purpose-built secret-detection model.
Tracking feedback and reaching Azure Repos
The update also adds a way to see what happens after Copilot leaves feedback. A new Repository insights page, the first item in a repository’s Insights tab, brings review outcomes together. Its four summary measures track developer responses, later human review and work merged without a Copilot review:
Copilot review comments that developers acted on.
High-severity Copilot comments that developers acted on.
Copilot-reviewed pull requests with no later human request for changes.
Merged pull requests that did not receive a Copilot review.
The page also shows the share of evaluable comments addressed, including the underlying counts, excluded unevaluable comments and trends over time. Feedback categories are available too. Anyone with read access to the repository can view its outcomes; the Copilot usage metrics policy must be enabled to support the functionality.
For teams hosting code in Azure DevOps, Copilot code review for Azure Repos has moved from public preview to general availability. Administrators can enable it at the organization, project and repository levels. Branch policies can request reviews automatically, or developers can request them manually from the Reviewers list.
Azure Repos teams can set Lite or Balanced as their default review effort and apply and commit Copilot’s suggestions while resolving the related comments. Usage is billed in GitHub AI credits to the Azure subscription linked to their Azure DevOps organization, with cost tracking and budget alerts available through Azure Cost Management.
Reader comments
Newest comments first. Replies stay oldest first.