How Claude skills work
Anthropic introduced skills in October 2025 to solve a simple problem: Claude is a capable generalist, but most real work follows a specific process. Your brand voice, your deploy checklist, the way your team formats a report. Instead of pasting that context into every chat, you package it once as a skill.
Skills load in three layers, which is why you can install dozens without slowing Claude down:
- Name and description. At the start of a session Claude sees only a line or two per skill, enough to know what each one is for.
- The SKILL.md body. When a request matches a skill's description, Claude reads its full instructions.
- Bundled files. Scripts, templates and reference docs inside the folder are opened or run only if the instructions call for them.
You don't have to call a skill by name. Ask for “a slide deck from this spreadsheet” and Claude picks up a presentation skill on its own. In Claude Code you can also run a skill directly as a slash command.
What's inside a skill
A skill is just a folder. The only required file is SKILL.md: YAML frontmatter with a name and description, followed by Markdown instructions. Here is a complete, minimal skill:
--- name: release-notes description: Writes release notes from merged pull requests. Use when the user asks for a changelog, release notes or "what shipped". --- # Release notes 1. List the pull requests merged since the last tag. 2. Group them under Features, Fixes and Internal. 3. Write one plain sentence per change, from the user's point of view. 4. Link each line to its pull request.
Bigger skills add a scripts/ folder for code Claude can run, a references/ folder for detailed docs, and templates or assets. Anthropic's own skills for Word, Excel, PowerPoint and PDF files work this way, and they're open source.
Where Claude skills work
- Claude Code. Put skill folders in
~/.claude/skills/for every project or.claude/skills/for one repo, or install them as part of a plugin. - The Claude app. Upload a skill as a ZIP under Customize → Skills. Custom skills need a paid plan with code execution turned on.
- The Claude API and Agent SDK. Developers can attach skills to requests and to agents they build.
- Other agents. The SKILL.md format is an open standard, so many skills also work in tools like Cursor and OpenAI Codex.
Claude skills vs MCP, CLAUDE.md and prompts
Skills are often confused with MCP servers. The short version: MCP gives Claude access to tools; skills teach Claude how to use them well. They work best together.
| Tool | What it is | Best for |
|---|---|---|
| Skills | Packaged know-how: instructions, scripts and templates, loaded only when a task needs them | Repeatable procedures, house styles, file formats, checklists |
| MCP servers | Connections that let Claude call outside tools and read live data | Acting in other apps: GitHub, Slack, databases, browsers |
| CLAUDE.md / project instructions | Notes Claude reads at the start of every session in a project | Conventions that apply to every task, like build commands |
| Prompts | Instructions you type or paste each time | One-off requests |
How to install a Claude skill
There are three ways, and every skill page on StealIt shows the exact steps:
- One command.
npx skills add owner/repo --skill namedownloads a skill into the right folder for Claude Code and other agents. - A ZIP file. Download the skill, then unzip it into your skills folder or upload it to the Claude app.
- A Claude Code plugin. Some publishers ship a whole suite of skills, agents and scripts as a plugin you add with
/plugin install.
The install guide covers each method step by step.
How to write your own Claude skill
- Pick one job you explain to Claude over and over.
- Create a folder named after it and add a
SKILL.md. Use lowercase letters, numbers and hyphens for the name. - Write the description carefully. Say what the skill does and when to use it, including the words people actually type, because that's what Claude matches against.
- Write the steps as you'd brief a sharp new colleague: concrete, ordered and short.
- Move long reference material into separate files, and repeatable work into scripts.
- Test it on real requests and tighten the description until it triggers reliably.
Anthropic's skill-creator skill automates most of this: it interviews you, drafts the SKILL.md and helps you test it.
Are Claude skills safe?
A skill that contains only Markdown can't do anything on its own. Skills with scripts can run code on your machine, so treat them like any other software you install: prefer publishers you trust, read the SKILL.md and skim the scripts first. Every skill on StealIt shows whether it includes scripts and links straight to its source on GitHub.