GitHub Copilot app for Beginners: How to teach Copilot about your codebase
Source
github.blog
Author
Christopher Harrison
Date
Context
A coding assistant needs a repository's conventions as well as the immediate task. GitHub's Copilot guide recommends storing recurring guidance in the project so developers need not explain it in every conversation. Repository instructions establish broad rules, while path-specific instructions add guidance for matching files.
Repeatable procedures belong in AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → skills, which can bundle instructions with scripts and supporting resources. This separates a standing convention, such as how components should be organized, from a procedure, such as which checks to run after an interface change. The guide suggests starting with one convention the team keeps explaining.
Key takeaways · AI-distilled
Repository-wide rules go in .github/copilot-instructions.md, while path-specific files in .github/instructions use an applyTo field, so Copilot applies both the broad rules and the file-type guidance when it works on a matching file.
Project skills live in .github/skills, one subfolder each, with a agent skillA reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.Full definition → whose description tells Copilot when to load it. GitHub's example quality-checks skill lists which scripts run tests and linting, what each checks, and how to troubleshoot common failures.
Custom agent profiles in .github/agents define a purpose, instructions, tools and project context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition → for recurring multi-file work. GitHub suggests them for tasks such as an accessibility review spanning components, templates, styles and interactions.
With the Playwright MCPThe Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.Full definition → server configured, Copilot can open the application in a browser and interact with it, so it can check that an updated interface behaves as expected, not only that the code passes automated checks.
GitHub's advice is to start small: pick one convention you keep explaining, add it to .github/copilot-instructions.md, and see how Copilot applies it on the next task. The Awesome Copilot community collection offers ready-made instructions, skills and agents to review and adapt.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
