Skills
Why have skills
Most of the work you do with the agent is one-off. But some things you do over and over — and every time you explain them again.
A skill is a way of explaining it for the last time.
It's a plain text file with a process: what should be done, in what order, what to watch out for, what the output should look like. You save it, and from then on you call it with a slash, like any built-in command.
How a skill comes about
Not by sitting down and writing it.
It comes out of work you've already done once. You set a task, fine-tune the result until you're happy with it, and then say: save this as a skill. The process you've worked your way to gets written into a file.
The second time, you call it with a single word, /skill-name. And when you find something's missing, you tell Claude to add it — so the skill gets better with every use. Exactly the way I describe it for briefs.
Skill examples
I have dozens and use them every day. Three to give you an idea:
/firma-info
From a company name or registration number it builds an overview of the company: financial data from public registers, the ownership structure, key people, its media footprint. The output is a page, not a paragraph of text.
/research
Market research in a given category. Market size, the main players, trends, who is growing. The Czech Republic and Europe.
/zadani
Turns messy notes into a usable brief. It thinks through what's missing, asks about what matters and clearly marks what it filled in itself.
Where to get them
They're all public on GitHub: jmech-pepa/pepa-claude-toolkit
There are ten of them, and not one is about programming. The skills are written in Czech — Claude will happily translate the one you want:
| Skill | What it does |
|---|---|
/research | Research into a market or a field — size, main players, trends. The Czech Republic and Europe. The output is a page with charts. |
/firma-info | Company analysis from public registers: finances, owners, key people, media footprint. Also handles Slovak, German, Polish and British companies. |
/zadani | Turns messy notes into a usable brief. Asks about what matters and marks what it filled in. |
/brainstorm | Ideas from five different techniques at once, then merges them and picks the five best. |
/data | Spreadsheets — CSV, Excel and the like. Overviews, comparisons, finding anomalies. It calculates outside the conversation, so file size doesn't matter. |
/pdf | Batch extraction of data from a pile of PDFs into a single Excel file. |
/pdf-extract | Pulls your highlights, underlines and handwritten notes out of a book in PDF. |
/banana | Generating and editing images — changing the background, adding or removing an object. |
/knihovnik | A personal library of articles, videos and books that links itself together. |
/ship | Publishes a website and checks that the deployment really worked. |
The examples/ folder also has a template for a skill that writes in your own style, and an excerpt from my CLAUDE.md on when it makes sense to reach for a more expensive model.
Download the repositoryA project folder stored on GitHub, together with the history of every change. You can download all of it at once as a ZIP. and run install.sh. The skills are copied to where Claude Code looks for them and are available right away. Three of them (/banana, /pdf, /pdf-extract) need a few extra things installed — the repository says what for each one.
You don't have to take them all. Feel free to take one, rewrite it in your own image and throw the rest away — that's exactly why they're out there.