How to Teach GitHub Copilot Your Codebase
GitHub outlines four ways beginners can give Copilot project context, from custom instructions to external tool integration.
Original source published: October 12, 2026
GitHub’s latest beginner‑focused guide explains how to give Copilot the project context it needs to follow a repository’s conventions. By adding a .github/copilot-instructions.md file—or path‑specific files in .github/instructions—teams can embed architecture, naming, and testing standards directly in the repo.
The guide also introduces three additional mechanisms. Agent skills are folders placed in .github/skills that contain a SKILL.md and optional scripts, allowing Copilot to load repeatable procedures such as quality checks. Custom agents live in .github/agents and provide a dedicated role for larger, multi‑file tasks like accessibility reviews. Model Context Protocol (MCP) servers link Copilot to external tools or data sources, for example a Playwright server that can drive a browser test.
GitHub suggests starting with a single convention, using community‑shared instructions from Awesome Copilot, and combining the four options as needed to make Copilot’s output align with a project’s workflow.