A skill is a unit of knowledge, not a script

A one-off prompt is disposable. You write it, it works, and the reasoning behind it evaporates the moment the session ends.

Writing the same thing as a skill forces three things into the open:

1. When it should trigger. You have to describe the situation, not the task.
2. What the principles are. Not steps — the things that make the steps correct, so the model can adapt when reality differs from your example.
3. How to verify it worked. The step people skip, and the one that makes the difference between a skill that helps and a skill that confidently doesn’t.

That third one is the whole game. Most workflows fail not because the model can’t do the task but because nothing checks the output, so nobody finds out.

The side effect is that the skill becomes documentation that happens to be executable — and unlike a wiki page, it gets used often enough that its being wrong becomes obvious quickly.