What are Claude Skills?

What is a Claude skill?

A Claude skill is a self-contained packet of expertise that Claude can call on when a task matches it. Think of it as a recipe Claude reads to do a specific kind of work the right way: produce a clean PDF, deploy an Azure resource, review a React component, write an internal status update.

A skill is just a directory with a SKILL.md file at its root. The file has YAML frontmatter at the top, then a markdown body that contains the actual guidance:

---
name: pdf
description: Use this skill whenever the user wants to do anything with PDF files. This includes reading or extracting text/tables from PDFs, combining or merging multiple PDFs into one, splitting PDFs apart, rotating pages...
license: Proprietary. LICENSE.txt has complete terms
---

# PDF Processing Guide

## Overview

This guide covers essential PDF processing operations using Python libraries and command-line tools...

The description is the trigger

The description field is not metadata, it is how Claude decides whether to use the skill. Most descriptions follow the pattern "Use this skill whenever..." or "Use when...", listing the situations where the skill applies. When Claude considers a task, it scans available skill descriptions and pulls in the matching ones.

That means a skill's description has to be precise. Too narrow and Claude misses opportunities to use it. Too broad and Claude pulls it into tasks where it does not help. The best descriptions read like a usage policy more than a tagline.

The body is the actual guide

The body is markdown that Claude reads before doing the work. It usually contains an overview, the steps, code snippets, and references to other files in the skill directory (REFERENCE.md, FORMS.md, etc.). This is where the real expertise lives, and why two skills with similar descriptions can produce very different results.

Skills vs MCP servers

Both extend Claude, but they are not the same thing:

• MCP servers give Claude new tools to call. A Slack MCP server adds the ability to send messages. A Postgres MCP server adds the ability to query a database. They are about new capabilities. See What is MCP? for the full picture.
• Skills give Claude know-how about using existing tools well. A skill does not add a new tool, it tells Claude how to produce a styled slide deck, structure a PRD, or fix a Kubernetes deployment using whatever is already available.

In practice you use both. An MCP server gives Claude the keys, a skill tells it how to drive.

Where Claude skills come from

Anthropic publishes a set of official skills at anthropics/skills covering common formats and workflows: PDF, DOCX, XLSX, presentations, brand guidelines, MCP server creation, and more.

Beyond the official set, a much larger community has emerged. Vendors like Vercel, Microsoft, and Google Workspace publish skills for their platforms. Individual developers publish skills for specific frameworks, languages, and workflows. The format is open, so anyone can ship one.

Finding skills

VaultPlane indexes them. The skills registry currently catalogs 12,000+ Claude skills sourced from public GitHub repositories, each with parsed frontmatter, a trust score, and a category. Browse by area, filter by platform, or search directly.

If you are deciding whether to add a skill to your environment, the trust score is the fastest signal. It factors in verification status, repository activity, frontmatter completeness, and invocation safety, so you do not have to read every SKILL.md to know which ones are worth running.

The shape of the ecosystem

The 12,000+ skills currently in the registry break down roughly like this:

• Code and developer tooling: ~5,200
• AI / ML and agent building: ~1,900
• Backend and APIs: ~1,100
• Frontend and web: ~1,100
• DevOps and cloud: ~700
• Everything else (security, data, testing, business workflows, design, docs): ~2,400

It is a fast-moving format. Six months ago Claude skills were mostly Anthropic's official set; today they are an ecosystem.

Getting started

Browse the VaultPlane skills registry to see what exists. Add a skill to your environment by cloning the source repository and placing the skill directory where Claude loads skills from. The skill activates automatically when your task matches its description.

To build your own, Anthropic's skill-creator walks through the whole process: drafting, evaluating, and benchmarking.