Reading 765 Skills: What the Ecosystem Looks Like Now
A single-page snapshot of the directory, with links to the analysis behind each number.
| Skills indexed | 765 |
|---|---|
| Fully evaluated | 591 (77.3%) |
| Distinct authors | 65 |
| Distinct source repositories | 66 |
| Median score | 8.5 |
| Average score (evaluated only) | 8.53 |
| Median detail views per skill | 2 |
| Skills never downloaded | 152 (19.9%) |
The distribution
Scores cluster tightly around the middle:
| Band | Skills | Share |
|---|---|---|
| 9.5+ | 18 | 2.4% |
| 9.0 – 9.5 | 165 | 21.6% |
| 8.0 – 9.0 | 293 | 38.3% |
| 7.0 – 8.0 | 97 | 12.7% |
| Below 7.0 | 18 | 2.4% |
More than half the directory sits between 8.0 and 9.0. The extremes are rare in both directions.
The supply
Six authors supply 535 of 765 skills. The largest single catalogue is 119 skills. The median author publishes three.
That concentration is the single most important fact about the directory, because it means most category-level statistics are really author-level statistics.
The dimensions
Averages across evaluated skills:
| Dimension | Average | Weight |
|---|---|---|
| Trigger quality | 7.72 | 20% |
| Structure | 8.37 | 15% |
| Workflow design | 8.20 | 25% |
| Content | 8.88 | 10% |
| Engineering | 8.53 | 10% |
| Security | 9.69 | 20% |
Trigger quality is the weakest dimension and security the strongest — a pattern that has held since the first analysis and is partly structural (security is a floor almost everyone clears).
The attention
The demand side is much thinner than the supply side:
- Median views: 2
- Median downloads: 5
- 20% of skills have never been downloaded
- The top 10% of skills account for 31% of downloads
A flat, young distribution. There is no entrenched head.
What changed since the last snapshot
We published a version of this at 591 skills. The shape has not moved much: the median score is up 0.1, the distribution is still centred between 8 and 9, and the download curve is still flat.
The more significant change is not in the numbers but in the data quality story — the 174 placeholder rows described here, and the 12 placeholder taglines found here. Both are defects we introduced, both are now documented.
How to read a directory like this
Three rules that apply to ours and to any other:
- Check completeness before comparing averages. A placeholder is not a neutral default.
- Treat download counts as exposure, not quality. Ours correlate with age and placement far more than with anything the user decided.
- Read the rationale, not the decimal. Our own re-runs move scores by up to a point.