Claude skills: what they are, how they work in Claude and Claude Code, and the free ones worth installing

Quick picks

What are Claude skills?

A Claude skill is a saved way of doing one job. Instead of pasting the same long instructions into every chat, you put them in a folder once, and Claude uses them whenever your request fits. Anthropic’s help centre describes skills as folders of instructions, scripts and resources that Claude loads dynamically for specialised tasks.

A skill folder looks like this. Only SKILL.md is required:

meeting-notes/
├── SKILL.md          required: name, description, instructions
├── references/       optional: longer docs Claude opens when needed
├── scripts/          optional: code Claude can run
└── assets/           optional: templates and other files

And this is a complete, working SKILL.md. The block between the --- lines is the frontmatter; everything under it is ordinary Markdown.

---
name: meeting-notes
description: Turns a pasted meeting transcript into decisions,
  owners and deadlines. For meeting notes and call transcripts.
---

# Meeting notes

1. List every decision, one line each.
2. List action items as: owner - task - due date.
   If no date was said, write "no date". Never guess.
3. End with the questions nobody answered.

Two things make a skill different from a saved prompt. Claude decides by itself when to load it, based on the description, so you don’t paste anything. And a skill can carry files (examples, templates, scripts) that a prompt can’t. The format is published as an open standard at agentskills.io, which is why the same folder also works outside Claude.

How Claude skills work

Claude loads a skill in three stages:

1. Discovery. At the start of a session Claude sees only each skill’s name and description.
2. Activation. When your request matches a description, Claude reads that skill’s full SKILL.md.
3. Execution. It follows the steps, and opens reference files or runs bundled scripts only when the steps call for them.

Because only the short descriptions stay in view, installed skills cost little until they are used. Simon Willison put it at “a few dozen extra tokens” per skill when he wrote about the launch. The flip side: the description decides whether a skill ever gets used. Claude Code’s docs say to put the key use case first, because the description is cut at 1,536 characters in the skill list.

Where Claude skills work

The same folder works in every Claude surface and in other tools that read the standard. Follow each tool’s own guide:

In the Claude app, skills need code execution turned on. Plan availability changes; the help centre has the current list.

How to install a skill in the Claude app

1. Open the skill’s page here and download the folder (the ⋯ menu), or copy it from the GitHub link in the credit line.
2. Make sure the folder is named after the skill and contains SKILL.md, then zip the folder.
3. Check that code execution is on in settings.
4. Go to Customize → Skills, press +, choose Create skill → Upload a skill, and pick the ZIP.
5. Start a new chat and describe the job in plain words. You don’t have to name the skill.

Zip the whole folder, not only SKILL.md. Many skills point to other files next to it, and without them the steps break halfway. The step-by-step install guide has a screenshot of every screen.

Claude Code skills: install, call and share them

In Claude Code, a skill is a folder on disk. Where you put it decides where it loads:

Every skill page on AgentAlley has a one-line command that copies the full folder, extra files included. For example, Systematic debugging:

npx degit obra/superpowers/skills/systematic-debugging#main ~/.claude/skills/systematic-debugging

Claude uses the skill when a request matches, or you call it by typing / and the folder name, for example /systematic-debugging. Two frontmatter fields are worth knowing: disable-model-invocation: true means only you can start the skill, and allowed-tools pre-approves the tools it may use. Claude Code also ships its own built-in skills, such as /code-review and /debug.

Skills on your machine don’t follow you into Cowork or cloud sessions. Those load the skills enabled on your claude.ai account, plus project skills committed to the repository.

Best Claude skills, picked by job

Not a ranking by stars. Stars belong to a whole repository, not to one skill in it. Every skill below does one clear job, carries a licence you can use, needs no account, and was read in full before it went on this list. Each reason comes from the skill’s own file. The command copies the folder into ~/.claude/skills/ for Claude Code; for the Claude app, zip the same folder.

Claude skills marketplace: where to find more

There is no single app store for skills. There are four kinds of places to find them:

• Anthropic’s own skills. The anthropics/skills repository works as a Claude Code plugin marketplace: run /plugin marketplace add anthropics/skills, then /plugin install example-skills@anthropic-agent-skills. Its example skills are Apache-2.0; the document skills (docx, pdf, pptx, xlsx) are source-available, not open source.
• Claude Code plugin marketplaces. The official one, claude-plugins-official, is added automatically; browse it with /plugin or at claude.com/plugins. Plugins can bundle skills with hooks and MCP servers, and Anthropic’s docs warn they can run code with your permissions.
• GitHub lists. Awesome-lists link to hundreds of repositories. You still open each repo to check the licence and files.
• Directories like this one. AgentAlley shows the whole SKILL.md on the page with the author, licence and a link to the original file, so you can read before you copy. Free, no sign-in.

Either way, the install comes down to a folder in the right place. See which repositories we collect from.

Browse Claude skills by job

Filed by the work you need done, not by tool name. The number is how many skills are on that shelf today.

What makes a good Claude skill

Four things the good ones in the list above share:

• One job. Copywriting writes page copy and hands email copy to a separate skill. A skill that tries to do everything gets loaded for the wrong requests.
• A description that says when, not how. Writing Skills warns that if the description summarises the steps, Claude may follow the summary and never read the skill.
• Detail kept in side files. SEO Audit sends Claude to references/international-seo.md for the full multilingual detail, so that part is read only when a site has several languages.
• Stated limits. Systematic Debugging stops after three failed fixes; Verification Before Completion bans “done” without evidence. Limits written into the file are limits Claude follows.

Are Claude skills safe?

A skill is mostly text, but it can include scripts, and Claude runs those with your permissions. Simon Willison’s caution applies: skills are only as safe as the environment that runs them. Before you install one from anywhere, check four things:

• Scripts. Any .py, .sh or .js file in the folder runs on your machine. Know what it does.
• Accounts and network calls. Some skills need a login or an API key and send your data to someone’s server. That is why the list above leaves out skills that need a third-party account before they work.
• Override lines. “Ignore previous instructions” or “don’t tell the user”. A real skill never needs these.
• Author and licence. A named author and a real licence are the minimum. No licence means you have no right to use it.

Every skill page on AgentAlley shows the entire file on the page, with no login and nothing cut. We read the files; we never run them. How listing works.

Skills vs prompts, MCP servers and subagents

They combine: a skill can tell Claude how to use an MCP tool well, and a skill like Requesting Code Review hands work to a subagent. The longer comparison: agent skills vs prompts vs MCP.

How we pick

• Skills are collected from public GitHub repositories and shown in full, with the author, licence and a link to the original file.
• Repositories that don’t state a licence are left out, however good the skill.
• Stars are shown as stars on the whole repository, never as a score for one skill.
• We never run a skill’s scripts. The list on this page was read line by line by a person, and every reason above points to a line in the file.

FAQ

Keep going