What are Claude skills? They are folders of instructions, scripts and resources, built around a SKILL.md file, that Claude loads only when a task needs them, so the same procedure can be reused without pasting it into every conversation (Claude Help Center).
Applies to: Claude Agent Skills, Claude docs and Help Center as of October 2026. Last checked: 2026-10-05.
On this page
What a skill is
Anthropic’s Help Center describes skills as “folders of instructions, scripts, and resources that Claude loads dynamically to improve performance on specialized tasks”, such as applying brand guidelines to documents or running a team’s own data analysis steps (What are skills?).
The developer docs call them Agent Skills: reusable, filesystem-based resources that load on demand, unlike a prompt, which only applies to one conversation (Agent Skills overview). Once a skill is available, Claude uses it automatically when a request matches it.
There are four kinds, per the Help Center (What are skills?):
- Anthropic skills: built and maintained by Anthropic, such as document creation for Excel, Word, PowerPoint and PDF files.
- Custom skills: ones you or your organization write for your own workflows.
- Organization provisioned skills: on Team and Enterprise plans, owners can provision skills that appear for every member.
- Partner skills: skills from partners such as Notion, Figma and Atlassian, built to work with their MCP connectors.
What is inside a skill
Every skill needs a SKILL.md file that starts with YAML frontmatter. This is the structure from Anthropic’s docs (Agent Skills overview):
---
name: your-skill-name
description: Brief description of what this Skill does and when to use it
---
# Your Skill Name
## Instructions
[Clear, step-by-step guidance for Claude to follow]
## Examples
[Concrete examples of using this Skill]
Only name and description are required. The docs set these limits (Agent Skills overview):
| Field | Rules |
|---|---|
name | Up to 64 characters; lowercase letters, numbers and hyphens only; no XML tags; cannot contain “anthropic” or “claude” |
description | Non-empty; up to 1024 characters; no XML tags; must say what the skill does and when to use it |
A skill folder can also hold extra Markdown files (for example FORMS.md or REFERENCE.md), scripts such as scripts/fill_form.py, and reference material like schemas or templates (Agent Skills overview).
How Claude loads a skill
Skills use what Anthropic calls progressive disclosure: Claude reads a skill in stages instead of putting all of it into context at the start (Agent Skills overview).
| Level | When it loads | Token cost (per the docs) | What it is |
|---|---|---|---|
| 1. Metadata | Always, at startup | About 100 tokens per skill | name and description from the frontmatter |
| 2. Instructions | When the skill is triggered | Under 5k tokens | The body of SKILL.md |
| 3. Resources and code | Only when needed | None until accessed | Bundled files; scripts run through bash and only their output enters context |
The description is what Claude matches a request against, which is why the docs say it must cover both what the skill does and when to use it. Until a skill is triggered, only its name and description take up context, so many skills can be installed at once (Agent Skills overview).
Where skills work
Skills run in a code execution environment with filesystem access, and their availability differs by product (Agent Skills overview):
- Claude apps: the Help Center lists skills for Free, Pro, Max, Team and Enterprise plans, with code execution turned on. You enable example skills and upload your own as a ZIP file in Customize > Skills (Use skills in Claude).
- Claude Code: custom skills are folders with a
SKILL.mdplaced in~/.claude/skills/(personal) or.claude/skills/(project). The pre-built document skills are not available in Claude Code. See the Claude Code skills guide for frontmatter options and invocation. - Claude API: pre-built skills are used by
skill_id(pptx,xlsx,docxorpdf) and custom skills are uploaded through the/v1/skillsendpoints; both need the code execution tool. API skills run in a sandboxed container with no network access and no runtime package installation.
Custom skills do not sync between these surfaces: a skill uploaded in the Claude apps has to be uploaded to the API separately, and Claude Code skills live on your filesystem apart from both (Agent Skills overview).
Skills compared with projects, MCP and custom instructions
The Help Center draws these lines (What are skills?):
- Projects give static background knowledge that is always loaded in that project’s chats; skills activate only when needed and work everywhere in Claude.
- MCP connects Claude to external services and data; skills hold the procedure for a task. The two can work together, with MCP providing the tools and a skill describing how to use them.
- Custom instructions apply to every conversation; skills are task-specific and load only when relevant.
The skill format is also published as an open standard at agentskills.io, so the same skill can work in other tools that adopt it (What are skills?).
Where to get skills
Anthropic’s public anthropics/skills repository holds example skills plus the document skills (docx, pdf, pptx, xlsx). Its README says the document skills are source-available rather than open source, and that the skills are provided for demonstration and educational purposes. In Claude Code, the README gives these commands to add the repository as a plugin marketplace and install a set:
/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills
/plugin install example-skills@anthropic-agent-skills
Individual entries such as the PDF skill and Skill Creator are listed in the skills directory, and the Claude skills marketplace guide covers where else to find them.
Warning
Anthropic advises using skills only from trusted sources: ones you wrote or got from Anthropic. A malicious skill can direct Claude to invoke tools or run code in ways that do not match its stated purpose, which could lead to data exfiltration or unauthorized system access. If you must use a skill from an unknown source, review every bundled file (
SKILL.md, scripts, images and other resources) and treat it like installing software (Agent Skills overview).
On Enterprise plans, organizations can turn on skill scanning to check uploaded skills and plugins for malicious content (What are skills?).
