GitHub Copilot Skills: How to Use SKILL.md with Copilot Agent Mode

GitHub Copilot added SKILL.md support in April 2026 through its agent mode. You can now teach Copilot custom workflows using the same SKILL.md files that work across Claude Code, Cursor, Codex CLI, and 20+ other agents. Here's how it works.

Quick Answer: GitHub Copilot reads SKILL.md skills through its agent mode in VS Code and GitHub.com. Place skills in .github/skills/ in your repo for project-specific skills, or in your global Copilot config directory for skills that work across all projects. Copilot loads the right skill automatically based on what you ask it to do.

What changed in April 2026?

Microsoft announced SKILL.md compatibility for GitHub Copilot as part of the broader agent mode rollout. This means Copilot's agent can now read structured markdown instructions from SKILL.md files and follow them during code generation, review, and other tasks.

Before this, Copilot relied on its built-in training and the .github/copilot-instructions.md file for project context. SKILL.md adds task-specific skills on top of that general context. The two systems complement each other: copilot-instructions provides always-on project context (similar to .cursorrules or CLAUDE.md), while SKILL.md skills activate only when relevant.

Recommended skills

designing-hybrid-context-layers

by https://loreto.io · 15

Architects the right retrieval strategy for every query — teaching your agent wh…

$10

ai-automation-qa-pack

by Roy Yuen · 9

Professional QA & UAT documentation generator for AI automation agencies and com…

$5

Bounty Security Pattern Master Library — 399 Vulnerability Patterns

by Atlas Agent Suite · 8

A premium library of 399 vulnerability patterns and DeFi attack vectors for AI-d…

$75

Browse all Github Copilot skills

How is this different from copilot-instructions.md?

.github/copilot-instructions.md is always-on context that Copilot reads for every interaction. It contains project-wide conventions like coding style, preferred libraries, and architectural patterns.

SKILL.md skills are task-specific. A code review skill only activates when you ask Copilot to review code. A testing skill only activates when you ask it to write tests. This keeps Copilot's context focused and prevents instruction overload.

Think of it this way: copilot-instructions.md is "how this project works." SKILL.md skills are "how to do this specific task well."

Where do I put skills?

In your repository: .github/skills/<skill-name>/SKILL.md. These skills are version-controlled and available to every team member who opens the project.

Globally: ~/.config/github-copilot/skills/<skill-name>/SKILL.md on macOS/Linux, or %APPDATA%\github-copilot\skills\<skill-name>\SKILL.md on Windows. Global skills apply to every project.

The folder structure is the same as every other agent:

.github/skills/
├── code-reviewer/
│   ├── SKILL.md
│   └── references/
│       └── owasp-top-10.md
└── test-generator/
    └── SKILL.md

Do existing skills work with Copilot?

Yes. Any SKILL.md file built for Claude Code, Cursor, or any other agent works with Copilot. The format is the same: YAML frontmatter with a name and description, followed by markdown instructions. Copilot reads the description to decide when to activate the skill and follows the markdown body as instructions.

Some agent-specific metadata fields in the frontmatter (like Claude Code's context: fork or Cursor's globs) are ignored by Copilot, but the core instructions work without modification. This is the whole point of the SKILL.md open standard: one skill, every agent.

How does Copilot pick which skill to use?

The same way other agents do: by matching your request against the description field in each skill's YAML frontmatter. If you ask "review this code for security issues" and you have a skill with description: Use when the user asks to review code, find bugs, or check for security issues, Copilot loads that skill.

If multiple skills could match, Copilot picks the most relevant one based on the specificity of the description match. Writing clear, trigger-based descriptions with multiple phrasings helps Copilot pick the right skill consistently.

For guidance on writing effective descriptions, read How to Write a SKILL.md Description That Triggers.

Copilot skills for teams

The project-specific approach (.github/skills/) is particularly powerful for teams. When you commit skills to your repository, every team member gets them automatically. This means you can standardize workflows across your engineering team:

A shared code review skill ensures every review follows the same criteria. A shared testing skill ensures tests follow the same patterns and conventions. A shared deployment skill ensures pre-deploy checks are consistent.

Skills in .github/skills/ are just files in your repo. They go through the same PR review process as any other code change. When someone improves a skill, the whole team benefits on the next pull.

Where do I find skills for Copilot?

The Agensi marketplace has 200+ skills across 8 categories, all compatible with GitHub Copilot. Every skill passes an 8-point security scan before publication.

You can also install any Agensi skill with a one-liner curl command that fetches and unpacks the archive into Copilot's skills directory in a single step.

---

For the VS Code-specific setup, see How to Use SKILL.md in VS Code. For a comparison of Copilot's approach vs other agents, see GitHub Copilot Skills vs SKILL.md.