Claude Skills: how reusable skills work in Claude

AISkills.guru is independent and is not affiliated with Anthropic. Always confirm specifics against Anthropic's official documentation.

Claude was an early home for the Agent Skills format. Skills let you give Claude and Claude Code reusable expertise that loads on demand, instead of repeating long instructions in every conversation. This guide explains how skills fit across Claude products and how to think about them in practice.

For the underlying format, see What are AI Agent Skills? and the SKILL.md guide.

Claude, Claude Code, and the API

"Claude" spans several surfaces, and skills apply differently to each:

• Claude — the assistant product, where skills can extend what Claude knows how to do across conversations.
• Claude Code — Anthropic's agentic coding tool that reads your codebase, edits files and runs commands. Skills here are especially powerful for codifying team conventions and workflows. See the Claude Code skills docs.
• Claude API — where developers build their own agents; skills can be loaded into a custom agent's workflow where the implementation supports them.

The common thread is the same open skill format: a folder with a SKILL.md describing what the skill does and when to use it.

Custom skills

A custom Claude Skill is a folder you create containing a SKILL.md. The skill's description tells Claude when it is relevant; the body holds the instructions Claude follows once activated. Because activation is driven by the description, a precise description is the single biggest lever on whether a skill is used.

Skill structure

Claude Skills use the standard Agent Skills structure:

my-claude-skill/
├── SKILL.md
├── scripts/
├── references/
└── assets/

SKILL.md carries the metadata and instructions; scripts, references and assets are optional extensions. This is the same structure covered in the SKILL.md guide, so a skill written for Claude can often be reused in other compatible agents.

Activation

Claude uses progressive disclosure: it loads each skill's name and description at startup, then reads the full instructions only when a task matches. That keeps context lean even when many skills are available. The practical implication is the same everywhere — write descriptions that clearly state when to activate.

Scripts and resources

A Claude Skill can include scripts Claude runs (for example a linter or generator) and reference files Claude reads selectively. Scripts run with real capabilities, so apply the same scrutiny you would to any code; the security guide walks through what to inspect.

Where skills fit alongside other Claude features

Skills complement — rather than replace — other capabilities. Where tools and MCP connect Claude to external systems, skills package the know-how for using them well. You might, for instance, have a skill that documents exactly how your team uses a particular MCP tool.

Getting started

The fastest path is to write a small skill for a workflow you repeat often, then refine its description until it activates reliably. The creation guide covers this end to end.

Frequently asked questions

Are Claude Skills the same as Agent Skills? Claude Skills use the open Agent Skills format. "Claude Skill" is the product-specific name for the same underlying concept.

Do I need the Claude API to use skills? No. Skills are usable directly in Claude products such as Claude Code; the API is relevant when you build your own agents.

Can I reuse a Claude Skill elsewhere? In many cases yes, because the format is open and portable. Check that the skill doesn't depend on Claude-specific behaviour.

Where are skills stored? Locally, as folders you control and version — there is no required central registry, though directories and registries are emerging.