I picked the skill with more stars. Its repo had 46 skills tied for first place.
Описание и применение
Выбор навыка по количеству звезд репозитория приводит к коллапсу ранжирования: 46 навыков из популярного репозитория получают одинаковый балл. Решение – оценивать содержимое навыка по четырём независимым измерениям, а не выдавать один общий балл.
Задача
Choosing a skill based on repository stars collapses the ranking, making it impossible to distinguish individual skills within a popular repo.
Результат
The original metadata‑based ranking collapsed, producing a 46‑way tie. A single overall LLM score produced a flat distribution (many 85s). Re‑engineering the prompt to score four independent dimensions yielded a more granular distribution, enabling better skill selection.
Рабочий процесс
- 1. Observe that repository stars are used as a proxy for skill quality.
- 2. Notice that 46 skills from the popular repo tie for first place in the ranking.
- 3. Realize that metadata signals are identical for skills within the same repo, causing the collapse.
- 4. Understand that content must be read to distinguish skills.
- 5. Implement a new scoring method that asks an LLM to evaluate four dimensions (specificity, actionability, completeness, distinctiveness) instead of a single overall score.
- 6. Sum the four dimension scores to obtain a final skill score.
- 7. Use the improved ranking to select skills more accurately.
Доказательства
Forty-six skills from that repo, tied for first place.
The system's honest answer to "which of these should I install?" was a list of 46 things, in no particular order.
The obvious implementation: send the skill, ask for a 0–100 quality score, store it.
Instead of one 0–100 score, the judge now scores four dimensions independently, 0–25 each, and the code sums them.
- Дата записи: 2026-10-10T05:55:50.436Z