How to create a Skill for GitHub Copilot

This is probably old news for hardcore AI enthusiasts, but I only recently stumbled across Skills. Since we are fully invested in the GitHub Copilot tooling world in our product primedocs, I thought this might be worth a quick write-up.

What is a Skill?

A Skill is technically “just” a Markdown file with instructions for the LLM. In practice, a Skill can include all kinds of things:

• Use tool X for task Y.
• Execute these steps in exactly this order.
• Watch out for these product-specific rules.

A Skill bundles a workflow you would otherwise have to explain again and again.

How do you create a Skill?

I can only speak for GitHub here, but other tools (for example Claude) understand similar concepts too and may look for files in different locations.

For GitHub Copilot, create a folder under .github/skills named after your skill. Inside that folder, add a SKILL.md with metadata and your actual instructions. GitHub and VS Code also support alternative locations such as .claude/skills, .agents/skills, and personal skills in ~/.copilot/skills.

Schematically, it looks like this:

your-repo/
|-- .github/
|   |-- copilot-instructions.md
|   |-- skills/
|   |   |-- review/
|   |   |   |-- SKILL.md
|   |   |
|   |   |-- architecture-check/
|   |       |-- SKILL.md

Note: You can also place additional assets/tools inside the skill directory.

A minimal SKILL.md can look like this:

---
name: review
description: "Use when reviewing pull requests for risks, regressions, and compatibility issues."
---

# Review Workflow

1. Summarize what changed.
2. Identify functional risks and breaking changes.
3. Check backward compatibility.
4. Suggest concrete fixes with file references.

Quick checklist:

• SKILL.md must be named exactly SKILL.md.
• The name field is required.
• The description field is required.
• Use lowercase names with hyphens (for example review-skill).
• Keep the directory name and name aligned.

Isn’t there already a copilot-instructions.md file?

Yes, fair question: is this duplicate?

Short answer:

• copilot-instructions.md is more “always on” (global behavior).
• Skills are more “on demand” for a specific workflow.

Or in other words: Instructions are guardrails, Skills are playbooks.

Example: My Review Skill

Right now, my main task is less writing code and more reviewing and judging it: is this actually good for the product or not?

For that, I built my own review skill that combines two things:

• It respects our general coding guidelines.
• It focuses on product-specific pitfalls, for example when components are no longer backward compatible.

That saves me time in every review and gives me much more consistent results.

How do you use the Skill?

Option 1: via Copilot CLI

In Copilot CLI, you can invoke a skill in chat with a slash command:

/review "Review this PR and focus on backward compatibility risks."

If you need to start an interactive session first, use your local copilot CLI entrypoint and then run the slash command.

Option 2: in Visual Studio Code

In VS Code you can use skills as well (for example via chat slash commands). In full Visual Studio, I still have not seen equivalent skill support as of March 2026.

Summary

Skills are not magic: in the end, they are structured instructions in a Markdown file. Still, the leverage is huge, because you can standardize recurring workflows instead of prompting from scratch every single time.

If you already work heavily with GitHub Copilot, Skills are, in my opinion, a quick “low effort / high impact” step.

Hope this helps!