What Are Claude Skills? How They Work and How to Create Your Own (2026)

Claude Skill (https://claude.com/skills) is a reusable file of instructions that teaches Claude how to do a specific task. No code, no API, no server required. You write a text file describing what you want, save it, and Claude loads it whenever the task comes up.

Think of it as a standing brief for a colleague. Instead of repeating your writing style, format preferences, and quality standards in every conversation, you pack them into a Skill once. Claude picks it up automatically when the task matches.

Skills vs Prompts vs MCP Connectors

These three get mixed up a lot. Here's the short version:

A prompt is one conversation. You type it, Claude responds, it's gone. Next time you need the same thing, you type it again.
A Skill is a permanent instruction file. Write it once, it loads across conversations, projects, and months. It's a prompt that doesn't expire.
An MCP connector is a data pipe. It connects Claude to external systems like Google Drive, Slack, or GitHub. Skills tell Claude what to do with that data.

A practical example: a connector lets Claude read your Gmail. A Skill tells Claude how to triage those emails, which ones need replies today, which ones can wait, and what format the triage report should follow.

Why Skills Matter

Skills solve a real problem with AI assistants: context decay. Every new conversation with Claude starts from zero. You rebuild context by re-explaining your preferences, your project structure, your audience.

Skills break that cycle. When you create a Skill for "weekly reports," Claude knows your template path, your preferred tone, and the data sources to pull from, every time, without being told again. The instruction persists across sessions.

This is also why Skills are the engine behind Claude Cowork plugins. A Cowork plugin is essentially a bundle of Skills, MCP connectors, slash commands, and sub-agent configurations wrapped together for a specific job role. Individual Skills are the atomic unit inside those bundles.

What Goes Into a Skill

A Skill file is plain text, typically saved as skill-name.md in Claude's skills directory. At minimum, it contains:

A name and description. What the Skill does, when Claude should use it.
Instructions. Step-by-step or principle-based guidance on how to complete the task.
Examples. Sample inputs and expected outputs, optional but dramatically improve reliability.

A simple Skill for formatting meeting notes might look like this:


# Meeting Notes Skill

## When to use
When the user asks you to format or clean up meeting notes.

## Instructions
- Extract action items and tag owners at the top
- Group discussion points by topic with H3 headings
- Remove filler phrases ("I think," "maybe," "sort of")
- Keep decisions and deadlines in bold

## Output format
# Meeting: [Topic] , [Date]

## Action Items
- [Owner]: [Task] by [Deadline]

More advanced Skills include conditional logic ("if the user mentions X, do Y"), reference to project-specific templates, and integration with MCP connectors for data sources.

How Skills Load

Claude uses a relevance matching system to decide which Skills to load. When you send a message, Claude scans your Skill files and loads the ones that match the task. You don't manually activate them.

This means Skill naming and descriptions matter. If your Skill for quarterly reports is named qreport.md with no description, Claude might not recognize when to use it. Name it quarterly-report-skill.md and describe it as "used for generating quarterly business review reports from financial data," and Claude will load it reliably when you ask about quarterly reports.

Creating Your First Skill

The simplest way to create a Skill is through Claude Desktop settings:

Go to Settings → Skills → Add Skill
Give it a name that describes what it does
Write the instructions in the editor
Save it

Claude also supports "Record a Skill," a feature added in July 2026. Instead of writing instructions, you record your screen while you demonstrate a task and narrate what you're doing. Claude watches the recording and generates the Skill file automatically. This works well for repetitive computer tasks, sorting files, formatting spreadsheets, and pulling data from specific apps.

For Claude Code and Cowork, Skills are stored as .md files in a project-or-user-level skills directory. You can share them with your team, version-control them in Git, and install community Skills from repositories.

Skills Across Claude Products

Anthropic maintains official Skill repositories on GitHub for different domains. The community has built thousands of Skills covering everything from financial modeling to content writing to DevOps automation.

In enterprise environments, Skills are managed through the "Customize" panel, where admins control which Skills are available and enforce company-wide standards. Teams can set up private Skill marketplaces so everyone uses the same report templates, code review standards, and documentation formats.

As of mid-2026, Skills are supported across the Claude product line: claude.ai chat, Claude Code, Claude Cowork, and through the API when using the Agent SDK.