How concentrated is adoption across published agent skills?
Registry listing counts are widely quoted as a measure of ecosystem health. They measure publishing, not use. Here is the difference, in one corpus.
n = 20,590 · measured 2026-10-19 · recomputed on every page loadResult
| Observation | Count | Share |
|---|---|---|
| Reports zero installs | 9,976 | 48.5% |
| Reports fewer than five installs | 16,559 | 80.4% |
| Reports 100 or more installs | 289 | 1.4% |
| Highest install count observed | 18,479 | — |
| Median install count | 1 | — |
| Share of all installs held by the top 10% | 85.8% | — |
What this does and does not say
It says reported adoption is extremely concentrated: a median listing reports 1 installs while the top tenth accounts for 85.8% of the total. It does not say the long tail is low quality, unmaintained or abandoned. New, niche, private-use and personal skills all legitimately report low counts, and a count of zero may simply mean the registry does not track installs for that listing.
Why it matters here
It is the reason we do not attempt to watch everything. Monitoring effort is directed at what people actually install, because watching a listing nobody uses protects nobody.
Method
- Install counts are as self-reported by the registry. We record them; we do not verify them.
- Latest observation per listing. Listings that report no install figure at all are excluded.
- No individual listing is named in this report.
Full method and known limitations: methodology. Believe a specific record is wrong? Request a correction.