Every codebase carries its own house rules: how components are structured, which scripts to run, which conventions apply per framework and file type. Copilot works better when it knows those rules, and much of that knowledge can live inside the repository so it is picked up automatically rather than re-explained at the start of each task.

Four mechanisms carry that context, and they differ in scope:

  • Custom instructions — the conventions Copilot should follow.
  • Agent skills — packaged, repeatable procedures.
  • Custom agents — specialized guidance for larger tasks.
  • MCP servers — external tools and data sources.

Repository and path-specific instructions

Adding a custom page size to an application involves more than making the change; the new code should also match existing team patterns. Repository custom instructions supply that baseline. A .github/copilot-instructions.md file holds guidance that applies across the repository — preferred architecture, naming conventions, expectations for testing and validation.

Path-specific files inside .github/instructions narrow that scope. Each uses an applyTo field to declare when the guidance is relevant. When Copilot touches a matching file, it can draw on both the repository-wide instructions and the specific ones, so broad rules stay in one place without forcing every file type into the same mold.

Skills for procedures that repeat

Some guidance only matters during a particular procedure: running quality checks, preparing a release, debugging a failed workflow. An agent skill is a folder holding instructions and, optionally, scripts and other resources that Copilot loads when relevant. Project-level skills live in .github/skills, each in its own subfolder.

A .github/skills/quality-checks folder, for instance, could contain a SKILL.md describing when to use the skill, which scripts run the project's tests and linting, what each script checks, and how to troubleshoot common failures. The description in SKILL.md is what helps Copilot decide the skill applies. After an interface update, Copilot can load that skill and follow the established testing and linting process instead of guessing at commands.

Agents for work beyond a single procedure

An accessibility review may require inspecting components, templates, styles and interactions across many files — more than one procedure. Custom agents create specialized versions of Copilot for that kind of work. A repository-level agent profile lives in .github/agents and can define the agent's purpose, instructions, tools and relevant project context. Reach for one when a recurring type of work benefits from a dedicated role or perspective, particularly when the task spans multiple files and several tools or context sources.

MCP servers extend reach outside the repository

Instructions, skills and agents shape how Copilot works within your project. MCP servers change what it can access and do. They connect Copilot to external tools, data sources and services — a documentation server for framework guidance, or Playwright to open and interact with your application in a browser.

Available servers can be browsed in the Copilot app, or you can add a custom one. Once Playwright is configured, a prompt such as Use the Playwright MCP server to test the functionality manually lets Copilot open the application and interact with it much as you would. That verifies not only that automated checks pass, but that the updated interface behaves as expected.

Where the four fit together

  • Custom instructions define conventions across the repository or for particular files.
  • Agent skills package repeatable procedures Copilot loads when relevant.
  • Custom agents provide focused expertise for broader tasks.
  • MCP servers connect Copilot to outside tools, services and data.

They solve different problems and are most useful combined. Four questions point at where to start:

  • What project convention do I keep explaining?
  • What multi-step procedure do I repeat?
  • What complex task could benefit from a dedicated custom agent?
  • What outside tool or source does Copilot need?

Ready-made instructions, skills, agents and more are shared through Awesome Copilot. Find one addressing a workflow you already repeat, review its contents, and adapt it to your repository's tools and standards. A small first step works: pick one convention you keep explaining, add it to .github/copilot-instructions.md, and watch how Copilot applies it on the next task.