Does Copilot code review use the PR branch’s instructions?

Does Copilot code review use the branch’s own instructions and agent skills when I test changes in the PR

Yes. Copilot code review reads repository custom instructions, agent instructions, and agent skills from the pull request’s head branch, not the base branch. That means changes you make to instruction files or skills in the PR are the ones Copilot uses during review.

That detail matters when you are testing a change to .github/copilot-instructions.md, a path-specific *.instructions.md file, AGENTS.md, or a skill under .github/skills. If the PR updates those files, Copilot can review the branch with the new rules in place before they merge. GitHub documents that same behavior with the example of merging my-feature-branch into main.

The part people get wrong is assuming Copilot reviews the base branch’s guidance and only the code diff from the PR. GitHub says the opposite for pull request review: Copilot reads the head branch’s instructions and skills. If the PR changes both code and the review guidance, the review reflects the branch state that exists in that PR.

The inconvenient part is that this behavior can hide mistakes in instruction changes if you expect a comparison against the old rules. A PR that edits the instructions and the code at the same time can make Copilot look “correct” for the new branch while still being inconsistent with the repository’s previous guidance. To check that case, review the instruction files themselves as part of the PR, not just the code output.

Copilot can also use repository-level agent skills and MCP servers when they are relevant to the review. GitHub says it is more likely to use that context when the repository or pull request gives clear signals, such as review-focused skill directory names, instructions that reference MCP context, or PR descriptions that mention identifiers tied to configured MCP servers.

That does not mean every review will use every skill you added. GitHub’s documentation says Copilot can use relevant skills, but it is not deterministic, and it is not guaranteed to spot every problem. If your branch adds a new skill, the safest check is to look for attribution at the bottom of Copilot’s review comment or open the linked review session logs.

A practical workflow is simple. Put the instruction or skill change in the feature branch, open the PR, request Copilot review, and inspect whether the review reflects the new guidance. If Copilot still behaves as if the old instructions are in force, first confirm the files are actually in the head branch and then check whether the skill name, description, or PR description gives Copilot enough signal to use it.

If you are comparing this with other Copilot surfaces, keep the contexts separate. GitHub’s docs for code review on GitHub.com say the head branch is used for review instructions and skills. Some other Copilot clients and pages discuss different instruction sources for other workflows, which is why the review-specific documentation is the one to trust for pull request review behavior.

If you want the shortest accurate rule, it is this: branch-specific review instructions and agent skills in the PR are the ones Copilot code review reads. Use the PR to test changes to those files, then validate the review comments instead of assuming the tool applied every rule exactly as intended.

For teams that want a single place to start, the GitHub Docs page on Copilot code review is the right reference, and DevConnect can be used to organize real testers and reviews around your own app work if you need a separate workflow for product testing: https://devconnectplatform.com. That link is for your own process, not a source for Copilot behavior.

What files does Copilot code review read from the branch

Copilot code review uses repository custom instructions, agent instructions, and agent skills from the head branch of the pull request. GitHub documents .github/copilot-instructions.md for repository-wide guidance, .github/instructions/**/*.instructions.md for path-specific guidance, AGENTS.md for shared standing rules, and skills in .github/skills for task-specific workflows.

Can I test a new review rule before merging it

Yes. GitHub explicitly says you can test instruction and skill changes in the same pull request without merging them first, because Copilot reads them from the head branch. That is the intended way to verify how a new instruction or skill affects code review.

Will Copilot always apply every skill or instruction

No. GitHub says Copilot review is non-deterministic and may overlook instructions, especially if the files are long or the signals are unclear. For that reason, GitHub recommends validating Copilot’s feedback carefully and supplementing it with human review.

How can I tell whether a skill was used in a review

GitHub says some review comments include attributions that identify the skill or MCP server used. You can also open the linked review session from the pull request timeline and inspect the session logs to see which context Copilot used.