The 18 Skills That Scored 9.5 or Higher
Eighteen skills out of 591. The list is short, and its composition says as much about the directory as about the skills.
| Score | Skill | Category |
|---|---|---|
| 9.8 | claude-api | Developer tools |
| 9.8 | xlsx | Documents |
| 9.7 | docx | Documents |
| 9.7 | pptx | Documents |
| 9.7 | hyperframes-cli | Video |
| 9.7 | remotion-to-hyperframes | Video |
| 9.6 | academy-guide | Productivity |
| 9.6 | create-video | Video |
| 9.6 | threejs-world-generation | Productivity |
| 9.6 | answer-me-with-html | Developer tools |
| 9.5 | baoyu-electron-extract | Developer tools |
| 9.5 | hyperframes-core | Video |
| 9.5 | hyperframes-media | Video |
| 9.5 | website-to-video | Video |
| 9.5 | constraint-driven-development | Developer tools |
| 9.5 | linkedin-reply-handler | Productivity |
| 9.5 | threejs-architecture-effects | Design |
| 9.5 | anatomy | Developer tools |
Who is on the list
Twelve of the eighteen come from two authors: calesthio (7 skills) and anthropics (5). The remaining six are one each from six different authors.
By category: video 6, developer tools 5, productivity 3, documents 3, design 1. Marketing, finance, data, and writing have none.
That last fact is the most interesting one. The categories with the largest skill counts — marketing (119), productivity (118), developer tools (125) — do not dominate the elite set in proportion. Marketing has 119 skills and zero at 9.5+.
What the elite skills have in common
Reading the rationales, a pattern repeats across the list:
- They are narrow.
xlsxis about spreadsheets.docxis about Word documents.hyperframes-cliis about one CLI loop. None are "a toolkit for X." - They ship working code. Every developer-tools and video entry here includes runnable scripts, not just instructions.
- They verify. The top entries include a step that checks the output — render the page, open the file, run the tests — before claiming done.
- They are maintained. Most are versioned packages with a current release, not a single committed file.
Why the list is narrower than it looks
Eighteen is 2.4% of the evaluated set, and it clusters by author. That means the list is partly a measure of who writes skills with these conventions, not a ranking of all possible approaches.
It is also not stable at the margin. We measured evaluator variance directly: re-running the same evaluation twice moves scores by up to a point. A skill at 9.4 and a skill at 9.6 are the same finding. The 9.5 cutoff is a useful conversation device, not a property of the skills.
Treat the eighteen as a sample of what the best current work looks like. Do not treat the nineteenth as meaningfully worse than the eleventh.
