Forty New Skills From Seven Repos: What This Batch Looks Like
We imported forty skills from seven GitHub repositories today and ran the same six-dimension static evaluation we use for the rest of the directory. This is the batch-level view; the individual findings are in the posts that follow.
The directory went from 769 to 809 skills today. The new material is not a random sample of the ecosystem. It is a deliberate attempt to fill domains where we had almost nothing: education, HR, CRM and sales operations, customer support, project-management tooling, e-commerce, and Google Workspace.
Here is what the seven repositories look like after evaluation.
| Repository | Skills | Mean | Range |
|---|---|---|---|
| anthropics/k12-teacher-skills | 4 | 9.07 | 8.7 – 9.3 |
| genspark-ai/genoffice | 4 | 8.50 | 8.4 – 8.6 |
| NEON-Rutger/B2B-revops-skills | 6 | 8.47 | 7.7 – 9.0 |
| jdrhyne/agent-skills | 2 | 8.20 | 6.9 – 9.5 |
| gemini-cli-extensions/workspace | 5 | 7.00 | 5.9 – 8.3 |
| claude-office-skills/skills | 9 | 5.84 | 4.9 – 7.5 |
| nexscope-ai/eCommerce-Skills | 10 | 4.67 | 3.2 – 9.1 |
The first thing to notice is the spread between repositories. The best repository averages 9.07; the worst averages 4.67. That is a larger difference than the gap between most of our categories.
The second thing is that the ranking is not the one you would get from GitHub stars. genspark-ai/genoffice has 11,420 stars and averages 8.50. nexscope-ai/eCommerce-Skills has 1,305 stars and averages 4.67. anthropics/k12-teacher-skills has 606 stars and averages 9.07. claude-office-skills/skills has 552 stars and averages 5.84. Across these forty skills, the correlation between repository stars and evaluation score is r = 0.203.
That number is not an argument against stars as a popularity signal. It is an argument against using stars as a quality signal. A repository can be widely noticed and still contain skills that do not survive a closer read.
The batch is useful anyway
A 4.67 average does not mean every skill in the repository is bad. dynamic-pricing-ecommerce scores 9.1; shopify-seo scores 3.5. The repository average hides a five-point spread. The same is true in the other direction: jdrhyne/agent-skills averages 8.20 across two skills, but one of them is 9.5 and the other is 6.9.
This is the practical lesson for anyone sourcing skills from GitHub: choose skills, not repositories. A good repository raises the prior, but it does not replace the evaluation.
The detailed posts cover the dimensions (workflow is the weak one), the strongest domains (Anthropic's K-12 set), the bimodal e-commerce batch (9.1 strategy, 3.2 checkout), and what forty skills do to our category averages (marketing drops 0.31).
What we did not get
Accessibility is still zero. Translation and localization still have two skills. Healthcare, real estate, manufacturing, and supply chain are still empty. Forty skills is a useful batch, not a complete map.
The scores are static evaluations of the SKILL.md files, not dynamic sandbox runs. Treat them as a first filter, not a verdict. The rubric is public at our methodology page.