SClaude Has SkillsThe Folder

Signature experience · editorial specimen

The Folder

A Claude Skill is a folder with a SKILL.md file — a name, a description of when to use it, and instructions Claude follows when the task matches. Optional references and examples sit beside it and load only when pointed at. Find official examples in theanthropics/skills repository. Write one by naming a single recurring job and describing exactly when to use it.

The desk below opens a labeled editorial specimen so you can read that anatomy without copying an Anthropic skill body. Claude Has Skills is an independent publication and is not affiliated with Anthropic.

The desk

Files on a desk. Click SKILL.md.

Three papers. The front door is required. The other two exist because this job has policy and an output shape. Click a paper, or walk the anatomy in order.

vendor-invoice-auditCHS-FOLDER-001

Open a file

Arrow keys move between papers. Keys 1 2 3 light the name, description, and instructions when SKILL.md is open. Space walks. Nothing is uploaded.

Required front door

vendor-invoice-audit/SKILL.md

Required

Why this file

Walk starts at the name field. Each beat is a labeled part of authoring, not a copied Anthropic workflow.

Readout

Three labels on SKILL.md, then the papers that keep it small.

This is the writing order. Name the job, write the trigger, write the procedure, then park anything that would bloat the front door. Official Skills documentation describes the same split: discovery metadata first, instructions when the task matches, supporting files on demand.

1 · name

name

The name is a stable identifier, not a title. Keep it lowercase, hyphenated, and easy to mention. Humans and Claude both use it to point at the folder.

2 · description

description

The description is the trigger contract. Write when to use the skill in the language people actually type, and write the nearby tasks that must not load it. If you cannot write this sentence, the job is not scoped yet.

3 · instructions

instructions

The body is the operating procedure: imperative steps, stop conditions, and one-level-deep pointers to files in this folder. Keep variant policy out of this file so the front door stays small enough to load.

references/

exception-rules.md

Long policy would bloat the front door and weaken the trigger. A named reference holds the exception table so SKILL.md can stay a procedure.

examples/

good-review.md

An example is a finished artifact the output should resemble. It is not a second set of instructions and it is not a copied Anthropic skill body.

Key facts

What You Are Looking At

A Claude Skill is a folder. Anthropic's own Skills documentation — the Help Center articleWhat are Skills?, the Claude Code Skills page, and theAgent Skills specification — describes the same object: a SKILL.md entrypoint with a name, adescription of when to use it, and instructions, plus optional files the agent reads only when the procedure points at them. That is the whole mechanism. It is not a prompt snippet, because it can carry files. It is not an MCP server, because the unit is a package you can open.

The Folder on this page is a labeled editorial specimen. Every line invendor-invoice-audit was written here to teach that anatomy. It is not a file from anthropics/skills, and it is not a substitute for the official examples in the directory. If you need a pattern from Anthropic, open their repository. If you need to see which part ofSKILL.md does which job, stay on this desk.

How To Write One

Name one recurring job. A skill that "helps with finance" will fire on everything and help with nothing. A skill that reviews a vendor invoice against a contract and a purchase order can be tested. Write the non-goals next to the goal. The Help Center page on creating custom Skillsand the Agent Skills best practicesboth start from a focused workflow rather than a dump of instructions.

Write the description before the body. That field is not marketing copy. It is the matching rule Claude uses to decide whether to load the folder. Include the task, the artifacts, and the exclusions. "Use when reviewing vendor invoices against contracts and purchase orders" can be evaluated. "Use for finance excellence" cannot. If the skill fails to trigger, revise this sentence before adding paragraphs.

Write the procedure as steps a careful operator would follow. Tell the agent what to collect first, which file to open next, what to stop for, and what the output should look like. Point at references/ and examples/ with paths one level deep. Leave scripts and assets out of this specimen on purpose: they are part of the portable folder format, which Convergent Skillsowns, not a requirement for a first Claude Skill.

Evaluate before you share it. Positive prompts should load the skill. Near-miss prompts should not. Safety prompts should pause. Thetrigger tester on this site is a scratchpad for that wording; the evaluation guide is the loop once the folder is real.

Where The Other Desks Are

The portable skill format — how an agent loads metadata, then the body, then scripts, references, and assets, and how far that directory travels across clients — isThe Skill Folder at convergentskills.com. That route is theirs. This one is /folder/, and it stays on Claude Skill authoring.

Using a skill inside a Claude Code engineering practice — hooks, subagents, CI, MCP, marketplace plugins as a workflow — belongs atclaudeshipscode.com. This site tells you where a Claude Code skill folder is installed. It does not become a Claude Code tutorial.

Why The Specimen Is Editorial

Copying an official Anthropic skill onto this page would teach the wrong lesson: that the way to understand Skills is to republish someone else's procedure. Official examples are already public, already linked from the directory, and already changing. A teaching specimen has a different job. It has to be short enough to label, dull enough not to be stolen as a product, and honest about being ours.

So we wrote a vendor-invoice audit. The job appears in the authoring guide as a description example; the folder here is the same job made inspectable. No real vendor, invoice, or contract is in it. The exception rules are invented policy for the specimen. Cite it as an illustration of anatomy, never as an Anthropic artifact.

FAQ

What is a Claude Skill?

A Claude Skill is a folder with a SKILL.md file: YAML frontmatter that names the skill and describes when to use it, followed by Markdown instructions. Optional references and examples sit beside that file. Claude loads the skill only when the task matches the description.

How do I write a Claude Skill?

Name one recurring job. Write a description that is a trigger contract — when to use it, and which nearby tasks must not use it. Write the procedure as imperative steps with stop conditions. Add a reference or an example only when it reduces error. Then test prompts that should fire, prompts that should not, and prompts that should pause.

Is this folder a real Anthropic skill?

No. It is a labeled editorial specimen authored by Claude Has Skills so you can see the anatomy without copying a file from anthropics/skills. The official examples belong in that repository and in the directory on this site. Cite this page as a teaching artifact, not as an observed Anthropic package.

Where do I find official Claude Skills?

Start with the official anthropics/skills repository on GitHub, which held nineteen top-level example folders when this page was last checked on September 19, 2026. This site’s directory and directory tool index those folders as patterns, not endorsements.

Is this the same as the portable skill-format desk on Convergent Skills?

No. This page teaches what a Claude Skill is and how to write the three labeled parts of SKILL.md. The portable directory shape — how agents load metadata, then the body, then scripts, references, and assets across clients — belongs at convergentskills.com/skill-folder/. Do not treat this Folder as that format walk.

How do I use a skill inside Claude Code?

Installation is in-lane here: project skills live in .claude/skills, personal skills in ~/.claude/skills, and marketplace plugins can ship more. Using skills as engineering practice — hooks, subagents, CI, MCP — belongs at claudeshipscode.com. This page stops at the folder you write.

Primary Sources

Claude Code Skills docs

Anthropic Docs

Canonical source for Claude Code skill locations, types, and runtime behavior.

Introducing Agent Skills

Anthropic News

Product announcement for skill availability, plan eligibility, the /v1/skills endpoint, and the Claude Code marketplace install path.