Claude Skills Guide

A skill is a folder that teaches Claude one specific task. It holds a SKILL.md file with instructions, plus optional scripts, templates, and reference files.

A skill can do what a saved prompt cannot:

• Load on its own when relevant
• Include scripts for Claude to run
• Bundle templates and examples
• Hold long reference material back until needed
• Be shared across a team
• Run in claude.ai, Claude Code, and the API

Think of a skill as an expert procedure: write it once, reuse it many times.

claude.ai and the apps

Available on the Free, Pro, Max, Team, and Enterprise plans.

Claude Code

Supports personal, project, plugin, and enterprise skill folders.

Claude API

Through the Skills API.

Skills first shipped in October 2025.

Skills stay light through progressive disclosure:

• Claude always sees every skill’s name and description
• The full instructions load only when the skill runs
• Supporting files load only when something references them

So you can install many skills without filling Claude’s context.

The description decides when a skill runs. Write it with care.

Free, Pro, Max

Go to Settings, then Capabilities, and turn on Code execution and file creation.

Team, Enterprise

An admin turns on Cloud code execution and file creation and Skills under Organization settings, then Plugins & skills.

Anthropic’s built-in skills for Word, Excel, PowerPoint, and PDF then kick in on their own when relevant.

Strong candidates are tasks that are:

• Repeated often
• Done the same way every time
• Too complex for one prompt
• Reliant on templates or examples
• Better handled by a script
• Shared across a team

Examples: weekly reports, on-brand posts, contract checklists, code reviews, and invoice processing.

Every skill needs a SKILL.md file that opens with a name and a description, followed by the instructions.

--- name: weekly-report description: Build our weekly client report from the metrics spreadsheet. Use when asked for a weekly report or a client update. --- 1. Read the latest metrics file. 2. Summarize the five biggest changes. 3. Follow the template in template.md. 4. Flag every metric that dropped more than 10%.

The description tells Claude when to use the skill. Cover:

• What the skill does
• When to use it
• Words people would naturally use
• What it must not be used for

Make it too vague and the skill never fires. Make it too broad and it fires when it should not.

• Number the steps
• State the goal before the steps
• State the inputs and outputs
• Add one or two examples
• Name the common mistakes to avoid
• Explain how to check the result

Keep SKILL.md tight. Put long reference material in separate files and link to them.

Templates

Documents or formats Claude fills in

Examples

Strong outputs for Claude to imitate

References

Policies, style guides, or data to consult

Scripts

Code for steps that must run exactly

Use scripts for anything that needs precision. Use instructions for anything that needs judgment.

Anthropic’s skill-creator skill helps you build new skills: it asks questions, drafts the files, and tests the result.

Prompt:

Use the skill-creator skill to turn this workflow into a skill. Ask me whatever you need to know, draft SKILL.md and any supporting files, then test the skill on one example.

• Zip the skill folder into one ZIP file
• Open Customize, then Skills
• Select +, then + Create skill
• Choose Upload a skill and pick the ZIP

On Team and Enterprise plans, admins can provision skills for the whole organization.

Personal

~/.claude/skills/<name>/SKILL.md, for all your projects

Project

.claude/skills/<name>/SKILL.md, for a single repository

Plugin

Inside an installed plugin, run as /plugin-name:skill-name

Run a skill yourself with /skill-name, or let Claude pick it up when the description matches.

Default

You and Claude can both run it.

disable-model-invocation: true

Only you can run it. Best for actions such as deploys.

user-invocable: false

Only Claude uses it, as quiet background knowledge.

Add allowed-tools to pre-approve specific tools while the skill runs.

In Claude Code, add Anthropic’s example skills by running:

/plugin marketplace add anthropics/skills

Many tool makers publish official skills too, such as Remotion and HyperFrames for video.

Read every third-party skill before you install it. Skills can run code.

• Run the skill on three real examples
• Confirm it fires when expected
• Confirm it stays quiet when it should
• Compare the output with your best manual work
• Fix the instructions, not only the output

If a skill fires too often, make its description more specific.

• Commit project skills to the repository
• Bundle related skills into one plugin
• Provision organization-wide skills on Team or Enterprise
• Version each skill and log its changes
• Assign one owner to each skill

• Never store passwords or API keys in a skill
• Review every script before you run it
• Install skills only from trusted sources
• Restrict tools with allowed-tools where possible
• Require a human to approve risky actions

A skill is exactly as powerful as the instructions and code inside it.

Choose

Pick a valuable workflow you repeat.

Describe

Write a precise description.

Instruct

Write clear, numbered steps.

Equip

Attach templates, examples, and scripts.

Test

Try it on real examples.

Refine

Tighten triggers and instructions.

Share

Hand it to your team.

Help me turn [workflow] into a Claude skill. First, interview me about the goal, the inputs, the steps, the quality bar, and common mistakes. Then draft SKILL.md with a precise name and a description that states exactly when to use it. Move long reference material into separate files, and use scripts only for steps that must run exactly. Last, test the skill on one example and list anything that needs my decision.