The 152 Skills Nobody Has Downloaded
The skills with zero downloads are, on average, better than the ones with downloads. That sentence only sounds paradoxical if you assume the download count measures quality.
Of 765 skills in the directory, 152 (19.9%) have never been downloaded once.
Their profile:
| Metric | Zero downloads | At least one |
|---|---|---|
| Skills | 152 | 613 |
| Average score | 9.03 | 8.52 |
| Have a source repo | 100% | 100% |
Every single one of them is a real skill with a real source repository. None are broken imports or test entries. They are simply invisible.
Why the inverse
Three mechanisms, all of which point at exposure rather than quality:
One: they are newer. The directory added 195 skills in October. New entries have had less time to be discovered, and the import dates cluster — a large batch that arrived the day before this analysis has had roughly zero discovery window.
Two: they are in categories with less traffic. The 119-skill marketing catalogue and the 74-skill gamedev catalogue both sit in niches where the site's current visitors are not browsing. A good skill in a category nobody visits gets no downloads.
Three: the ones that do get downloads include the automated ones. A bot sweeping the directory downloads whatever it encounters first — typically the top-rated list, not the long tail. That raises the downloaded group's count without raising its quality.
Put together: the downloaded group is a mix of genuinely popular skills and whatever the bots happened to hit, while the zero-download group is simply the set that nothing has reached yet.
What this means for the directory
The single most misleading thing about a skill directory is the download number, because it encodes distribution and reads as quality. We publish both the score and the count side by side, which invites exactly that conflation.
The honest presentation would be to label the download count as "times fetched" rather than as popularity, and to stop using it to rank anything until the directory has enough human traffic for the number to mean something.
What this means for a skill author
If you publish a skill and it gets zero downloads, the data says nothing about whether it is good. The 152 zero-download skills average 9.03 — higher than the directory average.
What it does tell you is that the skill has not been encountered. The fix is distribution, not revision: get it into one more list, one more search result, one more README.
That is unsatisfying advice, and it is the same advice we give ourselves.