How to Use Claude Skills: A Step-by-Step Tutorial

To use Claude Skills, enable code execution in your settings, then either let Claude trigger a skill automatically or invoke one by name. On claude.ai that means turning on "Code execution and file creation" under Settings, then opening Customize > Skills. In Claude Code a skill is just a folder you drop in place. This tutorial walks all three surfaces in order.

Skills are folders of instructions and optional scripts that teach Claude a repeatable task, and Claude loads them only when a job needs them. If you want the concept first, read what are Claude Skills. This page is the hands-on version: switch them on, run a pre-built one, add your own, and know how to invoke and remove them. Details below follow Anthropic's help center and Claude Code docs as of August 2026, and menu labels shift often in this product, so treat exact paths as current-not-permanent.

What do you need before you start?

One prerequisite covers every surface: code execution has to be on. Skills run in a sandbox where Claude can read files and run scripts, so without that environment there's nothing for a skill to execute against.

Per Anthropic's help center, Skills are available on the Free, Pro, Max, Team, and Enterprise plans, and the feature requires code execution to be enabled. Claude Code and the API also support skills, described as beta for API users through the code execution tool. So the checklist is short:

• A Claude account on any current plan (Free included, as of August 2026).
• Code execution and file creation enabled.
• On Team or Enterprise, an owner who has switched Skills on for the organization.

That last point trips up people on managed plans. If you don't see Skills at all, the setting is gated above you, not missing.

How do you turn on Skills in claude.ai?

Three steps, and the first one is the setting people miss.

1. Open Settings, go to Capabilities, and enable Code execution and file creation. Skills won't appear or fire without it.
2. Open Customize > Skills. You'll see the built-in document skills listed there.
3. On a Team or Enterprise plan, an owner enables both "Code execution and file creation" and "Skills" under Organization settings > Skills first. Members then reach their skills under Customize.

That's the whole setup. Once code execution is on, Anthropic's built-in skills work with no further action: "With Code execution and file creation on, Claude will automatically use these tools when relevant." You don't invoke them, you just ask.

How do you use a pre-built Skill?

The fastest way to see Skills work is to use one you didn't write. Anthropic ships built-in skills for Excel, Word, PowerPoint, and PDF files, maintained by Anthropic and used automatically.

Ask claude.ai something that matches one, for example:

Build me a Q3 sales spreadsheet with monthly columns, a totals row, and a bar chart.

Claude recognizes the request as a spreadsheet job, loads the Excel skill on its own, runs the underlying code, and hands back a real .xlsx file instead of a wall of text describing one. The same pattern produces a .pptx deck, a formatted .docx, or a filled PDF. There's no button to press and no skill to pick. That automatic matching is the point: Claude reads each skill's one-line description at startup and pulls in the full instructions only when your request fits. For a walkthrough of the document skills specifically, see Claude Skills for Excel and PowerPoint.

You can also be explicit. Naming the skill or the outcome ("use my brand-deck skill to draft the client presentation") nudges Claude toward the right one when a request could match several.

How do you install a Skill on claude.ai?

Built-in skills cover documents. For anything else, you add a skill folder yourself. The community and Anthropic both publish ready-made ones, and the upload flow is the same regardless of source.

1. Get the skill as a folder containing a SKILL.md file, then package that folder as a ZIP.
2. Open Customize > Skills.
3. Click +, choose Upload a skill, and upload the ZIP.

The skill now shows in your list, and Claude treats its description like any built-in one: it loads when your request matches. On Team and Enterprise plans, owners can provision skills org-wide from Organization settings, and members can share a skill with specific people, groups, or the whole organization once sharing toggles are on.

One caution before you upload anything. A skill can tell Claude to run code and call tools, so an untrusted skill is a genuine risk. Read the section on safety below before you install one from a stranger.

How do you use Skills in Claude Code?

Claude Code is the loosest surface: a skill is a folder on disk, no upload step. Where you put the folder decides who can use it.

To install a skill you were handed, drop its folder into one of those paths. Personal skills follow you everywhere; project skills travel with the repo, so committing .claude/skills/ shares a skill with everyone who clones it. After adding one, run /skills to confirm it loaded, or type / and look for it in the autocomplete menu. Claude Code watches those directories and picks up new or edited skills within the session, so you usually don't restart.

Invoking works two ways, the same as claude.ai plus a direct command:

• Automatic: ask something that matches the skill's description, and Claude loads it.
• Direct: type /<folder-name>. A skill at .claude/skills/deploy/SKILL.md runs as /deploy.

Claude Code also ships bundled skills you can use immediately, invoked the same way. /code-review reviews a diff, /doctor runs a setup checkup, and /run and /verify launch your app to confirm a change works. Type / to browse them.

Writing your own custom skill in Claude Code is a short exercise: make a directory, add a SKILL.md with a description, and Claude can trigger it. The full anatomy, including frontmatter fields and bundled scripts, is its own topic covered in Claude Skills vs Agents vs MCP, which draws the lines between the three so you don't reach for the wrong one.

How do you use Skills through the Claude API?

On the API, skills run inside the code execution tool. You reference a skill by its ID in a request, or upload custom skills so they're available across your workspace, and Claude invokes them during a run the same way it does elsewhere. Anthropic describes API skills as available in beta as of August 2026. Two constraints matter for API skills specifically: they run with no internet access and can't install packages at runtime, so a skill that depends on fetching a remote file or a fresh library won't behave the way it does on your laptop. Plan the skill's bundled files accordingly.

How do you know a Skill actually fired?

A quiet failure mode: you enabled a skill, asked for the thing, and got a generic answer. A few checks tell you what happened.

• On claude.ai, a triggered built-in skill returns a real file (an .xlsx, a .pptx) rather than a text description. If you get prose instead of a file, code execution is probably off.
• In Claude Code, run /skills to see everything loaded and where it came from, including bundled skills and any synced from your claude.ai account. If your skill isn't in the list, the folder is in the wrong place or the SKILL.md frontmatter is malformed.
• Description too vague? Claude decides whether to load a skill from its description alone. If a skill never fires on its own, the fix is almost always a sharper description that says both what it does and when to use it, not the skill body.

How do you manage and remove Skills?

Turning a skill off is as direct as adding one.

• claude.ai: remove or disable the skill from your list under Customize > Skills.
• Claude Code personal or project skill: delete its folder (~/.claude/skills/<name>/ or .claude/skills/<name>/). Claude Code drops it from the session.
• Claude Code bundled skill: these stay available by default; you disable them through settings if you need to.
• Keep a skill but stop Claude from auto-running it: set disable-model-invocation: true in the skill's frontmatter, so only you can invoke it with /name. Handy for anything with side effects, like a deploy skill you don't want Claude firing because the code looked ready.

Because a skill in Claude Code is just files, version control is your management layer: a project skill lives in the repo, gets reviewed in pull requests, and updates when you edit SKILL.md.

Are the Skills you install safe to run?

Only when you trust the source, and this is the one step worth slowing down for. A skill hands Claude new instructions and code. In Anthropic's own words, "a malicious Skill can direct Claude to invoke tools or execute code in ways that don't match the Skill's stated purpose," which in practice means data exfiltration or unauthorized access depending on what Claude can reach.

Treat installing a skill like installing software:

• Prefer skills you wrote or got from Anthropic. Audit anything from an unknown source before it runs.
• Read every file in the bundle (SKILL.md, scripts, resources) for behavior that doesn't match the stated job: surprise network calls, odd file access, tool use unrelated to the task.
• Watch skills that fetch content from external URLs. That's the highest-risk category, because fetched content can carry its own instructions.

None of this is exotic. It's the standard supply-chain problem applied to agent instructions, and the mitigation is boring: know where your skills come from.

Where should you go from here?

You now have the mechanics: enable code execution, let built-in skills fire automatically, upload or drop in the ones you want, and invoke or remove them by hand. The natural next move is to stop hunting for skills and start collecting good ones, then encode your own repeated instructions.

For a curated starting set, see the best Claude Skills roundup, or the Claude Skills for marketing and SEO picks if that's your lane. When you're ready to build instead of install, the record a Claude Skill approach lets you teach Claude by screen recording, no code required. The moment you catch yourself pasting the same instructions into Claude for the third time, that's the signal to turn them into a skill.