Free guide Comment SHEET Checked Sep 9, 2026

The 5 Claude Code file skills — install sheet

The two commands that install five Claude Code skills, what each one hands back as a file, and how to read the bill before you add a sixth. Checked 2026-09-09.

Everything below is the order to do it in, not a list to browse.

1. Install (two commands, after a one-time setup line)

Open Claude Code and run:

/plugin marketplace add anthropics/skills

/plugin install document-skills@anthropic-agent-skills

/plugin install example-skills@anthropic-agent-skills

The first line is a one-time setup: it registers the marketplace. The two install commands are the two that matter: the first puts four skills in at once, the second adds the fifth. That is the “two commands” the video means.

Two things to know before you run it. This is Claude Code. Cowork and scheduled cloud sessions don’t read the skills on your machine; they load the skills enabled on your claude.ai account instead. And a plugin brings more than the skill you wanted: the official reference lists a plugin’s components as “skills, agents, hooks, MCP servers, LSP servers, and monitors.”

2. What you just installed, and the file each one makes

Skill	You ask for	You get
`pdf`	a PDF form filled in	the filled `.pdf` back
`docx` (Word)	a draft document	a `.docx` file
`xlsx` (Excel)	a summary of your numbers	an `.xlsx` file
`pptx` (PowerPoint)	a deck	a `.pptx` file

And the fifth, from the example bundle:

canvas-design: you ask for a design and it comes back as a .png or .pdf you can open, instead of a paragraph describing one. Its own description: “Create beautiful visual art in .png and .pdf documents using design philosophy.” Same check as the four above. The file is in your folder or it isn’t. It does not claim, and I don’t either, that the words inside got better.

3. Your first real test

Give Claude Code a task that has to end in a file. A folder of receipts into a monthly sheet. Meeting notes into a deck. A form PDF filled in.

You’re checking whether a file appears, not whether the answer sounds better.

4. Read the bill before you add a sixth

Run:

/skill-doctor

(Claude Code v2.1.252 or later.) Two columns matter: what each skill costs you, and how often it gets used.

A skill with a used-count of zero and a cost near the top of the list is your first candidate to turn off. A skill you use weekly earns its place no matter what it costs.

Anthropic’s docs explain why this matters. Every installed skill puts its name and description in a listing Claude sees on every turn, whether or not it uses the skill. That listing’s budget is 1% of the model’s context window, and each entry is capped at 1,536 characters. Go over it and descriptions get shortened, starting with the skills you invoke least. The skill stays installed. Claude just stops seeing why it would pick it. (skillListingBudgetFraction raises the budget if you want it raised.)

Anthropic publishes no recommended number. It publishes the budget and the tool to measure it. “Too many skills makes Claude worse” is what people report; “descriptions get cut, least-used first” is what’s documented.

5. The one question before installing anything else

Can I name the file this makes?

If you can, install it. If it only promises better output, leave it for now. The one public benchmark that ran a skill with and without it, blind-judged, found no confirmed better answer on any of its three tasks.

keep reading

The rest is free too.

One address opens the rest of this guide and every other one here, for good. I’ll email you only when a new one is out.

No card, no upsell. One click to leave.