A Claude skill is a folder with a SKILL.md file that teaches Claude how to do a specific job. Claude always sees the skill's short description, and it reads the full instructions only when a task calls for them or when you type /skill-name.
How do Claude skills work?
Each skill has two parts: a short description and a body of instructions. The description is always in Claude's context, so Claude knows the skill exists. The body loads when the skill is used and then stays in context for the rest of the session. That is why a long skill costs almost nothing until you need it, and why Anthropic advises keeping SKILL.md under 500 lines.
A skill can start in two ways. Claude can load it on its own when your request matches the description, or you can call it yourself with /skill-name. An author can turn either route off in the skill's settings.
Skills follow the open Agent Skills standard, so many also work in other coding agents. Some fields, such as running a skill in a subagent, only work in Claude Code.
What is inside a SKILL.md file?
A short block of settings, then plain instructions in Markdown. Two settings matter most: name, which becomes the command, and description, which tells Claude when to use it. A skill folder can also hold reference files and scripts that Claude opens only when needed.
Where are Claude skills stored?
| Where | Path | Who gets it |
|---|---|---|
| Personal | ~/.claude/skills/name/SKILL.md | You, in all your projects |
| Project | .claude/skills/name/SKILL.md | Everyone who works in that repository |
| Plugin | plugin/skills/name/SKILL.md | Wherever the plugin is on, as /plugin:name |
When two skills share a name, an organization's managed skill beats a personal one, and a personal one beats a project one.
How do you install a Claude skill?
There are three common ways.
- With the skills command line tool:
npx skills add https://github.com/owner/repo --skill name. Every skill page in this directory shows its exact command. - As part of a plugin: run
/pluginin Claude Code and pick one from the Discover tab. - By hand: copy the skill's folder into
~/.claude/skills/. Claude Code picks up the change without a restart.
Can you use skills in the Claude app?
Yes. Skills work in the Claude app on every plan, from Free to Enterprise, as long as code execution is on. Turn on Code execution and file creation under Settings, Capabilities, then open Customize, Skills to switch skills on and off.
Anthropic’s Excel, Word, PowerPoint and PDF skills are built in, so Claude uses them on its own once code execution is on. To add another skill, zip its folder so the folder itself is at the top of the ZIP, with its SKILL.md inside, then choose Customize, Skills, +, Create skill, Upload a skill. The name can be up to 64 characters and the description up to 200. Team and Enterprise owners manage skills for everyone under Organization settings.
A skill that relies on programs installed on your own computer, such as a command-line tool, needs Claude Code, which runs on your machine. Source: Anthropic’s help pages on using skills in Claude and creating custom skills, checked November 8, 2026.
Claude skills vs MCP, plugins, commands and subagents
| Thing | What it is | Use it when |
|---|---|---|
| Skill | Instructions Claude loads when needed | You keep repeating the same process or advice |
| MCP server or connector | A link that gives Claude tools for an outside service | Claude needs to reach data or act in another system |
| Plugin | Skills, agents, hooks and MCP servers packaged as one install | You want a ready setup, or want to share yours |
| Custom command | Merged into skills: .claude/commands/deploy.md and a deploy skill both create /deploy | You have old commands; they keep working |
| Subagent | A separate Claude worker with its own context | A task should run apart from your conversation |
In short: a skill teaches Claude how, and a connector gives it access. They often work together.
Do skills slow Claude down or use up tokens?
Only a little until they run. The name and description of every skill Claude can call on its own count toward context on every turn, and Claude Code cuts each listing at 1,536 characters. The full text costs tokens from the moment it loads until the session ends. Installing dozens of skills you rarely use still adds up, so install for the work you actually do. Each page here estimates both costs.
Are Claude skills safe?
Not automatically. A skill is instructions plus sometimes scripts, and what it runs, runs as you. When Snyk studied 3,984 skills from two public marketplaces in February 2026, it confirmed 76 malicious ones and found critical issues in 13.4%. It also warned that popularity is no proxy for safety.
We saw the same thing while building this list: several repositories with fewer than 30 GitHub stars report over a million installs. Before you install a skill:
- Read its
SKILL.mdand any scripts it ships. - Be wary of anything that pipes a download into your shell.
- Check who publishes it, and whether the repository looks like a copy of a better-known one.
- Check its scan results, knowing that scanners disagree and miss things.
How do you make your own Claude skill?
Create a folder in ~/.claude/skills/ and add a SKILL.md with a name, a description that says when to use it, and your instructions. Make one when you keep pasting the same instructions into chat, or when part of your CLAUDE.md has turned into a procedure. Anthropic's skill-creator skill can walk you through it, and the official docs list every setting.
How do you find a good skill?
Start from the job, not the name. Pick what you want done and read the ranked list for it, or use the search on the directory's home page. We explain how the ranking works so you can disagree with it.