The Preinstalled Claude Skills Nobody Tells You About (We Counted Them)

Anthropic says Claude comes with built-in skills and loads them automatically when they’re relevant. That’s true, and it’s also where most explanations stop. Nobody publishes the list. So we went looking for the preinstalled Claude Skills in an actual session, counted what was there, and ran two of them end to end to see whether they were real capabilities or shelf decoration.

The count surprised us, and the gap it exposed is the useful part of this article.

What we did

Skills live as folders on the virtual machine Claude runs in. That’s not a secret — it’s how Anthropic describes them in the developer documentation, and it means the folder can be listed the same way you’d list any directory. So we asked Claude to do exactly that inside a live chat, with code execution enabled, and then to open each skill’s instruction file and report what it found.

Everything below with a number attached came out of that session on 30 July 2026. Where we’re relying on Anthropic’s documentation instead of our own check, we say so and we link it.

The count: 35 installed, 11 announced

Thirty-five skill folders were mounted in the container. They sat in three separate places:

Location	How many	What’s in it
Public	8	The document workhorses: Word, PDF, PowerPoint, Excel, plus file reading, PDF reading, frontend design, and one that makes Claude check Anthropic’s own docs before answering questions about Anthropic products
Examples	26	Everything else — writing, design, personal admin, developer tooling
User	1	A skill we built and uploaded ourselves

Here’s the part nobody documents. Of those 35, only 11 were announced to Claude in that session — listed up front, in the context Claude reads before it does anything. The other 24 were present on disk and completely absent from the list.

That’s not a rounding error. It’s roughly two thirds of what’s installed.

Announced and present are two different things

This distinction turns out to matter more than the raw number, so it’s worth being precise about it.

Announced means Claude knows the skill exists without looking. It’s in the model’s working context, so Claude can reach for it on its own when your request matches. Ask for a Word document and the docx skill fires without you naming it.

Present means the folder is there, the instructions are there, the bundled code is there — but Claude isn’t holding a note that says so. Something has to point at it.

The honest limit of our test: we found the other 24 by listing the directory, not by saying a skill’s name and watching it trigger. We can prove those skills work once you get to them. We can’t prove a plain request would surface them on its own. If you want one of them, name it.

The preinstalled Claude Skills worth knowing about

Most of the 26 in the examples folder are not useful to a content creator or a small business owner. Nine of them are personal-admin flows — ordering groceries, refilling a prescription, dealing with the DMV, cancelling a subscription — and they’re written for a Claude that can drive a browser and pop up structured questions. On a plain chat surface, several of those assumptions don’t hold. We didn’t test them, and we’re not going to pretend we did.

Three are genuinely worth your attention.

1. A skill that learns how you write

There’s a skill whose entire job is to read writing you have already sent, extract your voice, and save it as a reusable profile so future drafts sound like you instead of like generic AI. Its instruction file runs to about 42 KB — the longest of the 35 by a distance — and a good chunk of that length is guardrails: it asks permission before reading anything, refuses to treat anyone else’s writing as your voice, and explicitly forbids writing names, numbers, or account details into the saved profile.

For anyone publishing regularly, that’s the most valuable thing in the folder. It was not on the announced list.

2. A skill that audits your recurring charges

Another one takes a bank or card statement, finds the charges that repeat on a schedule, presents them as a checklist, and then handles the cancellations you tick. If you’ve ever suspected you’re paying for three tools you stopped using in March, that’s the skill for it — and it’s the same exercise we walk through in our own connector cost work.

3. A skill for building MCP servers

There’s a full guide to building Model Context Protocol servers sitting in there, with separate reference files for Python and Node. If you’ve read our coverage of the MCP specification and wondered where you’d even start, the starting point is already installed.

Beyond those three: ten ready-made visual themes with full colour palettes, a watercolour painting engine, an animated-GIF builder, a document co-authoring workflow, a tutoring mode, a financial calculator, and a skill that applies Anthropic’s own brand colours to whatever you’re making.

We ran two of them to see if they actually work

A folder full of instructions proves nothing. So we picked two skills that were not on the announced list and ran them to completion.

The painting skill

We followed its own instructions and nothing else: read the guide, wrote a scene, rendered a preview, looked at it, revised the two weakest things, rendered final. A 1600×900 image came out in 1.50 seconds.

It also broke first. The skill’s reference table lists an argument called taper without saying what type it takes. Passing it a single number crashes the render with a type error. The correct answer — it wants a sequence describing how the brush width changes along the stroke — is not in the documentation. It’s in the source code, which we had to open to get unstuck.

That’s a small thing, but it’s the kind of small thing that decides whether a tool is usable. A skill’s instructions are written by people, and people leave gaps.

The GIF skill

We generated 24 animation frames and asked it to save an emoji-sized GIF. The file it produced contains 12.

Not a bug. The skill knows Slack’s constraints — dimensions, colour count, duration — and quietly resamples your frames down to fit them, capping the palette at 48 colours along the way. Final file: 6.2 KB.

That silent correction is, we’d argue, the whole point.

What a skill actually gives you

It’s tempting to think a skill is a feature — Claude can now paint, Claude can now make GIFs. That’s the wrong frame, and it’s why “what are skills” explainers tend to feel hollow.

What a skill really carries is a constraint you didn’t know you had. Slack GIFs have a size ceiling. Word tables break in Google Docs if you set widths as percentages. Spreadsheets you hand to someone else need a legend and an example row. None of that is knowledge about the task — it’s knowledge about the destination, learned the expensive way by someone else and written down.

That’s also the cleanest line between a skill and a connector, which we drew in detail in Claude Skills vs MCP. A connector gives Claude access to something outside the chat. A skill gives Claude judgment about something it could already do badly. If your problem is “Claude can’t see my data,” you need a connector. If your problem is “Claude does this, but the output is wrong in a way I keep having to fix,” you need a skill.

Your list is probably not our list

This matters, so we’ll be blunt about it: what we found is what was mounted in our session, on our plan, on 30 July 2026. The files in that container were dated 24 July. Yours may differ, and there is no version number to compare.

Three things we know can change the picture:

• Code execution has to be on. Anthropic’s help documentation is explicit that skills require it, across Free, Pro, Max, Team and Enterprise plans. Without it there’s nothing to find, on any plan.
• Your organisation’s settings matter. On Enterprise, owners have to switch skills on before anyone sees them — and they can push additional skills to everyone in the organisation, which changes the count in the other direction.
• One of the 35 was ours. A skill we wrote and uploaded sat alongside the built-in ones. If you’ve ever uploaded one, your count is inflated by the same amount.

How to check your own preinstalled Claude Skills

You don’t need our list. You can produce yours in one message.

Make sure code execution is enabled in your settings, then ask Claude, in plain words, to list the skill directories available in its environment and show you the name and description at the top of each one. It’s reading its own filesystem, which it’s allowed to do, and the answer takes seconds.

Two habits worth keeping when you do this. First, read before you run: Anthropic’s own security guidance is to review a skill’s contents before enabling it, especially one shared by someone else, and to pay attention to bundled scripts. That advice applies with more force to skills you install than to the ones that shipped with the product, but the habit is the same one we argued for in MCP security for non-developers. Second, don’t assume a skill that exists is a skill that fits — several of the ones we found are built for a Claude with capabilities a plain chat window doesn’t have.

The short version

Thirty-five skill folders installed, eleven announced, twenty-four sitting there unmentioned. Two of the unmentioned ones ran to completion when we pointed at them, one of them after we had to read its source code to fix an undocumented argument.

The practical takeaway isn’t a list to memorise. It’s that the inventory is checkable in about thirty seconds, and almost nobody checks. Before you go looking for a tool to add, it’s worth finding out what’s already sitting in the box.

You Might Also Like

If this was useful, these go deeper on the same ground — what to connect, what to skip, and what it all actually costs.