Claude Skills Explained: What They Are and Where They Run

A Claude Skill is a folder of instructions, scripts and resources that Claude loads when a task matches the skill's description. Each skill has a SKILL.md file. Anthropic's documentation says skills work in claude.ai, the Claude API and Claude Code, but custom skills do not sync between those places. Plan availability is described differently on different Anthropic pages.

How this article was made

We fetched and read the Anthropic pages listed in the sources on 2026-10-03. We did not test skills, and Anavem has not audited any skill. Passages are labeled "What the sources say" (facts, each with its source) and "What to consider" (our assessment). Product pages change, so treat each statement as dated 2026-10-03.

What the sources say: what a skill is

Anthropic's skills overview defines skills as directories containing instructions, scripts and resources that Claude dynamically loads to handle specific tasks. Each skill has a SKILL.md file that defines when it should be activated and what instructions Claude should follow.

The platform documentation (Agent Skills overview) describes three levels of loading:

1. Metadata, always loaded. The name and description from the file's header. Until a skill is triggered, only these occupy Claude's context.
2. Instructions, loaded when triggered. The body of SKILL.md.
3. Resources and code, loaded as needed. Extra markdown files, reference material and scripts. For scripts, the docs say only the output enters context, not the script code.

The platform page lists the required header fields. The name has a maximum of 64 characters and may contain only lowercase letters, numbers and hyphens. The description must be non-empty, at most 1,024 characters, and say both what the skill does and when to use it.

Types of skills

The skills overview lists four sources: Anthropic skills (pre-built skills for creating Excel, Word, PowerPoint and PDF documents), partner skills (built to work with partners' MCP connectors), organization-provisioned skills (deployed by an Owner on a Team or Enterprise plan) and custom skills you create. The platform page names the pre-built skills pptx, xlsx, docx and pdf.

An open standard

The skills overview says skills follow the Agent Skills specification, an open standard, so a skill written for Claude also works in other tools that adopt it. The agentskills.io site says the format was originally developed by Anthropic and has been adopted by a growing number of agent products. We did not test other products.

What the sources say: where skills run

The platform documentation describes three surfaces. The table repeats its statements and nothing else.

Surface	Availability and sharing	Network access
claude.ai	Pre-built skills and custom skills. Custom skills are uploaded as zip files. They are individual to each user, not shared organization-wide, and not centrally managed by admins.	"Varying": depending on user and admin settings, full, partial or none.
Claude API	Pre-built and custom skills. The page says custom skills are shared workspace-wide, and that using skills needs the code execution tool.	None. The page also says no runtime package installation.
Claude Code	Custom skills, stored in ~/.claude/skills/ (personal) or .claude/skills/ (project). Plugins can also carry skills. The pre-built document skills are not available there; the page notes that an open-source Claude API skill comes bundled with Claude Code.	Full: the same network access as any other program on your computer.

Source: platform.claude.com, Agent Skills overview, "Limitations and constraints," read 2026-10-03. The Claude Code row also draws on the Claude Code skills page.

The same page states that custom skills do not sync across surfaces. A skill uploaded to claude.ai must be uploaded separately to the API, and Claude Code skills are filesystem-based and separate from both.

What the sources say: turning a skill on in claude.ai

The skills overview says all your skills are listed at Customize > Skills in claude.ai and the desktop app. The "Your skills" tab groups them as Created by you, From your organization, Shared with you, and From Anthropic & Partners. You select Turn on, then describe your task and Claude loads a matching skill. You can also type "/" in the message box and pick one.

The same page says skills run in Claude's code sandbox, so "Code execution and file creation" must be on. On Team and Enterprise plans an Owner turns it on in organization settings.

Menu names differ between pages. The platform documentation says custom skills are uploaded under Settings > Features. The claude.com skills page uses Customize > Skills. The Help Center article Use skills in Claude also says Customize > Skills. Check your own interface.

What the sources say: which plans include skills

Anthropic's pages do not agree, so we report each one without choosing.

• The claude.com skills overview says skills are available on Pro, Max, Team and Enterprise plans.
• The platform documentation says custom skills in claude.ai are available on Pro, Max, Team and Enterprise plans with code execution enabled.
• The Help Center article What are skills? says skills are available for users on Free, Pro, Max, Team and Enterprise plans. The article Use skills in Claude says the same.
• The pricing page lists "Skills" with "Yes" in the Free column of its plan comparison table.
• The README of Anthropic's skills repository says its example skills "are all already available to paid plans in Claude.ai."

What to consider: the pages differ in age and audience, and we cannot tell which is current. Check Customize > Skills in your own account and the pricing page on the day you decide. Do not buy a plan for skills on the strength of this article.

What the sources say: skills, connectors, plugins and projects

According to Anthropic's skills overview: skills are task-specific procedures that Claude loads when a request matches. Plugins are packages containing several skills together with connectors, commands and agents. Projects hold background knowledge that is always loaded in that project's chats. MCP connectors let Claude reach external services and data. The page Create custom skills adds that a skill can tell Claude how to use a connector, but cannot reach the service itself. For a map across vendors, see the terminology map. For connectors, see what is an MCP server.

What the sources say: examples and licences

Anthropic's skills repository on GitHub says its skills are provided for demonstration and educational purposes. It says many skills in the repository are open source under Apache 2.0, while the docx, pdf, pptx and xlsx skills are source-available, not open source. Anavem lists skills under Claude skill listings, and has not audited them.

The platform page also says Agent Skills is not covered by zero data retention arrangements. Skill definitions and execution data are retained under Anthropic's standard data retention policy.

What to consider

• A skill is the right shape when you give Claude the same instructions repeatedly. Anthropic's guide says a one-off task belongs in the conversation, and instructions for every conversation belong in preferences or project instructions.
• Because custom skills do not sync, a team that uses claude.ai, the API and Claude Code has three places to upload and update. Plan for that.
• A skill can carry scripts. Read what a skill contains before you turn it on. See are Claude Skills safe.

Limitations, and when a skill is not the right choice

• Plan availability is unresolved in Anthropic's own pages, as shown above.
• Surfaces and menus are changing. The pages we read mention features (organization sharing, scanning, plugins) that vary by plan and by organization settings.
• A skill does not give Claude access to a live service. Anthropic's guide says that is what a connector provides.
• If you cannot read a skill's files or do not trust their source, the right choice may be not to turn it on. Anthropic's skills page says skills that others share with you, or that you upload yourself, are not reviewed by Anthropic.
• Developers who want Claude Code details should read the Claude Code profile and Anthropic's own docs.