SkillMD documentation

SkillMD is the open, safety-reviewed registry of Agent Skills. A skill is one SKILL.md file that any compatible agent loads on demand. Learn the format, install skills from the CLI, and wire the registry into your agent over MCP.

Get started Read the format spec Browse skills

What is an Agent Skill?

An Agent Skill is a single SKILL.md file: YAML frontmatter naming the skill and describing when it applies, followed by markdown instructions. An agent keeps only the description in context and loads the body on demand, the moment a task matches. That is what lets an agent carry hundreds of skills without spending context on any of them until one is needed.

The format is the one Anthropic documents for Agent Skills, and it is deliberately not vendor-specific: the same file is read by Cursor, OpenAI Codex and 60+ other agents. Skills sit alongside, not instead of, the Model Context Protocol: a skill is knowledge about how to do something, an MCP server is a capability the agent can call. Most useful setups have both. The longer version lives in what is an Agent Skill, and the field-by-field rules in the format spec.

Guides

Start

Getting started

Install the CLI and add your first skill, or scaffold and publish one of your own.

Reference

SKILL.md format

The frontmatter fields, body rules, lint checks, and quality score a skill is measured against.

Reference

CLI reference

Every skillmd command and flag: lint, scan, search, add, publish, and more.

Integrate

MCP server

Connect the registry to Claude, Cursor, and other MCP clients. Seven tools over stdio.

How to install a skill

Install the CLI, find a verified skill, and drop it into your agent's skills directory:

# install the CLI
npm i -g skillmds@latest

# find a skill, then add it to your agent
skillmd search "pdf extraction" --verified
skillmd add anthropic/docx -a claude-code

Install the CLI. Run npm i -g skillmds@latest. The CLI is the one dependency; the skills themselves are plain files with no runtime.
Find a skill. Run skillmd search "pdf extraction" --verified, or browse the registry on the web. Every result names its owner, its safety verdict and the capabilities it declares.
Add it to your agent. Run skillmd add <owner>/<name>. With no -a flag the CLI writes the skill into every agent directory it finds on the machine; -a claude-code pins one. Restart the agent to pick it up.

No -a flag? The CLI installs to whichever agent directories it detects, defaulting to Claude Code. See the CLI reference.

Per-agent guides

Each agent keeps skills in its own directory. These guides are plain text, so an agent can read them directly over HTTP:

Claude Code

.claude/skills

Cursor

.cursor/skills

Codex

.codex/skills

Windsurf

.windsurf/skills

OpenCode

.opencode/skills

All agents

Index of every guide.

For machines

Building an agent that reads SkillMD directly? Two entry points are designed for you:

/llms.txt: a compact, linkable map of the registry and its API.
https://api.skillmd.com/v1/search?q=<query>: JSON search that returns a raw_url per result, a direct fetch of the installable SKILL.md.

Frequently asked questions

What is an Agent Skill?

An Agent Skill is a single SKILL.md file: YAML frontmatter naming the skill and describing when it applies, followed by markdown instructions. The agent reads only the description until a task matches, then loads the full body. That on-demand loading is the whole point — an agent can have hundreds of skills installed without spending context on any of them until one is needed.

How do I install a skill?

Run npm i -g skillmds@latest, then skillmd add <owner>/<name>. The CLI detects every agent installed on the machine and writes the skill into each one's skills directory. You can also copy any SKILL.md by hand into your agent's skills folder — the format is plain text with no build step.

Which agents can use skills from SkillMD?

Any agent that reads SKILL.md files. That includes Claude Code, Claude.ai, Cursor, OpenAI Codex, Windsurf, OpenCode and 60+ others; each one's skills directory is listed at /agents. The format is not vendor-specific, so one file works everywhere.

Are skills on SkillMD safe to install?

Every skill passes an automated safety review before it becomes publicly visible, and the verdict is shown on its page along with the capabilities it declares (network access, filesystem writes, shell commands). Skills are still instructions an agent will follow, so read the SKILL.md before installing one — every page shows the full file and offers a markdown twin at /skills/<owner>/<name>.md.

Do I need an account to use SkillMD?

No. Searching, reading and installing skills are free and need no account. You only sign in to save skills, publish your own, or manage a plugin.

How is a skill different from an MCP server?

A skill is knowledge: instructions an agent loads into context to do a task a particular way. An MCP server is capability: a running process exposing tools the agent can call. They compose — a skill often tells the agent how and when to use an MCP server's tools. SkillMD itself ships both: a registry of skills, and an MCP server for searching it.