Skip to content
Agent skills

Claude Agent Skills

Anthropic’s official public repository of reusable skills, examples and patterns for teaching an agent specialized workflows.

Skills

Who it’s for. People learning the official skill shape before they write a company skill. The example files stay in the repository.

What it is

A useful starting point—not another unexplained bookmark.

Agent Skills package instructions, references and supporting files into reusable capabilities. Anthropic’s repository is where those example folders, the specification and the template live. This page explains the idea. The files stay on GitHub.

Repository

Open the skill files on GitHub.

The examples, the Agent Skills specification and the starter template are in Anthropic’s repository. Use that link when you want the folders. They are not copied onto this page.

Useful for

When to use it

  • Teaching an agent a repeatable company workflow
  • Packaging domain instructions and reference files
  • Studying how official skills are organized

What you will find

Inside the resource

  • Skill folders and SKILL.md files
  • Reusable workflow instructions
  • Supporting references and scripts
  • Examples for document and development tasks

Quick start

How I would use it

  1. 01

    Pick one repeated job

    Choose a task with a clear trigger and result, such as reviewing a document or preparing a report.

  2. 02

    Study a similar skill

    Look at its name, description, instructions and supporting files before designing your own.

  3. 03

    Write a narrow first version

    Explain when the skill applies, the required workflow and how the agent should verify completion.

  4. 04

    Test with real requests

    Check that the skill activates at the right time and does not take over unrelated tasks.

Ready?

Continue to GitHub · Anthropic

Open original resource ↗

Keep exploring

Where to go next

Next step

Want to use one of these in a real product?

Tell me the user problem. I will help you choose the resource and the safest first integration.