Dataset report · source-based Claude Skills ecosystem report: 87 public candidates analyzed
Data updated 2026-09-15 · Rules and source signals, not hands-on ratings
Snapshot
This report is generated from the same structured records that power the directory.
It does not estimate output quality and does not convert repository stars into
per-skill adoption. A candidate passes only when its exact source, skill file,
maintenance, license, category, platform, and confidence signals satisfy the current gate.
| Measure | Count | Share |
| Public candidates analyzed | 87 | 100% |
| Index-eligible profiles | 40 | 46.0% |
| Exact path-level SKILL.md found | 73 | 83.9% |
| Usable license signal | 71 | 81.6% |
| Updated or verified within 30 days | 53 | 60.9% |
| Updated or verified within 90 days | 70 | 80.5% |
Qualified skills by category
Qualified skills by creator type
Source concentration
Multiple skills can live in one repository. Official vendor repositories are shown in
full; community repositories are capped at ten index-eligible profiles so the directory
does not become a mirror of a single collection.
| Repository | Qualified profiles |
| anthropics/skills | 14 |
| obra/superpowers | 10 |
| sanjay3290/ai-skills | 10 |
| mhattingpete/claude-skills-marketplace | 3 |
| NeoLabHQ/context-engineering-kit | 3 |
License signals across all candidates
| License signal | Candidates |
| MIT | 32 |
| Apache-2.0 | 30 |
| Not declared | 16 |
| GPL-3.0 | 4 |
| Source-available | 4 |
| Unknown | 1 |
What the pass rate means
The 46.0% pass rate is a publication threshold, not a quality rating. Profiles fail
or remain references when a source is stale, archived, missing an exact skill file,
missing a usable license signal, or too weakly mapped to a platform or category. The gate
is recalculated after source enrichment; it is not adjusted to produce a target page count.
Limitations
- The dataset is a curated sample, not a census of every public skill repository.
- GitHub repository stars describe the repository, not usage of an individual skill inside a monorepo.
- A recent commit can affect any file in a repository; it does not prove that every contained skill changed.
- Source verification does not replace security review or hands-on output testing.
- Non-GitHub and incomplete sources require separate manual verification and may remain noindex.
Reproduce and inspect
Read the scoring methodology, inspect the
public JSON feed, browse the
skill directory, or use llms.txt
to find the site's machine-readable entry points.