Claude Skills: What They Are and How to Install Them
A skill is a folder of instructions Claude loads only when it needs them. Here is where it lives, how to install it, and how it differs from MCP.
A Claude skill is a folder containing a SKILL.md file with instructions, and optionally scripts and reference files. Claude reads the description, decides the skill is relevant, and loads the rest only then. Claude Code skills work the same way and are invoked with /skill-name. Skills follow the Agent Skills open standard, so the same file works in other AI tools too.
Claude skills vs MCP, plugins and projects
These four get mixed up constantly, and picking the wrong one is why people end up installing something that never fires. The difference is not what they contain. It is when Claude loads them, and what job they do.
| What it gives Claude | When it loads | Use it for | |
|---|---|---|---|
| Skill | A procedure — how to do a task | On demand, when relevant | Repeatable work: a report format, a review checklist, a house style |
| MCP | Access to an external tool or data source | When the tool is called | Reaching Jira, Slack, a database, a file store |
| Plugin | A bundle: skills, agents, hooks, MCP servers | When the plugin is enabled | Shipping a whole workflow as one installable unit |
| Project | Static background knowledge | Always, in every chat in that project | Context that must never be missing |
- MCP gives Claude the tools. A skill teaches Claude how to use them. They stack.
- A project is always loaded. A skill is loaded only when it matches the task.
- Custom instructions apply to every conversation. A skill applies to one kind of work.
Where Claude Code skills live
In Claude Code, a skill is a directory on disk. Where you put that directory decides who can use it. There are four locations, and they resolve in a fixed order when two skills share a name.
| Location | Path | Applies to |
|---|---|---|
| Enterprise | Set in managed settings | Everyone in the organisation |
| Personal | ~/.claude/skills/<skill-name>/SKILL.md |
All your projects |
| Project | .claude/skills/<skill-name>/SKILL.md |
That project only |
| Plugin | <plugin>/skills/<skill-name>/SKILL.md |
Wherever the plugin is enabled |
When two skills share a name, enterprise wins over personal, and personal wins over project. Plugin skills are namespaced as /plugin-name:skill-name, so they never collide. Claude Code watches these folders and picks up an edit inside the session — no restart.
The mistake almost everyone makes
Cowork and cloud sessions do not read ~/.claude/skills/ on your machine.
They load the skills enabled for your claude.ai account, synced when the session starts. A skill that exists only in your local folder will come back as “skill not found” there. To use it in Cowork, enable it for your account instead.
How to install a Claude skill
Three surfaces, three different paths. Pick the one you actually use. Each takes under two minutes and none of them require writing code.
Claude (claude.ai and desktop)
- Open Customize in your account, then Skills.
- Click + and either upload your own
SKILL.mdfolder or click Browse skills to open the directory. - Toggle the skill on.
- Start a chat and describe the task. Claude loads the skill when it matches.
Requires code execution to be enabled on your account. Available on Free, Pro, Max, Team and Enterprise.
Claude Code
- Create the folder:
~/.claude/skills/my-skill/for every project, or.claude/skills/my-skill/for one repo. - Drop
SKILL.mdinside it. - Type
/my-skillin your session, or just describe the task and let Claude pick it up.
No restart needed. Claude Code watches those folders and picks up changes live.
Cowork
- Enable the skill for your claude.ai account — Customize in the Desktop sidebar, or skills settings on claude.ai.
- Start a Cowork session. Your account skills sync at session start.
- Invoke it by name, or let Claude choose it.
Cloud sessions additionally load project skills committed to the repository’s .claude/skills/ folder.
What is actually inside a SKILL.md file
A skill is Markdown with a small YAML header. That is the whole format. If you can write a checklist in a text editor, you can write a skill.
--- name: quarterly-report description: Builds our quarterly revenue report from a raw CSV export when_to_use: When someone asks for the quarterly report or hands over a revenue export allowed-tools: Read Grep --- ## Steps 1. Read the CSV and check the column headers match the expected schema. 2. Group revenue by product line, then by month. 3. Flag any month that moved more than 15% against the previous one. 4. Write the summary in the format below.
when_to_use.Everything below the header is just instructions in your own words. No SDK, no build step, no deployment.
Using Claude Cowork skills
Cowork runs Claude against your files and tasks rather than inside a chat window, and skills are how you tell it what “done properly” means for your team. The rule that trips people up: Cowork does not read the skills folder on your machine. Both interactive and scheduled Cowork sessions load the skills enabled for your claude.ai account, synced when the session starts.
So the workflow is: write the skill once, enable it for your account, and it follows you into every Cowork session and every scheduled run.
Anthropic Agent Skills: an open standard, not a lock-in
The Agent Skills specification is published as an open standard at agentskills.io, with a reference Python SDK for anyone implementing skills support in their own platform. Claude Code follows the standard and adds its own extensions on top — invocation control, subagent execution, dynamic context injection.
The practical consequence is the part worth reading twice: a skill you write today is a Markdown file governed by a public spec. If you move to a different AI tool that adopts the standard, the file moves with you.
When a skill does not fire
The skill never triggers
Nine times out of ten the description is the problem, not the instructions. Claude reads only the description to decide. Rewrite it to lead with the concrete use case, and add when_to_use with the phrases people actually type.
The skill triggers too often
The description is too broad. Narrow it to the specific situation, or set disable-model-invocation: true so it runs only when you type /skill-name yourself.
It works locally but not in Cowork or a scheduled run
It is in ~/.claude/skills/ and nowhere else. Those sessions start fresh and load your claude.ai account skills. Enable it for your account, or commit it to the repository’s .claude/skills/ for cloud sessions.
Claude skills: frequently asked questions
What is a Claude skill?
A folder containing a SKILL.md file with instructions, and optionally scripts and reference files. Claude reads the skill’s description, decides it is relevant to your request, and loads the full instructions only then. This is called progressive disclosure, and it is why having many skills installed does not fill up the context window.
Where do Claude Code skills live?
Personal skills go in ~/.claude/skills/<skill-name>/SKILL.md and apply to every project. Project skills go in .claude/skills/<skill-name>/SKILL.md inside the repository and apply to that project only. Plugins carry skills at <plugin>/skills/<skill-name>/SKILL.md, and enterprise skills are set through managed settings.
What is the difference between Claude skills and MCP?
MCP connects Claude to external services and data sources — it provides the tools. A skill provides procedural knowledge: the instructions for how to do a task properly. They are complementary, not alternatives. An MCP connector gives Claude access to your Jira; a skill tells Claude how your team writes a ticket.
How do I install a Claude skill?
On claude.ai, open Customize, then Skills, click + and upload the skill or browse the directory. In Claude Code, drop the skill folder into ~/.claude/skills/ or the project’s .claude/skills/ and invoke it with /skill-name. In Cowork, enable the skill for your claude.ai account and it syncs at session start.
Do Claude skills work in Cowork?
Yes, but not from your local folder. Cowork sessions, interactive and scheduled alike, load the skills enabled for your claude.ai account rather than ~/.claude/skills/ on your machine. Enable the skill for your account from Customize in the Desktop sidebar or from the skills settings on claude.ai.
Do I need to know how to code to write a skill?
No. A skill is a Markdown file with a short YAML header — name, description, and your instructions in plain language. Scripts are optional and only needed for advanced behaviour. If you can write a checklist, you can write a skill.
Are Claude skills free?
The feature is available on Free, Pro, Max, Team and Enterprise plans, and requires code execution to be enabled. Anthropic’s own skills, such as document creation for Word, Excel and PowerPoint, are available to all users. Skills you buy or download are priced by whoever made them.
Is it safe to install a skill someone else wrote?
A skill is instructions Claude will follow, so read it the way you would read a script before running it. On Enterprise plans, organisations can turn on skill and plugin scanning to check uploads for malicious content. For everything else, the rule is the usual one: install from sources you can identify.
What are Anthropic Agent Skills?
Agent Skills is the open specification behind the format, published at agentskills.io with a reference Python SDK. It means a skill is not tied to Claude — any platform that adopts the standard can read the same SKILL.md file. Anthropic also publishes its own maintained skills, which Claude invokes automatically when relevant.
What is the difference between a skill and a Claude project?
A project provides static background knowledge that is loaded into every chat inside it. A skill provides a procedure that activates only when the task matches, and it works everywhere rather than inside one project. Use a project for context that must always be present, and a skill for work you repeat.
Don’t want to write one?
1,000+ skills, already written and tested.
Role-specific skill files for Claude, ChatGPT and any AI chat — bookkeeper, SEO strategist, medical scribe, platform engineer. Written and tested by us, not scraped off GitHub. Download, drop in the folder, done. One-time price, no subscription.