What changed
GitHub’s October 12 Copilot tutorial turns a common frustration—re-explaining a project’s conventions—into four concrete places to put that context. Repository-wide `.github/copilot-instructions.md` files define baseline rules such as architecture, naming and tests. Path-specific files in `.github/instructions` apply narrower guidance only to matching files.
For repeatable procedures, GitHub points to agent skills under `.github/skills`, where a folder can include a `SKILL.md` plus scripts or other resources. A quality-check skill, for example, can tell an agent which commands to run after a UI change. Custom agent profiles under `.github/agents` are for broader recurring roles that may span multiple files and tasks.
What the result actually shows
MCP servers connect the agent to external tools and data. GitHub’s example is Playwright, so the assistant can use browser automation as part of a workflow. The practical distinction is useful: instructions set defaults, skills package procedures, agents define specialized roles, and MCP supplies connected capabilities.
This does not make Copilot automatically correct. It gives the model project-specific context that would otherwise be repeated in prompts, and it makes the workflow easier for a team to inspect in version control. Start with one convention you routinely explain, add it to the repo, then check the next change rather than trying to write an all-encompassing rulebook.
Source published 2026-10-12. Coverage is based on the maker’s announcement and demonstration.