Skill collections, ranked
A single skill has no numbers of its own — nobody counts installs, and stars belong to the repository it lives in. So the unit here is the collection, and the movement comes from our own daily snapshots of every source we track.
Stars the source repository gained over seven days, measured from our own daily snapshots — not a number GitHub shows. Repositories publishing a single skill are left out: their stars are for the product, not the collection.
Skills we actually publish from the repository, after folder-level deduplication and the description gate.
Total stars on the source repository. Attention, not adoption — nobody counts skill installs.
Share of the collection whose folder carries runnable files, not just a markdown instruction. Collections under 5 skills are excluded — one skill with resources would read as 50%.
Why rank collections instead of individual skills?
Because the movement is a property of the repository. Stars are gained by the repo, not by one folder inside it. Ranked per skill, every skill from the same source would show the same number — a collection of 864 skills would occupy 864 identical rows.
Where does the weekly change come from?
From our own snapshots. We record stars for every repository in the radar daily and difference them, which is why a seven-day number exists at all — GitHub does not publish one.
Does a high rank mean the skills are good?
No. It means the source repository is being starred, or that the collection is large, or that its skills ship runnable files. Those are the honest signals available; skill quality is not something stars measure.
What counts as a skill here?
A folder with a SKILL.md, deduplicated by folder name within its repository, and carrying a description long enough to say something. Copies of the same skill across a repository collapse into one.