RunLocalModel.com

What Are Agent Skills? A Beginner's Guide

By the RunLocalModel editorial team · October 10, 2026

Reference: https://support.claude.com/en/articles/12512176-what-are-skills

Short version An agent skill is a folder of instructions, and sometimes scripts or templates, that an assistant opens for one kind of task. The folder holds a repeatable procedure: a status-email format, a brand layout, the way your team files a ticket. The assistant reads a short description of every skill you have, and opens the full instructions only when the task matches.

This site's Agent Skills browser lists public skills you can read. This page is the short explanation of what those folders are, written for someone who has never used one. The product details below follow Anthropic's help article What are skills?.

The Monday email

Every Monday you ask for a team status email. You want three sections, the real project names, and no ticket numbers the assistant invented. You can paste those rules into the chat. Next Monday the chat is new, and the rules are gone.

A skill is those rules, saved once. The next time you ask for a status email, Claude can open the skill and follow it. The same idea covers a brand guide for slides, a meeting-note layout, or the way your team creates an issue in Jira, Asana, or Linear.

A simple skill is Markdown, the same kind of text as a README. Anyone who can write instructions can write one. Add a script when the job needs one, for example to fill a spreadsheet or check a file.

What goes in the folder

The Agent Skills specification names the instruction file SKILL.md. Claude's help center describes the same file. It starts with two required fields, a name and a description, and then the steps. Claude reads the description first, so say what the skill does and when to use it.

---
name: weekly-status
description: Write the weekly status email in the team format. Use when asked for a status update, weekly recap, or standup summary.
---

# Weekly status

Write a short email with these sections, in order:

1. Shipped this week
2. In progress
3. Blocked

Stay under 200 words. Use only the project names and ticket numbers given in the request.

The description is what Claude compares with your request. The steps under it stay unread until that description matches. Extra files are optional. A brand skill might add a color sheet. A data skill might add a script. Claude opens those files when the instructions call for them.

How Claude chooses a skill

Anthropic calls this progressive disclosure. The point is practical: a long library of procedures should not crowd out the task in front of you.

  1. Claude reads the name and description of each available skill.
  2. When a description matches the task, Claude opens that skill's instructions.
  3. When the instructions point at a template or a script, Claude opens that file.

Ask to turn meeting notes into the weekly status email, and the weekly-status skill is the one that matches. A skill for slide layouts stays closed. Ask for a branded deck and a status email in the same breath, and Claude can open both. Skills do not name each other. Claude still picks every skill that fits the request and can use them together.

That is where the consistency comes from. The procedure is written down, and it is available in a new chat, and to a teammate who has the same skill, whenever the task comes up.

Where you can use them

Anthropic says skills are available on the Free, Pro, Max, Team, and Enterprise plans. They need code execution enabled, because a skill can create files and run scripts. With that setting off, skills do not run.

Skills are also available in Claude Code. On the API, they are in beta for people using the code execution tool.

The folder format is an open standard. Other assistants that adopt it can use the same folder. A reference Python SDK is available for developers adding skill support to their own products. The catalog on this site collects public SKILL.md files used with assistants such as Claude Code, Codex, and Cursor.

Four places a skill can come from

Kind Who looks after it What you should expect
Anthropic skills Anthropic Included for every user. Claude uses them on its own when you ask for an Excel, Word, PowerPoint, or PDF file.
Custom skills You, or someone on your team A workflow you wrote and uploaded. Colleagues get the file after you share it.
Organization skills A Team or Enterprise owner Listed for every member. The owner chooses whether each skill starts enabled or disabled.
Partner skills Companies such as Notion, Figma, and Atlassian Listed in the skills directory, and written to work with that company's connector.

Custom skills are the ones most people end up writing. A few jobs they fit:

On Team and Enterprise plans, an owner can hand an approved skill to the whole organization. It shows up in every member's list, so a new person does not have to upload it. The owner can also leave it in the list and start it turned off. Anthropic documents the admin steps in Provision and manage skills for your organization.

Partner skills sit next to connectors. A Notion connector lets Claude reach Notion. A Notion skill describes how a page in your workspace should be written. Use them together: the connector is the access, and the skill is the procedure.

On the Enterprise plan, an organization can turn on skill and plugin scanning to check uploaded skills and plugins for malicious content.

Skills, projects, connectors, and custom instructions

Claude already has a few places to store guidance. Each one answers a different question.

Feature What you put there When Claude uses it
Project Background for one body of work: a client, a codebase, a research folder In every chat you start inside that project
Skill The steps for one kind of task When that task comes up, in any chat
Custom instructions Preferences for how Claude should treat you in general, such as tone or language Across your conversations
Connector (MCP) A link to an outside service or data source When Claude calls that service

A project is the background for one effort. A skill is a procedure you want followed wherever you are working. Custom instructions are the standing preferences. A connector, through the Model Context Protocol, is the link to a product. The skill is how your team wants that product used.

How to look at them in Claude

  1. Open Customize in your account, then Skills. Anthropic's document skills are already there.
  2. Click the plus button, then Browse skills, to open the directory.
  3. Add a skill when you want Claude to follow that workflow. It starts enabled. You can turn it off later and leave it in the list.

To see how other people write these files, open a listing in the Agent Skills browser. Read the source, including any scripts, before you use a skill you did not write. RunLocalModel lists the files. It does not install them, and it does not certify them.

The API overview of the same feature is Agent Skills in the Claude docs.

Questions people ask first

Common questions
Do I have to write code to make a skill?
A simple skill is a Markdown file with a name, a short description, and the steps. Add a script when the job needs one, such as filling a spreadsheet or checking a file.
Will all of my skills be loaded into the chat?
Claude reads the name and description of each available skill, then opens the instructions for the skills that fit the task. The other skills stay closed.
Do I need to install Anthropic's spreadsheet and document skills?
Those skills are already available. When you ask for an Excel, Word, PowerPoint, or PDF file, Claude can use them on its own.
What is a project for, if I also have skills?
A project stores background for one body of work, and that background is present in every chat you start there. A skill stores the steps for one kind of task, and Claude can use those steps in any chat when the task comes up.
When do I use a connector, and when do I use a skill?
A connector links Claude to an outside service, such as Notion or Jira. A skill stores the procedure for the work. The connector lets Claude reach the service, and the skill tells it to follow your team's way of doing the job.
Can my company give every employee the same skill?
On Team and Enterprise plans, an owner can provision a skill for the organization. It appears in each person's skill list, and the owner chooses whether it starts enabled or disabled.
Will the same folder work outside Claude?
The Agent Skills format is an open standard. A folder written to that standard can run in other assistants that support it. This site's catalog includes public skills used with Claude Code, Codex, and Cursor.

Related on this site