Claude Skills: What They Are, How to Build One, and 5 Skills Every Operator Should Have

A skill is a procedure you write down once so you never have to explain it to Claude again.

Most operators use AI the way they use a new hire on day one. Every task starts with a briefing. How you onboard a client. How you structure a weekly review. What a finished SOP looks like in your business. The briefing works, the output is good, and tomorrow you type the same briefing again.

That repetition is a tax on leverage. It is also the exact problem Claude Skills were built to remove. This article explains what a Skill is according to Anthropic's own documentation, how the SKILL.md file works, where Skills run and where they do not, how they differ from Claude Projects, and the five Skills a coach or owner-operator should build first.

Every product fact below is quoted from Anthropic's documentation as of October 2026. Features change. Check the linked pages before you build.

What a Claude Skill actually is

Anthropic's platform documentation defines them plainly: "Agent Skills are modular capabilities that extend Claude's functionality. Each Skill packages instructions, metadata, and optional resources (scripts, templates) that Claude uses automatically when relevant" (Agent Skills overview).

The launch announcement on 16 October 2025 put it in one sentence: "Skills are folders that include instructions, scripts, and resources that Claude can load when needed" (Anthropic announcement).

The word that matters is folder. A Skill is not a prompt you paste. It is a directory on a filesystem. Inside sits a file called SKILL.md, plus any reference documents or scripts you want Claude to have on hand. The documentation draws the line against prompting directly: "Unlike prompts (conversation-level instructions for one-off tasks), Skills load on demand, so you don't have to repeat the same guidance across conversations."

Anthropic ships pre-built Skills for documents. As of October 2026 the overview lists PowerPoint (pptx), Excel (xlsx), Word (docx) and PDF (pdf). Custom Skills are the ones you write yourself, and those are the ones that matter for an operator.

How Claude decides to use a Skill

The mechanism is called progressive disclosure, and understanding it tells you how to write a Skill that triggers when it should.

The overview describes three levels. At startup Claude loads only the metadata, the name and description from the top of every installed Skill, at roughly 100 tokens per Skill. When a request matches a description, Claude reads the body of SKILL.md, which the documentation says should stay under 5,000 tokens. Any further files, schemas or scripts load only when the instructions point to them.

The announcement describes the same behaviour from the user's side: "While working on tasks, Claude scans available skills to find relevant matches. When one matches, it loads only the minimal information and files needed."

The practical consequence is in one line from the overview: "The description is what Claude matches your request against when determining whether to trigger the Skill, so it must say both what the Skill does and when to use it." A vague description produces a Skill that never fires. A precise one produces a Skill you forget you installed because it simply works.

The SKILL.md format

Every Skill requires a SKILL.md file that opens with YAML frontmatter. Two fields are mandatory: name and description. The platform documentation sets these limits as of October 2026:

• name: maximum 64 characters, lowercase letters, numbers and hyphens only, no XML tags, and it cannot contain the reserved words "anthropic" or "claude".
• description: must be non-empty, maximum 1,024 characters, no XML tags, and it "must include both what the Skill does and when Claude should use it".

One discrepancy worth knowing. Anthropic's Help Center article on custom Skills for claude.ai states a 200-character maximum for the description field (Creating custom Skills), while the platform documentation states 1,024. Write a description that fits the shorter limit and it will pass on both surfaces.

Below the frontmatter comes the body: instructions, examples, and pointers to any bundled files. Anthropic's authoring guide adds rules that separate a Skill that works from one that merely exists (Skill authoring best practices):

• Write the description in the third person. The guide warns that "inconsistent point-of-view can cause discovery problems", and gives "Processes Excel files and generates reports" as the good form.
• "Keep SKILL.md body under 500 lines for optimal performance." Split anything larger into separate reference files.
• "Keep references one level deep from SKILL.md." Claude may only skim a file that is referenced from another referenced file.
• Name Skills consistently. The guide suggests gerund form, such as processing-pdfs or writing-documentation, and warns against names like helper or utils.
• Assume Claude is already capable. The guide's default assumption is "Claude is already very smart", and its test for every paragraph is "Does this paragraph justify its token cost?"

A minimal Skill, written out in plain text, looks like this:

Line one: three dashes.
Line two: name: writing-sops
Line three: description: Writes standard operating procedures in the house format from a rough description of a process. Use when the user asks for an SOP, a process doc, a checklist for a recurring task, or says "document how we do this".
Line four: three dashes.
Then the body: a heading, the steps Claude should follow, the output template, and two short examples.

Where Skills run, and where they do not sync

Skills are available on three surfaces, and the differences between them matter for an operator deciding where to invest.

claude.ai

Custom Skills are uploaded as zip files in settings. The Help Center instructs: "Create a ZIP file of the folder. The ZIP should contain the skill folder as its root (not a subfolder)." On plan access the two Anthropic pages differ as of October 2026: the Help Center states "Skills are available for users on Free, Pro, Max, Team, and Enterprise plans", while the platform overview describes custom Skill uploads as available on Pro, Max, Team and Enterprise with code execution enabled. Check your own settings rather than assuming.

Sharing is individual. The overview is explicit: "Custom Skills are individual to each user. They are not shared organization-wide and cannot be centrally managed by admins." If you run a small team, every person uploads their own copy.

Claude Code

Skills are plain directories on your machine. The Claude Code documentation lists personal Skills at ~/.claude/skills/ and project Skills at .claude/skills/, and states that "Where you save a skill decides which sessions load it" (Use Skills in Claude Code). Claude uses them automatically when relevant, or you can invoke one directly as a slash command. Project Skills can be committed to version control, which is how a two-person operation shares them.

The Claude API

Skills run inside the code execution tool's container. You reference a skill_id, and custom Skills are uploaded through the Skills API. Sharing here is workspace-wide. The constraint is the sandbox: the overview states Skills on the API have "No network access" and "No runtime package installation".

The limitation that catches people is portability. From the overview: "Custom Skills do not sync across surfaces." A Skill uploaded to claude.ai is not available through the API, and Claude Code Skills are separate from both. Build the Skill once as a folder, keep it in a repository, and install it wherever you work.

One more line from the overview belongs in every operator's head: "Use Skills only from trusted sources: those you created yourself or obtained from Anthropic." A Skill is instructions plus code that Claude will follow. Treat installing one like installing software.

Skills versus Projects versus custom instructions

Three tools overlap here, and operators waste time by using the wrong one.

Custom instructions and prompts are conversation-level. They shape tone and context for the chat in front of you. They are the right place for facts about you and preferences that apply everywhere.

Projects are workspaces. You load documents, set instructions, and every chat inside the Project inherits them. They are the right place for a body of knowledge: your offer, your client roster, your brand voice. The guide to Claude Projects covers five setups for coaches.

Skills are procedures. They load on demand, in any conversation, when the task matches. The Claude Code documentation gives the cleanest rule for when to build one: "Create a skill when you keep pasting the same instructions, checklist, or multi-step procedure into chat, or when a section of CLAUDE.md has grown into a procedure rather than a fact."

Facts go in instructions and Projects. Procedures go in Skills. If you can describe the thing as "how we do X", it is a Skill.

Five Skills every operator should build

These are the five procedures a coach or owner-operator repeats most, written as Skill specifications. Each description follows Anthropic's what-plus-when rule. Build the one you repeated most this week first.

1. onboarding-clients

Description: Produces the complete onboarding package for a new coaching client, including the welcome message, intake questionnaire, first-week schedule and the CRM checklist. Use when the user says a client has signed, paid, or is starting.

Body: the sequence of onboarding steps in order, the house template for each message, the fields the intake form must capture, and a reference file holding the current programme options. The Skill ends by producing a checklist the operator can paste into GoHighLevel or whichever CRM runs the business.

2. running-weekly-stoic-review

Description: Runs the Friday weekly review in the Stoic format. Use when the user asks for a weekly review, says "review my week", or pastes a week of notes, metrics or journal entries.

Body: the five questions of the review, the order to ask them, how to separate what was in the operator's control from what was not, and the output format: three sentences of judgement, one decision for next week, one thing to stop. This turns the Stoic weekly review from a habit you remember into a procedure that runs.

3. writing-sops

Description: Writes standard operating procedures in the house format from a rough description of a process. Use when the user asks for an SOP, a process document, or a checklist for a recurring task.

Body: the SOP template (purpose, owner, trigger, steps, done-criteria, failure modes), the rule that every step starts with a verb, the rule that every SOP names what "done" looks like, and two examples. A reference file holds completed SOPs so the new ones match.

4. briefing-content

Description: Turns a topic or idea into a content brief with the angle, the audience problem, the structure and the sources to verify. Use when the user asks for a brief, an outline, or says "I want to write about".

Body: the brand's banned words, the structural rules for articles and posts, the requirement that every claim carry a source, and the brief template. The ten strategy prompts that work well as one-off requests become permanent once they live inside a Skill like this.

5. checking-offer-pricing

Description: Sanity-checks a coaching offer's price against delivery hours, capacity and margin targets, and flags where the offer breaks. Use when the user proposes a price, a package, or a new tier.

Body: the capacity model (hours per client per month, maximum clients, target hourly value), the margin floor, the questions to ask before approving a price, and a short table format for the answer. The point is a procedure that refuses to let enthusiasm set the price.

How to build your first Skill in an afternoon

Anthropic's authoring guide recommends a process it calls evaluation-driven development, and it maps cleanly onto how an operator should work.

1. Do the task once without a Skill. Run a real onboarding or a real review with Claude in a normal chat. Notice every piece of context you had to type.
2. Write down the repeated pattern. The guide's instruction: "identify what context you provided that would be useful for similar future tasks." That list is the body of your Skill.
3. Ask Claude to write the Skill. The guide states that "Claude models understand the Skill format and structure natively" and that you can "Simply ask Claude to create a Skill". Describe the pattern and ask for a SKILL.md.
4. Cut. Remove every explanation Claude does not need. The guide's example edit: "Remove the explanation about what win rate means - Claude already knows that."
5. Test on three real cases. The guide's checklist asks for "At least three evaluations created" before sharing. Use last month's actual clients, reviews or offers, not invented ones.
6. Install it where you work. Zip it for claude.ai, drop the folder into .claude/skills/ for Claude Code, or upload it through the Skills API. Remember that it will not sync between them.

Then use it for two weeks and watch where Claude stumbles. The guide calls this observing "Unexpected exploration paths" and "Missed connections". Each stumble is an edit to the Skill, not a reason to go back to pasting prompts.

Frequently asked questions

What are Claude Skills?

Claude Skills are folders containing a SKILL.md instruction file plus optional scripts and reference documents that Claude loads automatically when a task matches the Skill's description. Anthropic's documentation describes them as "modular capabilities that extend Claude's functionality". They exist to stop you re-explaining the same procedure in every conversation.

What has to be in a SKILL.md file?

YAML frontmatter with two required fields, name and description, followed by the instructions in Markdown. As of October 2026 the platform documentation caps the name at 64 lowercase characters with hyphens and the description at 1,024 characters, while the claude.ai Help Center lists a 200-character description limit. The body should stay under 500 lines.

How are Skills different from Claude Projects?

Projects are workspaces that hold documents and instructions for every chat inside them, which makes them right for knowledge. Skills are procedures that load on demand in any conversation when the request matches, which makes them right for repeated workflows. Anthropic's rule of thumb is to build a Skill when you keep pasting the same checklist or multi-step procedure into chat.

Where can Claude Skills be used?

As of October 2026 Anthropic documents Skills on claude.ai, in Claude Code and through the Claude API. Custom Skills do not sync between these surfaces; a Skill uploaded to claude.ai must be separately added to the API, and Claude Code Skills are filesystem folders separate from both. Pre-built document Skills are available on claude.ai and the API but not in Claude Code.

Are Claude Skills safe to install from other people?

Anthropic's guidance is to use Skills only from trusted sources, meaning ones you created or obtained from Anthropic. A Skill is instructions and code that Claude will follow, so a malicious one can misuse tools or leak data. Audit every file in a third-party Skill before installing it, and treat the decision like installing software.

Write the procedure down once

The Stoics kept their rules in writing because memory bends under pressure. A Skill is the same discipline applied to leverage: the procedure exists outside your head, it runs the same way on a bad day as on a good one, and it compounds every time you use it.

Start with the task you repeated most this week. Build that one Skill. Then notice how much of the three-tool stack an operator actually needs is procedure, not software.

If you want the operating rhythm those procedures plug into, the free 5-Day Stoic Operator Challenge installs it in five days: a training block, a Stoic review and a daily structure you can keep running when the business gets heavy.