GitHub announced on November 6, 2026 that Copilot code review can now be managed through its REST and GraphQL APIs. Administrators can set policy from the enterprise level down to a single repository without opening the settings pages.
The same update shortens the overview comment Copilot leaves on a pull request and fixes several billing and reliability problems. The changes apply to the Copilot Pro, Pro+, Max, Business and Enterprise plans.
What Changed in the API
Until now the REST API let you ask Copilot for a review, pick how thorough it should be and fetch the outcome. The release adds attribution fields, so a script can tell which reviews and comments Copilot wrote.
A review record can also report the effort level it ran at, whenever GitHub has that detail. For each comment, the API names the skills, MCP servers and custom instructions that Copilot cited as its basis.
Who Can Manage Which Settings
Control follows the familiar GitHub role layers, and each layer touches only its own scope.
- Enterprise owners can switch code review on or off across all organizations, or hand that decision to each organization.
- Organization owners handle enablement, the default effort level and which runners are used.
- Repository admins choose the effort level and whether reviews draw on custom instructions and MCP tools.
Whatever an enterprise fixes stays fixed for the layers below it. The API offers the same effort choices as the web interface, and each settings response shows which levels the caller may use.
Token and Permission Limits
One limit matters for automation. Enterprise-level and organization-level settings endpoints, plus the organization runner endpoint, reject GitHub Apps and fine-grained personal tokens.
Repository endpoints accept both token types. Reading needs the Administration read permission and updating needs Administration write. A classic personal token works only if its owner administers the repository.
A More Compact Review Overview
The summary comment on a pull request now puts the findings first, in one block at the top. Effort details and survey prompts sit in a short footer below.
Anything Copilot skipped, or left out because of content exclusions, is grouped under a single "Files not reviewed" heading with file names and links to the reasons. Sections with nothing in them disappear, and a long list is cut off with a count of the rest.
Readers can also rate the overview. A thumbs-down opens a reason picker with options such as a disputed approval verdict, wording that runs too long, or findings that are wrong or petty.
Other Improvements and Fixes
Code review agent sessions now display the AI credits a review used. The repository rule that requests Copilot review automatically now honors a bypass list, so listed actors are not reviewed automatically.
When a pull request gets another review, findings that have nothing to do with your latest edits are folded away and later hidden. Balanced-effort reviews now run three reviewer models and a fourth step merges what they found.
- Turning code review off and back on no longer wipes billing and preview choices.
- A review that fails or times out is retried one time, and Copilot may fall back to a different eligible model.
- Actions workflows, Dependabot and organization-owned apps can once again ask for reviews on pull requests opened by bots, provided the organization covers the cost.
- A review no longer shows a false failure moments after it begins and loses its results, and reviews caught in a service update get extra time.
- Organizations that declined feedback collection no longer see survey prompts.
What It Means for Developers
The API is mainly useful for platform teams that manage many repositories. Instead of clicking through settings pages, they can script a standard review effort and consistent rules for custom instructions and MCP tools, and audit the result.
The attribution fields help reporting. You can now separate Copilot's comments from human ones when you measure how many findings get fixed. Check which token type your automation uses, because the organization and enterprise endpoints exclude GitHub Apps and fine-grained tokens.
The bypass list is worth a look for repositories where bots or release tooling open many pull requests. Exempting them can save review credits that would otherwise be spent on automated changes.
Conclusion
This update turns Copilot code review into something you can configure as code, from enterprise policy to repository rules. Together with the tidier overview and the reliability fixes, it makes reviews easier to run at scale. Review your token types and permissions before you automate it.
Sources
See my articles more often when you search Google.
