$ skilldirectory
— zsh — registry.skilldirectory.dev
skill libraryblogguidesabout

home / blog / claude skills

What Are Claude Skills? How They Work, With Examples (2026)

Quick answer: A Claude skill is a folder with a SKILL.md file that teaches Claude how to do one task the same way every time.

Claude reads a short description of every installed skill, and loads the full instructions only when your request matches.

Claude skills are reusable instruction packages for Claude. Instead of pasting the same long prompt into every chat, you save the method once and Claude applies it whenever it is relevant.

This guide explains what a skill is, how it works, where it runs and when it is worth using.

Claude skills at a glance

QuestionAnswer
What is a skill?A folder containing a SKILL.md file, and optionally scripts and reference files
What does it contain?A name, a description, and Markdown instructions
How does Claude find it?By reading the name and description of every installed skill
When is it loaded?Only when your request matches the description
Where does it run?Claude.ai, Claude Code, the Claude API and the Agent SDK
What does it cost?Skills are free to write. Running them uses your normal Claude usage

Anthropic announced skills on October 16, 2025, and they follow an open format called Agent Skills.

A simple example

Imagine you always want blog posts written a certain way: a quick-answer box, short paragraphs and a FAQ. Without a skill, you explain that every time.

With a skill, you write the rules once. When you later say "draft a post about X", Claude notices the skill matches and follows your rules.

How Claude skills work

Skills load in three stages, which Anthropic calls progressive disclosure. This is why you can install many skills without filling Claude's memory.

LevelWhen it loadsApproximate costWhat it holds
1. MetadataAlways, at startupAbout 100 tokens per skillThe name and description
2. InstructionsWhen your request matchesUnder 5,000 tokensThe body of SKILL.md
3. ResourcesOnly when neededEffectively unlimitedExtra files and scripts

A skill you never use costs almost nothing. A skill you do use costs only what Claude actually reads.

What is inside a skill

The only required file is SKILL.md. It starts with a short block of settings, then plain Markdown instructions.

my-skill/
  SKILL.md        required: name, description, instructions
  references/     optional: detailed docs loaded on demand
  scripts/        optional: code Claude can run
  assets/         optional: templates and other files

Two settings are required.

  • name: lowercase letters, numbers and hyphens, up to 64 characters.
  • description: what the skill does and when to use it, up to 1,024 characters.

The description matters most. Claude uses it to decide whether the skill applies, so it should say both what the skill does and when to use it.

Where Claude skills run

Skills work across Claude's products, but each one stores them separately.

WhereHow you add a skill
Claude CodePut the folder in ~/.claude/skills/ for yourself or .claude/skills/ in a project
Claude.aiZip the folder and upload it in Claude's settings
Claude APIUpload through the Skills API and reference it in your request
Agent SDKLoad skills from the filesystem

The catch: custom skills do not sync between surfaces. A skill you upload to Claude.ai is not available in the API or Claude Code, so install it in each place you want it.

Skills vs prompts, projects and MCP

Skills sit between a one-off prompt and a full integration.

  • Prompt: one conversation, and you retype it each time.
  • Skill: saved, reusable and loaded only when relevant.
  • Project: a shared workspace with files and standing instructions for a topic.
  • MCP server: a connection that gives Claude access to outside tools and data.

Skills teach Claude how to do something, and MCP gives it access to things. They work well together, because a skill can tell Claude how to use an MCP tool well.

What are skills good for?

Skills fit tasks you repeat and want done consistently.

  • Writing to a standard: reports, emails or blog posts in your format.
  • Working with files: creating and editing Word documents and spreadsheets.
  • Following a process: debugging, code review or test-first development.
  • Domain know-how: SEO audits, marketing plans or data analysis routines.

They are less useful for one-off questions, because there is nothing to reuse.

Examples from our directory

These skills show the range. Each page links to the source and licence.

See the best Claude skills or browse the skill library.

Are skills safe?

A skill is instructions, and sometimes code, so treat it like software from a stranger. Read it before you install it, and be cautious with scripts.

Our safety checklist for agent skills lists what to look for.

How to get started

  1. Pick one task you repeat every week.
  2. Find a skill for it in the library, or write your own.
  3. Install it where you use Claude.
  4. Run the task and see whether Claude follows the skill.

Steps are in our guide on how to install an agent skill. To understand the file format, read our guide to SKILL.md.

Frequently asked questions

What are Claude skills?

They are folders of instructions that teach Claude a repeatable task. Claude loads them automatically when your request matches their description.

Are Claude skills the same as prompts?

No. A prompt is typed each time. A skill is saved and loaded only when it is relevant.

Are Claude skills free?

Writing and using skills is free. Running them counts toward your normal Claude usage, and some surfaces need a paid plan.

Do Claude skills work in Claude Code?

Yes. Claude Code reads skills from a skills folder in your home directory or project. It loads them automatically based on their descriptions.

What is the difference between skills and MCP?

Skills hold instructions for how to do a task. MCP connects Claude to outside tools and data.

Methodology and disclosure

This guide draws on Anthropic's published documentation on Agent Skills and independent write-ups, checked in November 2026. Product details change often, so confirm against Anthropic's documentation.

We have no affiliation with Anthropic, and this article contains no affiliate links.