What Is a Claude Skill? A Plain-English Answer

I have started saying the word ‘skill’ a lot, in posts, in client conversations, and on the consultancy AI operating system page. It is doing a great deal of work in those sentences, and when I searched for a plain-English explanation to point people at, I could not find one. The answers are either developer documentation or vendor gloss. So here is the version I give clients.

A Claude skill is a folder of instructions that teaches Claude, Anthropic’s AI assistant, how to perform a specific task your way. At its core is a file called SKILL.md containing a short description of when the skill applies and detailed instructions for how to do the work: your templates, your rules, your quality bar. Claude reads it automatically whenever the task comes up.

The mechanics, briefly

You do not need to be technical to follow this, and that is rather the point. A skill is a folder. Inside it sits one required file, SKILL.md, written in plain language with a small amount of structure at the top: a name, and a description that tells Claude when the skill should be used. Below that come the instructions, written the way you would brief a capable new starter. The folder can also hold supporting files, such as reference documents, templates, or scripts, which Claude reads only when it needs them.

What makes this cleverer than it sounds is that skills load dynamically. Claude sees the short descriptions of every skill you have installed, and pulls in the full instructions only when a task matches. Ask for a proposal and the proposal skill activates; ask about a spreadsheet and it stays out of the way. You are not stuffing everything into one giant prompt. You are building a library, and the right volume comes off the shelf at the right moment.

Skills work across Anthropic’s products: the Claude apps, Claude Code for developers, and the API for anything you build yourself. Anthropic has also published the format as an open standard, which matters if you are nervous about locking your firm’s know-how into one vendor’s file format. The details live in Anthropic’s Agent Skills documentation if you want the technical depth.

A prompt is typed. A skill is owned.

This is the distinction I find lands hardest with business owners. A prompt is something an individual types into a chat box and loses. It lives in one person’s head, varies with their mood, and leaves when they do. A skill is a file the firm owns. It can be versioned, reviewed, improved, and handed to every member of the team, and from the day it improves, everyone’s output improves with it.

Here is what that looks like in practice. My proposal skill does not say ‘write a good proposal’. It encodes my discovery questions, my section order, my pricing logic, and the phrases I never want near a client document. When I refine how I handle a pricing objection, that refinement goes into the file once. Every proposal after that includes it. Twenty years of method, compounding in a text file rather than depreciating in my head.

What a skill is not

A skill is not a plugin or an app; there is nothing to install in the software sense, and nothing your IT team needs to approve beyond Claude itself. It is not fine-tuning or training a model, which is an expensive machine learning exercise most firms will never need. And it is not magic. A skill is only as good as the method written into it, which is why the hard part of building one is never the file. It is getting a firm’s actual working method out of the heads that hold it and onto the page.

Why this matters beyond the technology

I run my consultancy on a library of these: proposals, client reports, blog posts, SEO pages, SQL Server health check reports. Individually, each one saves hours. Collectively, they are something more interesting: the first practical mechanism I have seen for a small firm to own its method as an asset rather than renting it from whoever happens to be in the building. That collective library is what I call a consultancy AI operating system, and I have written a full explanation of the operating system model separately.

The honest caveat: the skill file format belongs to the AI world and will keep evolving. Your firm’s encoded method does not. Written down properly, it ports between platforms, survives vendor churn, and improves with use. That is the asset. The file is just where it lives this year.

See skills running on a real firm

If you want to see what a library of skills looks like when it runs an actual consultancy, the consultancy AI operating system page covers the full model, with published pricing and a 30-minute working demo where I run my own firm on screen.

Frequently asked questions

Do I need to be a developer to create a Claude skill?

No. A skill is written in plain language, structured the way you would brief a capable new starter. The technical wrapper is minimal: a folder, one Markdown file, and a short description at the top. The difficult part is articulating your method clearly, not the file format.

What is the difference between a Claude skill and a custom GPT?

They solve a similar problem in different ecosystems. A custom GPT is configured inside ChatGPT; a skill is a portable folder of instructions that works across Claude’s apps, Claude Code, and the API, and follows a published open standard, so the encoded method is easier to move and version.

Where do Claude skills work?

Across Anthropic’s products: the Claude apps on paid plans, Claude Code, and the Claude API. Because Anthropic has published the format as an open standard, other tools that adopt it can run the same skills, though support varies by platform.

Can a skill contain confidential business information?

A skill typically encodes your templates, tone, and commercial logic, so treat the files with the same care as any business document. Keep them in your own storage, control who can edit them, and only install skills from sources you trust, since a skill directs how the AI behaves.