What is a Skill, and why should you build one this week?

If you've used ChatGPT, Claude, Gemini or Copilot for more than a few months, you've probably typed the same instructions into a chat more times than you can count. Or opened a new chat and had to re-explain who you are and what you need, again. Or set up a custom GPT once, felt proud of yourself, and never opened it again.

A Skill fixes all three of those, and it goes further than most people realise. Here's what a Skill actually is, why it's a different thing from a saved prompt, and how to build, test and refine your first one this week.

What a Skill is

A Skill is an ability the AI unlocks and then applies on its own. Think of it as training that sticks. Once you've taught the AI the ability, it knows the ability exists, knows what it does, and reaches for it whenever the work calls for it, without you asking.

• Build a presentation, and the brand guidelines load themselves.

• Analyse a dataset, and the analysis framework is already running.

• Write a report, and the house structure shapes it before anyone mentions the structure.

Underneath the analogy, it's a plain folder. Inside sits one instruction file written in ordinary English, describing how the task gets done, alongside whatever the task needs on hand: reference documents, brand assets, worked examples, and code scripts for the steps that have to run the same way every time. The shorthand: a Skill equals the prompt, plus the tools, plus the assets.

Two properties matter more than the rest.

1. Nothing loads until it's needed. A Skill can carry far more reference material than you could ever paste into a single chat, because the AI only reads the parts a task calls for.

2. A Skill can run actual code. That keeps the steps that must be exact, exact. The AI handles the judgement, the code handles the certainty, which is what keeps a well-built Skill honest on the parts that have to add up.

That's what separates a Skill from a saved prompt.

• A saved prompt is wording you reuse. You still have to find it, paste it, and hope you remembered the version with the latest tweaks.

• A Skill is a portable piece of a workflow, written in plain English, that produces the same output from the same inputs every time it runs, without you touching it.

In Claude, ChatGPT and most major platforms, this is now a real product feature. You don't need to write any code to build one. You just describe the job in plain English, the same way you'd brief someone new on your team.

Here's what that looks like in practice. You might build a Skill that sorts incoming enquiries into the right category. Or one that summarises a call into your standard format. Or one that checks a document against your criteria, or rewrites a response in your tone. Whatever the job is, the Skill does it the same way every time, so the same kind of request always gets the same kind of result, and once it's built, your whole team can use it too.

Because a Skill is just a folder of plain files, it's portable too. Claude's version is built on an open format called SKILL.md, and that same format is starting to turn up across other developer tools, not just inside one app. I pulled a written Skill apart section by section back when the feature first launched, to show what a Skill looks like and the type of information a skill contains, check out this example and breakdown here: read the anatomy of a Skill →

A quick reality check on which tools call this "Skills"

Worth knowing before you go looking for the feature. ChatGPT for Work launched Skills in July 2026, and Claude has had Skills for a while now, both in Claude Code and inside Claude Cowork, where you can even record yourself doing a task once and it builds the Skill for you (more on that below). Copilot has something similar called Agent Skills, though right now it mostly lives in developer tools like Visual Studio and GitHub Copilot rather than the everyday Copilot most of us use at work. Gemini's version is called Gems. Different name, same idea: a standing set of instructions the AI applies consistently.

So the tools and the terminology vary a bit. The underlying shift doesn't. Every major AI platform is heading the same way, toward an assistant that remembers how you like work done.

Skills versus Projects, and why the difference matters

If you already use Claude Projects, this one is worth being clear on, because the two do different jobs.

A Project is a persistent workspace. You hold your documents, chat history and ongoing context in it, and you choose which Project to work in each time. It suits long-term research or a topic that keeps evolving over weeks.

A Skill works differently. It's available across your whole account, not just inside a single Project, and it loads automatically the moment Claude decides it's relevant, or the moment you name it yourself. Because it only loads what a task needs, it doesn't eat into your conversation space the way a Project full of background instructions can.

Think of the Project as where the context lives, and the Skill as the workflow that runs on top of it. Several Skills can stack inside one Project conversation. One reads a document, another formats the output, a third checks it against your standards, all in the same thread, without you switching tabs or re-explaining anything.

Build it, then test it. A Skill is never finished.

The first version of a Skill is rarely right, and that's normal. Run it on a real task, read the output properly, and correct it the same way you'd correct a new team member in their first week. Save that correction back into the Skill, and it stops making the same mistake. Skills aren't shipped once and left alone. They're sharpened, a bit more each time you use them.

A Skill you build once and never touch again is just a saved prompt wearing a better name.

What's good, and what still needs a workaround

Skills tend to be more consistent than the instructions you'd load into a Project. A Skill sticks to what you told it, handles a rough prompt without losing the intent, and gives roughly the same quality of output the tenth time as the first. Because context lives inside the Skill rather than being pasted into every conversation, Claude also stays more flexible mid-chat, it won't cling to an old instruction when you want to take the conversation somewhere new.

A few rough edges are worth knowing before you build your first one.

• It doesn't always trigger on its own. Even when the use case looks obvious, you'll sometimes need to name the Skill directly in your prompt, "use the [name] Skill for this," until you trust it to fire reliably.

• Claude needs a clear description to pick the right Skill. When you're building one, spend the extra few minutes on the line that tells Claude when to use it. Be specific, and test a few different ways someone might phrase the request.

• Building the folder by hand is fiddly. Skip the manual build. The Skill Creator, covered below, builds and checks the structure for you.

How to build one

A practical note first: Skills sit behind a paid plan on most platforms, not the free tier.

In Claude:

• Ask directly, in any chat or in Claude Code: "create a Skill called [name] that does [task]." Or, after Claude has already done something well for you, "can you save what you just did as a Skill and call it [name]?"

• Use the Skill Creator, a meta-skill that builds Skills for you through conversation. In Claude.ai, go to Settings, then the Capabilities tab, and toggle it on under Skills (a handful of other pre-built Skills live there too, covering things like brand guidelines and internal comms). Open a new chat, describe exactly what you want the Skill to do, and it builds the folder structure, writes the instruction file, and bundles in any code or assets you give it, ready to download and upload back into Capabilities.

In ChatGPT, you can ask it to build the Skill for you in chat, use the Skills editor in your workspace settings, or upload a Skill package if you already have one.

Either way, once it's built, run it on something real before you trust it on anything that matters.

A real example worth stealing

Say you've just come off a client call. Paste your rough notes and say "summarise call notes." Your Skill formats them into decisions, actions and owners, your way, every time.

Then say "turn this into a follow-up email for the client." A second Skill picks up that context, matches your tone, and signs off the way you like. Two Skills, one workflow, done in under a minute.

What could a Skill do for your week?

The best candidates are things you do repeatedly, that follow a consistent structure, and where the output always needs to come out the same shape.

If you do it every week, if it follows a pattern, and it always needs to come out the same shape, that's your Skill.

One thing worth knowing before you start

What I've described here is the simple end of what a Skill can do, a reusable instruction that saves you retyping yourself and gets sharper the more you correct it. That's useful on its own, and it's where everyone should start.

But Skills can go a lot further than that. Some read your files and search the web before they answer. Some act inside the tools you already use, updating a CRM record or scheduling a send. And the most advanced ones remember what happened last time and get better at the job the longer they run.

That's a bigger piece for another week. For now, pick one thing you do every week that always needs to come out the same way, and build the Skill for it. Ten minutes, and you'll never retype that brief again.

If you want to move deeper and faster to set up Skills from simple to the complex across your entire workstreams and transform your day to day productivity then our Claude Cowork Programme might be just right for you. Learn more here.