What Are Claude Skills? A Plain-English Guide for Professionals

What Are Claude Skills, Exactly?

A Skill is a folder of instructions, scripts and reference material that Claude loads when your request matches what that folder covers. Anthropic describes them as a way to teach Claude how to complete a specific task in a repeatable way — company brand rules, a reporting format, or in the case of the built-in ones, how to construct a valid Excel workbook.

The mechanism has a name: progressive disclosure. Claude reads the short descriptions of the Skills available to it, decides which are relevant to what you just asked, and pulls in the full instructions for those only. Nothing else gets loaded.

Why should you care about the plumbing? Because the practical answer to what are Claude Skills lives here, in the behaviour you will actually notice. You never pick a Skill from a menu. You describe what you want, and Claude decides. That is a real shift from the way most people have learned to work with AI, where every session starts by re-explaining the rules.

One detail worth knowing before you invest any time: the Agent Skills specification is published as an open standard at agentskills.io. A Skill you build is not locked to Claude, which matters if you are the person who has to justify the choice later.

The Four Built-In Document Skills

Anthropic ships four document Skills to every user, and Claude invokes them automatically when they fit. These are the ones doing the work in almost every professional scenario you will hit.

The distinction that matters is between a file and a description of a file. Older AI output gave you the second one and left the construction work to you. What actually changed is that Claude now runs code in a sandbox to build the file, which is why the switch controlling all of this is called code execution.

These Skills also follow you into the Claude add-ins for Excel, PowerPoint, Word and Outlook. Type a forward slash in the sidebar to see what is available, or just describe the task and let Claude recognise it.

How to Switch Skills On

Skills need one setting enabled before anything works, and this is where most people stall. If your Skills list looks greyed out or missing entirely, this is why — it is not your plan.

On a personal account, go to Settings, then Capabilities. Look for Code execution and file creation and make sure it is on. On a Team or Enterprise account this lives in Organization settings and an owner may need to enable it for everyone.

Go to Customize, then Skills. You will see the Anthropic-built Skills alongside anything you or your organisation has added. Each one has a toggle and a one-line description of when it applies.

Toggle the document Skills you actually want. There is no cost to leaving them all on, and Claude only loads one when the task calls for it — that is the whole point of progressive disclosure.

Do not test with something trivial. Take a document you genuinely need converted this week and ask for it. You will learn more from one real attempt than from any tutorial, including this one.

The Output Contract, Explained

Switching Skills on is the easy half. The half that determines whether you get something usable is how you phrase the request — and there are exactly four things worth specifying. We call it the Output Contract: Format, Structure, Source, Standard.

Format is the one people skip, and it is the one that causes the most complaints. If you ask for “a summary of this data” you may well get prose. If you ask for “an .xlsx”, the Excel Skill has an unambiguous trigger. Say the extension out loud.

Standard is the one people have never been taught. You are telling Claude what a correct result looks like so that you know what to check when the file lands — and so it can catch its own obvious errors before you do.

Most guides answering what are Claude Skills stop at the feature description and never get to this part. The feature is the easy half — the phrasing is what separates people who find Skills useful from people who conclude they are overhyped.

Use your [Excel / Word / PowerPoint / PDF] Skill for this.

Format: return a downloadable [.xlsx / .docx / .pptx / .pdf] file.
Structure: [the tabs, sections or slides it must contain].
Source: use only the attached [document] — do not add figures that are not in it.
Standard: [what must be true for this to be correct, e.g. every total is a live formula, not a typed number].

If anything in the source is ambiguous, list it at the end instead of guessing.

That last line does more work than it looks like it should. It gives Claude somewhere to put uncertainty other than into your file.

Two Workflows Worth Stealing

Rather than list a dozen shallow examples, here are two that cover most of the ground — one that pulls data out of a document, and one that turns rough material into something presentable.

Statement to spreadsheet

This is the highest-value use of the PDF and Excel Skills together, and the one that converts sceptics fastest. Claude reads the layout of the statement, extracts the line items, and builds a workbook with real formulas rather than pasted values.

Use your PDF and Excel Skills on the attached statement.

Format: a downloadable .xlsx file.
Structure: one row per line item, with columns for date, description, category and amount. Add a summary tab grouping totals by category.
Source: the attached PDF only.
Standard: every total is a live SUM formula, dates use a single consistent format, and amounts are formatted as currency.

List any line items you could not read confidently at the end of your reply, rather than guessing at them.

Rough notes to client-ready document

The Word Skill is quietly the most useful one for anybody who writes for a living. Its real contribution is not the words — you can already get those — it is the heading styles, the contents page and the consistent layout that normally eat an afternoon.

Use your Word Skill to turn my attached notes into a client-ready report.

Format: a downloadable .docx file.
Structure: executive summary, then one section per theme, then recommendations. Add a contents page.
Source: my attached notes only — do not invent findings or quotes.
Standard: proper Heading 1 and Heading 2 styles throughout so the contents page updates automatically, and consistent body formatting.

Keep my phrasing where it is already clear. Flag anything too thin to stand as a finding.

If your organisation has a house template, upload it alongside the notes and ask Claude to build into it. That single habit removes most complaints about generic-looking output.

Skills vs Projects, MCP, and Instructions

Understanding what are Claude Skills is only half the job. The other half is knowing which of four similar-sounding features you actually want, because they get confused constantly.

Each does a genuinely different job. They do genuinely different jobs, and picking the wrong one is the most common reason people conclude a feature “does not work.”

The pairing that trips people up most is Skills and connectors. Anthropic’s own framing is the clearest available: connectors give Claude access to tools, and Skills teach it how to use those tools well. They are complements, not alternatives — if you are weighing them up, our guide to Model Context Protocol for non-technical professionals covers the access side properly.

For the other two, the line is about permanence. Claude Projects hold knowledge that is always loaded within them, while custom instructions apply everywhere and never switch off. Skills sit between the two: specific, and dormant until needed. If you are new to the tool entirely, start with what Claude AI is and how it differs from ChatGPT before layering any of this on. The generated files themselves surface in the Artifacts panel, and the same Skills carry into Claude Cowork if you work there.

Who Gets Skills and What Stays Private

Here is the correction most articles on this topic need: Skills are not a paid feature. According to Anthropic’s Skills documentation, they are available to users on Free, Pro, Max, Team and Enterprise plans. The gate is the code execution setting, not your billing tier.

• What your plan does not gate: access to the four document Skills, or the ability to add your own
• What it does gate: how much you can run before hitting usage limits
• What a business plan adds: central provisioning, so an owner can push approved Skills to everyone

Paying does buy you something — usage headroom, so long conversions do not stall halfway. But if you have been holding off because you assumed the document Skills sat behind a subscription, they do not. Our breakdown of Claude’s free, Pro and Max tiers covers where the real differences fall.

Privacy is where you should slow down, because the boundary is not the one people assume. It is not free versus paid. It runs between consumer accounts and business accounts.

You can check this yourself on Anthropic’s plans page, where consumer tiers are marked opt-out and business tiers say no training by default. A Pro subscription is still a consumer account. If you are uploading client contracts or salary data through a personal Pro login because it felt like the “professional” tier, that assumption needs revisiting today.

The practical move for most people is a two-minute one: open your privacy settings and check the training toggle rather than assuming. Our fuller guide to whether Claude is safe to use at work walks through retention and what to anonymise.

When Skills Go Wrong

By this point what are Claude Skills should be a settled question. The live one is why yours is not behaving — and most failures are one of a small handful, each with a fix that takes under a minute.

The greyed-out case catches out a surprising number of people on work accounts, because nothing on screen explains that an administrator made the decision. If toggling does nothing, that is your answer.

The workbook you produced shows formula errors when I open it.

Rebuild it using only formulas compatible with older Excel versions — standard SUM, IF, VLOOKUP and similar. Avoid newer functions such as XLOOKUP or LET.

Keep the structure and data identical. Return a fresh .xlsx download.

What Skills Still Cannot Do

Skills are genuinely useful and genuinely oversold, and the gap between those two is worth naming plainly.

There is a security dimension too, and Anthropic is direct about it. Prompt injection and data exfiltration through untrusted third-party Skills are the named risks, and the standing advice is to install Skills only from sources you trust. Reading a colleague’s shared Skill before enabling it is not paranoia — it is the same care you would apply to a macro-enabled workbook. If that risk is new to you, our guide to prompt injection and what professionals should know is worth ten minutes.

The built-in Anthropic Skills are not where this risk lives. They are maintained by Anthropic and enabled by default reasoning. The caution applies to what you add on top.

Where This Leaves You

The honest summary is that Skills remove a category of work most people had stopped noticing — the retyping, the restyling, the rebuilding of things that already existed in another format. That is not a small category. It is most of what makes document work tedious.

If you arrived asking what are Claude Skills and wanted one thing to take away, it is the four-line habit that turns a vague ask into a finished file:

• Format — say the extension out loud
• Structure — the tabs, sections or slides it needs
• Source — exactly what it may draw from
• Standard — what has to be true for it to be correct

If you want to understand why loading the right context at the right moment matters so much, context engineering is the wider skill underneath all of this.

Switch on code execution, pick one document you would otherwise spend an hour reformatting this week, and write it a proper Output Contract. That is the whole test.