Describe a task. Get the one Claude Code skill that fits it, and watch how it was chosen.
Scores all skills by the words they share with your task. Each skill is indexed by its name, its description, and everyday requests written for it, so a task worded differently from the description still matches. No AI model, no cost. The best 50 go on.
Jev, a fast decision model from TypeSafe, reads all 50 candidates in one pass and gives each a probability, plus how sure it is overall. At 40% sure or more, its pick is the answer.
Only when Jev is unsure: an LLM on the LLM reads the top 5, with Jev's leanings, and makes the call in a sentence.