When you ask Claude to repeat a specialized task, such as formatting a report or applying brand rules, a Skill gives it reusable instructions and related resources for that work. Skills are folders that Claude loads when a task makes them relevant. They help Claude handle repeatable work consistently, without making you paste the same instructions into every chat.
You can use built-in Skills in Claude, create custom ones, or add them to Claude Code. Skills require code execution to be enabled. Setup depends on where you use Claude. In each case, choose or build a focused set of instructions, then ask Claude to do the task.
What a Skill contains
A Skill is a folder of instructions, scripts, and supporting resources. Its SKILL.md file contains YAML metadata, including a name and description, followed by instructions written in Markdown. The folder can also include reference materials or scripts for the work.
Claude uses progressive disclosure: it finds Skills that fit a request and loads their relevant information as needed. This means Claude does not load a Skill’s full contents into every conversation. Anthropic’s overview of Skills describes this structure and how Claude uses it.
Anthropic provides built-in Skills for creating documents such as Excel spreadsheets, Word files, PowerPoint presentations, and PDFs. Claude invokes those Skills when they fit the task. You can also build a custom Skill for work such as applying a company’s writing rules or following a repeatable reporting procedure.
Find and use Skills in Claude
In Claude.ai, open Customize in your account and go to Skills to find available Skills. Anthropic says its document-creation Skills are available to all users and run automatically when relevant. For a task, describe the result you want in your prompt, such as asking Claude to format a document according to a particular set of instructions.
Skills require code execution to be enabled. Check Claude’s current settings if a Skill does not appear or run. The setup details and available controls can vary by Claude product, so use the controls for the product where you work.
For Claude Code, you can install Skills as plugins through a marketplace or place them in a .claude/skills directory for automatic discovery. Anthropic’s Skills repository provides examples and implementation details.
Create a focused custom Skill
Start with a task you repeat and define the result Claude should produce. For example, a brand-writing Skill could tell Claude how to review a draft against a company style guide. Put the actual steps in SKILL.md, and keep supporting material, such as the style guide, in a separate reference file. Keeping the style guide separate makes the instructions easier to update.
Keep each Skill narrow. One Skill that handles a specific procedure is easier to trigger and test than one that tries to cover unrelated work. Try it with several realistic requests, then refine its description and instructions if Claude misses it or applies it when it should not. For Claude.ai, use the current Skills controls to add or manage a custom Skill; for Claude Code, use the supported plugin or directory method.
Choose the right Claude feature
Use Skills for task-specific instructions that Claude loads when relevant. Other Claude features serve different purposes:
- Custom instructions set preferences that apply broadly.
- Projects provide background information for work in a particular space.
- Model Context Protocol (MCP) connects Claude to external services and their data.
For example, store a company’s background information in a Project, use custom instructions for a general preference such as concise answers, and use a Skill for the steps Claude should follow when preparing a recurring report. Use MCP when Claude needs a connection to an external service.
Check access and security
Skills are available on Claude’s Free, Pro, Max, Team, and Enterprise plans, according to Anthropic’s current Skills documentation. The documentation also covers Skills for Claude Code and a beta for API users who use the code execution tool. For Team and Enterprise plans, organization Owners can provision Skills for users; Enterprise also offers Skill scanning for malicious content.
For the API, the implementation described by Agno’s Anthropic Skills documentation supports up to eight Skills per request. That documentation also says API Skills run in an isolated Anthropic container without network access or runtime package installation. Files generated there need to be downloaded through the Files API. API availability also depends on the hosting option: the cited documentation lists Claude’s API, Claude Platform on AWS, and Microsoft Foundry deployments hosted by Anthropic, but not Amazon Bedrock or Google Cloud Vertex AI.
Before enabling a Skill, review its instructions and scripts, especially if someone else created it. Do not put sensitive information in a Skill. If Claude produces an unexpected result, check whether the Skill’s description is too broad or its instructions overlap with another Skill. Then test the revised Skill again.