Copilot code review uses repository instructions, not only skills

Did Copilot code review start using repository skills or custom instructions

Yes. GitHub Copilot code review uses repository custom instructions, and it can also use agent skills when those are relevant. Custom instructions are the baseline, skills add task-specific context. GitHub documents both paths for code review, and it now treats them as separate ways to improve review quality.

The part people get wrong is thinking this is an either-or choice. It is not. Copilot code review can read repository-wide instructions from .github/copilot-instructions.md, path-specific instructions from .github/instructions/**/*.instructions.md, and AGENTS.md for shared repository context. GitHub also says review-focused skills can be used when they are relevant.

If you are asking what started first, the documented customization path for code review is custom instructions. GitHub’s code review docs describe custom instructions as a way to enhance the reviewer’s knowledge of the repository, then separate skills as a different mechanism for task-specific workflows. In practice, that means instructions are the persistent repository layer, while skills are the optional workflow layer.

Repository-wide custom instructions are the simplest place to put standing review rules. GitHub says .github/copilot-instructions.md is for repository-wide guidance that should apply across the codebase, while .github/instructions files are for narrower scopes. That is the material Copilot code review is meant to pick up automatically when it reviews a pull request in that repository.

Skills are different. GitHub’s recent changelog says agent skills for Copilot code review are generally available, and that a skill can extend analysis with context and instructions specific to a repository or organization. GitHub also describes skills as task-specific review workflows. So if you are choosing what to add for a stable review rule, custom instructions are the first stop. If you are defining a narrower workflow, a skill is the stronger fit.

The inconvenient part is that both can be true at once. A repository can have a Copilot instruction file, path-specific instruction files, an AGENTS.md file, and one or more skills. The reviewer may use whichever sources fit the current review. That means a missing or vague instruction file can leave you thinking Copilot “ignored” the repo, when the real problem is that the repo never gave it a clear rule.

A concrete example helps. If your team wants every review to flag missing tests, .github/copilot-instructions.md is the place for that standing expectation. If only files under payments/ need a special review lens, use a path-specific instruction file. If the review needs a repeatable workflow, such as checking a specific release process, a skill is the better fit. GitHub’s docs separate those cases for a reason.

One more thing people miss: support is not identical across every Copilot surface. GitHub’s custom-instructions support page shows that Copilot code review supports repository-wide instructions, path-specific instructions, and agent instructions, but support varies by product and platform. That is why a setup that works in code review may not behave the same way in every Copilot experience.

So the short answer is: Copilot code review did not switch from one single mechanism to the other. It uses repository custom instructions, and it can also use repository skills where those are available and relevant. If you want predictable results, start with instructions, then add a skill only when you need a separate workflow layer.

If you want the exact repo files to check in your own project, GitHub’s docs point to .github/copilot-instructions.md for repo-wide rules, .github/instructions/**/*.instructions.md for path-scoped rules, AGENTS.md for shared agent context, and .github/skills/.../SKILL.md for skills. That is the current shape of the system GitHub documents for code review.

If you are building a page or internal runbook around this, link the team to the source docs and keep the distinction explicit. On GitHub, custom instructions are the always-on repository guidance, skills are the task-specific extension point, and Copilot code review can draw on both when they are present. For a single place to start, GitHub’s own docs are the right reference.

If you need a working example inside a repository, GitHub shows how a .github/copilot-instructions.md file is applied to reviews, and it explains how path-specific instruction files narrow the scope. That is usually the fastest way to verify behavior in a real repo, because you can change one rule, open a pull request, and see whether the review follows the new instruction.