
A directory with twenty entries can be browsed. At a hundred and twenty, browsing stops working: the reader needs to know which category to look in before they can find anything. That changes what a category is for — it stops being a filing decision and becomes the first question the interface asks.
This page is about how the skill categories are drawn, using the current distribution as the example.
The short answer
Categories should map to the job a person is trying to do, not to the technology. Coding, writing, research, productivity, image, video, search — each is a job, and the skill inside it is the method.
The current distribution
| Category | Count | What it covers |
|---|---|---|
| Coding | 31 | Review, testing, migrations, CI |
| Productivity | 27 | Meetings, notes, decisions, inbox |
| Writing | 19 | Docs, changelogs, tone, outlines |
| Research | 18 | Synthesis, sources, interviews |
| Image | 11 | Alt text, generation workflow |
| Video | 8 | Shot lists, captions, retiming |
| Search | 6 | Intent, result evaluation, query logs |
The shape is not accidental. Coding and productivity are where skills started, because they are where repetitive instructions hurt most. Research is catching up because research work is full of rules that nobody writes down.
Why job-based beats technology-based
A technology category would be 'Claude Code skills' or 'MCP-adjacent skills', and it would be useless: the reader does not arrive thinking about the runtime. They arrive thinking 'I review a lot of code' or 'I write a changelog every week'.
The job framing also survives tool churn. A category named after a runtime decays when the runtime does; a category named after the work does not.
The section that matters most
Every entry carries a best-for and a not-for line, and the not-for is the useful one. At this size, the reader's problem is not finding a skill; it is filtering out the twelve plausible ones. A stated limit does that faster than any description.
The same logic applies to the whole category. A search category with six entries is a signal that search-specific skills are immature, which is useful information for a reader deciding where to invest.
What the distribution implies
Three observations from the counts. First, coding is saturated at the level of individual reviews and tests; the gaps are now in architecture and migration, which are harder to write as skills. Second, research is the fastest-growing area because evidence work has explicit rules. Third, search is thin because most search quality problems are evaluation problems, and evaluation is a habit more than a skill.
How to read the directory
Start from the job, open the two most likely entries, and read their not-for lines. If both are wrong, the category page's description tells you whether you are in the wrong category or the right category with a gap. That is a two-minute loop, which is the only kind that survives at scale.
Frequently asked questions
Why not fewer, broader categories? Because the category is the first filter. A category with sixty entries is a list, not a filter.
Will categories change? Yes. The test is whether a category answers 'what job is this' without the reader knowing any product names. See skills.
Where do new skills go? The category whose job they do, not the tool they run on. See writing a skill.


