What Are Claude Skills? A Plain Guide for Lean Ops Teams

You asked Claude to turn a messy export into a formatted invoice, and it fumbled the formulas. Or you keep pasting the same three paragraphs of "here is how our team writes a status update" into every chat. Claude Skills exist to fix exactly that kind of repeated, task-specific work.

Claude Skills, which Anthropic also calls Agent Skills, are reusable folders of instructions, scripts, and reference files that Claude loads only when a task needs them. Each one is built around a single file, SKILL.md, that tells Claude what the skill does and how to do it. Anthropic launched them on October 16, 2025.

The confusion is fair, because a skill overlaps with MCP and with a CLAUDE.md file without replacing either. Here is what a skill actually contains, how Claude loads one only when the task needs it, and the honest answer on whether a lean team should bother building its own.

What a Claude Skill actually is

A skill is a folder on a filesystem. At its center sits a required SKILL.md file with two parts: YAML frontmatter holding a name and description, and Markdown instructions Claude follows once the skill is selected. That is the whole minimum.

Around that file, a skill can bundle whatever the task needs:

• Scripts Claude runs when code is more reliable than free-form generation, like writing Excel formulas or filling a PDF form.
• Reference files Claude opens only when a detail is required, so the main instructions stay short.
• Assets such as templates, fonts, or a brand style sheet.

Anthropic shipped a set of pre-built skills at launch for document creation, covering .pptx, .xlsx, .docx, and PDF form-filling, per its Agent Skills overview. The point is portability. You write the skill once, and Claude reaches for it wherever the skill is installed.

For an ops team, the mental model is simple. A skill is the written-down version of a repeatable job you would otherwise re-explain every time: "format this data as our monthly board deck," "extract line items from a supplier PDF this exact way," "draft a refund reply in our tone."

How Claude Skills work: progressive disclosure

The reason skills do not bloat every conversation is a loading model Anthropic calls progressive disclosure. It works in three levels.

1. Claude reads only the lightweight metadata first, the skill's name and description, to decide whether the skill is even relevant.
2. If it is relevant, Claude loads the full SKILL.md instructions.
3. Only when the task actually needs them does Claude open the bundled reference files or run the scripts.

So a folder with 30 pages of reference material costs almost nothing until the moment Claude needs page 12. That keeps the context window lean, which matters when you are chaining several steps.

Skills are also composable. Claude can pull in more than one for a single job, stacking a data-extraction skill with a document-formatting skill. In Claude Code you can also trigger one directly by typing /skill-name, documented in the Claude Code skills reference.

Claude Skills vs MCP vs CLAUDE.md

This is where most people get confused, because all three feel like "ways to customize Claude." They solve different problems.

| | What it is | What Claude gets | |---|---|---| | Skills | Packaged instructions and scripts for a task | How to do a specific job, loaded on demand | | MCP | A protocol for connecting to tools and data | What it can call or access (your CRM, a database) | | CLAUDE.md | A standing instructions file | Always-on context for a project or repo |

Put plainly: a skill is procedural knowledge, MCP is connectivity, and a CLAUDE.md is the rules that apply to everything. They are not competitors. A real automation often uses all three at once. A skill tells Claude how to build the report, MCP hands it live access to the numbers, and the always-on file sets the guardrails.

If you only ever tried one of these, that gap is usually why an early experiment felt limited. A skill with no data connection can format beautifully but has nothing real to format.

Where you can use Claude Skills

Skills run across Anthropic's products, so a skill you build is not locked to one surface. According to Anthropic's support documentation, availability breaks down like this:

• Claude apps (claude.ai): available on Free, Pro, Max, Team, and Enterprise plans, but they require code execution to be enabled.
• Claude Code: supported in beta, with direct /skill-name invocation, which we cover in our guide to Claude Code skills.
• Claude API: available to all API users who are using the code execution tool.

Anthropic also publishes an open library of example skills on its GitHub repo, which is the fastest way to see how a real SKILL.md is structured before you write your own. If you want a shortlist worth stealing first, we rounded up the Claude skills that actually earn their place, and for where to find and share more of them, there is a growing Claude skills marketplace.

The catch worth flagging: because most skills lean on running scripts, code execution is the common dependency. If that is off, the fancier document and data skills will not do much.

Do you need Claude Skills, and who builds them?

A skill is a genuinely useful building block. It is not a running workflow. That distinction decides whether you touch this yourself or bring in help.

If your need is a personal, self-contained task, write the skill yourself. Copy a template from Anthropic's repo, drop your instructions into SKILL.md, and you are done in an afternoon. We walk through that end to end in how to build a custom Claude skill. There is no reason to hire that out.

The moment it becomes a business process, the picture changes. A skill that formats an invoice is one piece. The workflow around it is the rest: pulling the closed deal out of your CRM, connecting Claude to Stripe and Slack through MCP or an API, running it on a trigger instead of a person remembering to ask, and monitoring it so a silent failure at 2 a.m. does not surface as an angry client email a week later. That is where a skill stops being a document and starts being infrastructure, and where Claude earns its keep for a small business instead of just answering chats.

That build is what we do at bottta. Our AI Automation work is exactly this: turning a promising skill or prompt into a workflow that runs on its own, with the Integrations wired to your actual tools and the monitoring that keeps it honest. We take it on two ways, a $3K per month retainer (flexible hours, up to 3 active workflows, ongoing monitoring and fixes) or a $4K fixed-scope project with integrations included and 30-day post-launch support. You can see both on the pricing section.

You could also stitch the surrounding automation together in Zapier or n8n and call the API from there. That works for a simple two-step job. It tends to break the moment the logic gets real, and then you own a brittle setup nobody documented. Before you build anything, it is worth being honest about whether the task is even worth automating yet. We wrote a whole framework on that in when to automate a task and when not to.

Frequently asked questions

Are Claude Skills free? The feature is included on paid Claude plans and the free plan, but they require code execution enabled, and API usage bills through the code execution tool. Anthropic lists availability across Free, Pro, Max, Team, and Enterprise in its support docs.

Is a Claude Skill the same as a plugin? No. A skill is a folder of instructions and files Claude reads. It does not run as a separate app or expose a new interface. It changes how Claude approaches a task, not what external systems it can reach. That second job belongs to MCP.

Can I use Claude Skills without coding? Yes for using pre-built ones, and mostly yes for writing simple ones, since SKILL.md is just Markdown. But skills that run scripts, and the automation that wraps a skill into a real business process, need real engineering.

Do skills work in Claude Code? Yes, in beta, and you can invoke a skill directly with /skill-name. This is where most developers first try them.

What is the difference between a skill and a CLAUDE.md file? A CLAUDE.md is always-on context for a project. A skill loads only when its description matches the task at hand. Use the always-on file for standing rules and a skill for a specific repeatable job.

When Claude keeps almost getting a task right, another prompt rarely closes the gap. A workflow built around the skill does, with the data connection, the trigger, and the monitoring that a bare skill lacks. That is the build we scope and run, so start a project or book a call with the task that keeps slipping.