A prompt helps with one conversation. An agent skill helps every time. A skill is a small package of instructions (and sometimes scripts or templates) that teaches an AI agent how to do a particular job the way you want it done: reviewing code, writing release notes, testing a web app, turning a transcript into meeting notes.
Skills started with Claude and are now read by a growing number of AI agents and coding tools, including GitHub Copilot. The catalogue has more than 1,500 of them.
How a skill works
A skill is a folder with a file called SKILL.md in it. The file starts with a short header that gives the skill a name and describes when to use it, followed by the instructions themselves:
---
name: meeting-notes
description: Turns a meeting transcript into short notes with decisions and
action items. Use when the user shares a transcript or asks for minutes.
---
# Meeting notes
1. Read the whole transcript before writing anything.
2. List decisions first, then action items with an owner and a due date.
3. Keep the notes under 300 words. Quote numbers exactly.
4. End with any open questions that nobody answered.
The clever part is how the agent uses it. At the start it only reads each skill’s name and description, which takes very little space. When a task matches a description, it loads the full instructions, along with any scripts or reference files in the folder. That means an agent can have dozens of skills installed without being slowed down or confused by them.
Installing a skill
On every skill’s page in the catalogue, Download saves the file as SKILL.md. Where you put it depends on your tool:
- Claude Code: create a folder named after the skill and save the file inside it, either in your project at
.claude/skills/<skill-name>/SKILL.md(shared with everyone working on the project) or in~/.claude/skills/<skill-name>/SKILL.md(available in all your projects). - Claude apps: put the folder in a ZIP file and upload it where your settings list skills.
- Other tools: check your tool’s documentation for its skills folder. Many coding agents now look for the same
SKILL.mdformat.
Then just ask for the job in plain language. You do not need to name the skill: the agent picks it up when your request matches its description.
Skills, custom agents and rules files: what is the difference?
The catalogue has three kinds of files for AI agents, and they do different jobs:
| What it is | When it is used | |
|---|---|---|
| Rules file | Standing instructions about a project: its stack, conventions and do’s and don’ts | Always, in every session in that project |
| Agent skill | A packaged know-how for one kind of task, sometimes with scripts | Only when a task matches the skill’s description |
| Custom agent | A separate specialist with its own instructions and tools, for example a security reviewer | When the main agent (or you) hands it a task |
Custom agents are often called subagents. In Claude Code they are Markdown files in .claude/agents/; in GitHub Copilot they are .agent.md files in .github/agents/. Because each one works in its own context, they are good for focused reviews and research that would otherwise clutter your main conversation. Examples in the catalogue include the SEO Analyzer, the Web Accessibility Checker and the Next.js Architecture Expert. For rules files, see our guide to AGENTS.md, CLAUDE.md, Cursor rules and Copilot instructions.
Read before you install
A skill is instructions your agent will follow, and some skills include scripts that run on your computer. Treat them like any software you download:
- read the whole
SKILL.mdbefore installing it, and any scripts that come with it; - check which tools and commands it asks the agent to use;
- be wary of skills that send data to outside services or ask for keys and passwords;
- install only what you need, and remove skills you stop using.
We only include skills from openly licensed collections and leave out ones built for harmful purposes, but you know your own systems best.
Skills worth a look
- Web Application Testing Skill: checks a local web app in a real browser with Playwright, including screenshots.
- karpathy-guidelines: habits that reduce common AI coding mistakes, such as over-complicating changes.
- Prompt Refiner: turns a rough request into a tight, well-structured prompt.
- claude-md-master: creates and maintains a project’s CLAUDE.md file from what is really in the repository.
- Second Opinion: asks other AI coding tools to review Claude Code’s work.
Browse all of them in Agent skills or the Agent Skills collection, and filter by the tool you use.