What Are Claude Skills? A Plain-English Guide (2026)

TL;DR. A Claude Skill is a reusable folder of instructions, scripts, and reference files that teaches Claude to do one task well — and that Claude loads only when it's relevant. Skills are composable, portable across every Claude product, and you can build one by just describing it. They're the building block; plugins bundle many skills into a ready-to-install package for a profession.

If you've heard people talk about "Claude Skills" (Anthropic's formal name is Agent Skills) and weren't sure whether it's a feature, a product, or developer jargon — here's the plain-English version. A Skill is how you give Claude specialized, repeatable expertise without re-explaining it every time.

What is a Claude Skill?

A Skill is a folder that contains everything Claude needs to perform a specific task: written instructions, optional scripts (real code Claude can run), and any reference files. At its core is a SKILL.md file that describes what the skill does and when to use it.

The clever part is progressive disclosure: Claude scans the skills available to it and loads a skill's full contents only when the task in front of it actually calls for that skill. As Anthropic puts it, Claude "will only access a skill when it's relevant to the task at hand," and it "only loads what's needed, when it's needed." So you can have dozens of skills installed and none of them slow Claude down until the moment one is useful.

A simple analogy: a Skill is like a trained colleague's playbook for one job. It sits on the shelf until the job comes up — then Claude opens it and follows it exactly.

The four things that make Skills useful

Anthropic describes Skills as:

• Composable — multiple skills work together automatically. Claude can combine a "pull the data" skill with a "write the report" skill in one task.
• Portable — the same skill format works across every Claude product. Build it once, use it in the chat apps, in Claude Code, in the API, and in Cowork.
• Efficient — progressive disclosure means skills don't tax your context window until they're invoked.
• Powerful — a skill can include executable code, so the work that should be deterministic (math, formatting, file manipulation) runs as code instead of being re-derived by the model each time.

Where Claude Skills work

Skills are an open standard, so they run wherever Claude runs:

Surface	How skills show up	Who it's for
Claude apps (claude.ai / desktop)	Available to Pro, Max, Team, Enterprise users	Everyone on a paid plan
Claude Cowork	Delivered through plugins	Non-technical knowledge workers
Claude Code	Installed to ~/.claude/skills or via plugins	Developers
Developer Platform API	Via the Messages API + the /v1/skills endpoint (Code Execution beta)	Builders integrating Claude

The free Claude plan doesn't include Skills. On Team and Enterprise, an admin enables Skills for the organization before anyone can use them.

Skills vs. Projects vs. Plugins — the part everyone confuses

These three names get mixed up constantly. Here's the clean distinction:

• A Skill is one reusable capability — one task done well (e.g., "draft a discharge summary from de-identified notes").
• A Project is a persistent workspace — custom instructions plus uploaded knowledge for an ongoing body of work. It's about context, not a single task.
• A Plugin is a bundle — it packages a set of Skills (and the connectors they need) for an entire profession, so you install one thing and Claude instantly knows how to do many related tasks the right way.

In short: Skills are the building block, Plugins are the package, and Projects are the workspace you run them in.

How to create a Claude Skill

You don't need to write code or hand-edit files. The easiest path is the built-in skill-creator skill: you describe the task you want Claude to learn in plain language, and it generates the folder structure and SKILL.md for you — Anthropic's words: "No manual file editing required."

Developers who want more control can author the folder by hand and:

• drop it in ~/.claude/skills/ for Claude Code, or
• upload it through the API to use it programmatically.

Because skills are just folders, teams share them through version control — commit the skill to a repo and everyone gets the same expertise.

Where to find Claude Skills

• Anthropic's open directory of partner-built skills, and the official anthropics/skills repository on GitHub for ready-made examples.
• Profession-specific bundles: The AI Career Lab's free plugins library packages the skills for a given job — a social media manager, a real estate agent, a bookkeeper, and dozens more — so you install one plugin and get a coworker that already knows the work.

Who should care about Skills?

If you do the same kind of task repeatedly — the same report, the same note, the same review — a Skill turns that repeated expertise into something Claude does consistently, every time, without you re-explaining it. That's the whole point: you teach Claude the job once, and it stops being a blank slate.

For most professionals, the fastest way to benefit isn't building skills from scratch — it's installing a plugin that bundles the skills for your role. See what Claude Cowork is for how those plugins run as a no-code desktop coworker, or browse the plugins library to find your profession.

FAQ

What are Claude Skills in simple terms?

A folder of instructions and scripts that teaches Claude to do one specific task well, which Claude loads only when it's relevant. Anthropic's formal name is Agent Skills.

Do I need to code to use Claude Skills?

No. You can install ready-made skills (or a plugin that bundles them) with no code, and even create a skill by describing it to the built-in skill-creator. Writing the folder by hand is an option for developers, not a requirement.

Are Claude Skills the same as Claude Cowork?

No. Cowork is the desktop agent that does the work; Skills are the specialized know-how Cowork (and the other Claude surfaces) use to do specific tasks well. In Cowork, skills arrive bundled inside plugins.

Sources

• Anthropic: Introducing Agent Skills
• Anthropic: Agent Skills — open directory
• GitHub: anthropics/skills