GitHub shipped changes to Copilot code review on July 17 that change how it picks up instructions for reviewing your pull requests. Two changes matter for anyone who already leans on a written instructions file to steer an AI coding tool.
What changed
Copilot code review reads instructions from files like copilot-instructions.md, *.instructions.md, agent skill files, and AGENTS.md to know what to look for when it reviews a pull request. Until now, it read those files from the base branch, the branch you’re merging into. That meant you had to merge a change to your instructions before you could see it take effect.
Copilot code review now reads those files from the head branch instead, the branch that actually contains your pull request. You can edit your instructions and test them on the same branch you’re already working on, before anything merges.
The second change is broader file support. Copilot code review now also reads REVIEW.md, GEMINI.md, and CLAUDE.md files if they exist in your repository. If you already maintain a CLAUDE.md for Claude Code or a similar instructions file for another tool, Copilot code review picks it up automatically instead of requiring a separate, Copilot-specific file.
GitHub also added a copilot-code-review.yml file for configuring the runtime environment Copilot code review uses, including custom setup steps and independent runner settings. That part is aimed at teams that already manage GitHub Actions workflows for their repository; it’s not something you need to touch to get the instruction-file changes above.
Why it matters
A written instructions file only works if the tool reading it actually reflects what you wrote, and reflects it before you’ve committed to it. Reading from the head branch fixes that: you can iterate on your review instructions the same way you’d iterate on code, and see the result before merging.
The broader file support is a smaller but telling detail. Markdown instruction files like AGENTS.md and CLAUDE.md started as a way to steer one specific tool. Copilot code review now reading CLAUDE.md alongside its own formats suggests these files are becoming a shared convention across coding tools, not something you have to duplicate for every tool you use.
Who should care
If you maintain a CLAUDE.md, AGENTS.md, or similar instructions file in a repository that also uses GitHub Copilot code review, that file now feeds both tools without extra setup. If you’ve been tweaking review instructions and waiting for a merge to see whether they worked, that friction is gone.
The custom setup steps and runner configuration are relevant only if you already manage your repository’s GitHub Actions workflows. If that’s not part of your normal process, you can ignore that part of this release entirely.
What builders should do next
If you keep a CLAUDE.md or AGENTS.md in a repository reviewed by Copilot, no action is needed. It’s already read automatically. If you want to test a change to your review instructions, edit the file on your feature branch and open or update the pull request. Copilot code review will use the version on that branch, not the one already merged.
End of article