Claude Skills Explained

Claude Skills are folders of instructions that Claude loads on demand to do a specific job well. Anthropic launched them on October 16, 2025 as an open, portable format. At their core, a Skill is just a folder with one Markdown file inside — no special app or coding degree required.

This page explains what Claude Skills are and how they work, from the file structure to the clever way Claude decides which Skill to open. We keep it definitional and technical, so you understand the mechanism before you spend money on it.

You will learn the anatomy of a Skill, the idea of progressive disclosure, how to create a simple one, where Skills run, and how they differ from MCP, Custom GPTs, and Projects. Reviewed by Jonathan West, Founder of Layer3 Labs, on July 2, 2026. We research using primary vendor and regulator sources.

What Are Claude Skills?

A Claude Skill is a folder containing a SKILL.md file plus optional scripts and reference files that Claude loads when a task matches. The SKILL.md file holds plain instructions written in Markdown. The optional files can be Python scripts, templates, checklists, or reference documents Claude pulls in only when it needs them.

Anthropic describes Skills as onboarding materials that package expertise, turning Claude into a specialist on what matters to you. Think of one Skill as one saved playbook — a firm-standard way to draft an email, run a checklist, or fill a template.

The format is deliberately simple. Simon Willison, who reviewed Skills at launch, called them "conceptually extremely simple" — essentially instruction files with a few resources alongside them.

• A Skill is a folder, not an app or a plugin you have to compile
• The required piece is one file named SKILL.md written in plain Markdown
• Optional extras include scripts (.py), reference docs (.md), and data files
• One Skill captures one repeatable job — a template, a workflow, a standard
• Anthropic published the format as an open standard, so it is not locked to one product

What Is Inside a Skill Folder?

Every Skill folder has one required file — SKILL.md — and any number of optional supporting files. The SKILL.md file starts with a short YAML header, called frontmatter, followed by the instructions Claude should follow when the Skill is active.

The frontmatter has two required fields: name and description. The name must be lowercase with hyphens and 64 characters or fewer. The description must be non-empty and 1,024 characters or fewer, and it explains both what the Skill does and when Claude should use it.

Below the frontmatter you write the actual guidance — steps, examples, and rules — in normal Markdown. Supporting files sit in the same folder, and the SKILL.md body points Claude to them when a task needs the extra detail.

• SKILL.md — required; YAML frontmatter plus Markdown instructions
• name — lowercase, hyphens, max 64 characters; Anthropic suggests a gerund like "processing-invoices"
• description — max 1,024 characters; states what the Skill does and when to trigger it
• Supporting files — optional scripts, templates, and reference docs in the same folder
• The description carries the whole triggering burden — a vague one means Claude never opens the Skill

How Do Claude Skills Work Under the Hood?

Claude Skills work through progressive disclosure, meaning Claude reads only a tiny bit of each Skill until a task matches. At the start of a session, Claude loads just the name and description of every installed Skill — roughly 100 tokens each. That is enough to know a Skill exists and when to reach for it.

When your prompt matches a Skill's description, Claude then reads the full SKILL.md body into memory and follows its instructions. If those instructions point to a script or reference file, Claude opens that file only at the moment it is needed.

This three-level loading is why you can install many Skills without slowing Claude down or crowding its context window. It is also the key difference from an MCP server, whose tool definitions can consume tens of thousands of tokens up front.

• Level 1 — frontmatter (name + description) is always loaded, about 100 tokens per Skill
• Level 2 — the full SKILL.md body loads only when a task matches the description
• Level 3 — scripts and reference files open only when the instructions call for them
• Skills are composable — Claude can stack several at once and coordinate them
• Skills can run real code, so some tasks are done reliably instead of guessed at

How Do You Create a Claude Skill?

You create a Claude Skill by making a folder, adding a SKILL.md file, and writing a short YAML header plus your instructions. That is the whole minimum — a folder and one Markdown file. Everything else is optional.

Start with the frontmatter. Give the Skill a lowercase, hyphenated name and a clear description that says what it does and when to use it. Then, below the header, write the steps or rules Claude should follow, the same way you would brief a new hire.

If you would rather not touch files at all, Anthropic ships a "skill-creator" Skill that interviews you about the workflow and generates the folder and files for you. For a business team, this is usually the fastest path from idea to working Skill.

• Step 1 — create a folder named for the job, like "drafting-client-emails"
• Step 2 — add a SKILL.md file with name and description in the YAML header
• Step 3 — write your instructions in plain Markdown under the header
• Step 4 — optionally drop in a template, checklist, or script the Skill can use
• Shortcut — use Anthropic's skill-creator Skill to build it by conversation, no manual editing

Where Do Claude Skills Run?

The same Claude Skill runs everywhere Claude runs — Claude.ai, Claude Code, Claude for Small Business, the Anthropic API, and Enterprise. You author a Skill once and reuse it across surfaces, because the format is a portable open standard rather than a product feature.

Portability also spans models. A Skill built for one Claude model works with the others — Sonnet, Opus, Fable, Mythos, and Haiku — because a Skill is just instructions and files, not model-specific code.

On the Developer Platform, Skills pair with the code execution tool so scripts inside a Skill can actually run. In the consumer and Team apps, they load automatically when a task matches, with no setup for the end user.

• Claude.ai — available to Pro, Max, Team, and Enterprise users
• Claude Code — installed via marketplace plugins or dropped in manually
• Anthropic API — runs alongside the code execution tool for scripted steps
• Claude for Small Business and Enterprise — deployed team-wide
• Model-agnostic — one Skill works across Sonnet, Opus, Fable, Mythos, and Haiku

Skills vs MCP, Custom GPTs, and Projects at a Glance

Skills, MCP, Custom GPTs, and Projects solve different problems, so the choice depends on what you need. Skills package a repeatable capability as files. MCP is a live protocol that connects Claude to outside tools and data at request time. Custom GPTs live in OpenAI's runtime, and Projects give a single chat a static context folder.

Reach for a Skill when you want the same output shape every time without re-prompting. Reach for MCP when Claude needs live access to a database, ticketing system, or CRM. The two often work together — a Skill for the method, MCP for the live data.

This is only the quick contrast. For the full decision framework and business tradeoffs, use the dedicated pages linked below rather than choosing from this summary alone.

• Skills — packaged instructions and files; lightweight, portable, load on demand
• MCP — a live protocol for tools and data; more powerful, needs infrastructure
• Custom GPTs — a competing product in OpenAI's runtime; does not reuse Claude workflows
• Projects — static context files scoped to one chat, with no dynamic loading
• Common pattern — Skills for the how-to, MCP for the live connection, used together

What Can You Do With Claude Skills?

Claude Skills are best for repeatable, rules-based work where you want a consistent result every time. Anthropic's own public library shows the range — document Skills that build and edit PDF, Word, PowerPoint, and Excel files, plus creative, branding, and communications Skills.

For a business, common categories include drafting on-brand documents, running standard checklists, applying compliance-approved templates, and turning tribal knowledge into a shared asset. Each of these is one Skill any teammate can trigger by simply describing the task.

If you want to see concrete, industry-specific examples and a build-versus-buy view, that is the job of the business and vertical pages linked below. This page is the mechanism; those pages are the playbooks.

• Produce documents in your firm's exact format — proposals, memos, reports
• Run repeatable checklists and multi-step workflows the same way every time
• Enforce brand, legal, or compliance rules automatically inside outputs
• Manipulate real files — PDFs, spreadsheets, slide decks — with bundled scripts
• Turn one expert's know-how into a reusable Skill the whole team can call

Who Should Learn About Claude Skills?

Anyone whose team repeats the same knowledge work should understand Skills, because Skills turn that repetition into a reusable asset. Owners and operators benefit most, since one well-written Skill can standardize an output that used to depend on a single person.

You do not need engineers to start. A team lead who can write a clear procedure can author a first Skill, then hand harder or compliance-sensitive builds to a specialist.

If your work sits in a regulated field — law, healthcare, accounting, insurance, financial advice — the payoff is standardization plus built-in guardrails. The vertical guides below show what that looks like in each industry.

• Owners and operators who want consistent, repeatable output
• Team leads ready to capture their best playbook once and share it
• Regulated firms that need brand and compliance rules baked in
• Anyone evaluating Claude before committing a whole team to it
• Next step — see the business hub and vertical guides for real examples