Before a user runs a skill, they read its name. That’s the entire interface. One word, maybe two, to shape what they bring to the interaction and what they expect to take away.
Name a skill thesis-generator and they bring a task. They leave with a document. Nothing accumulates between uses. The skill stays external — a vending machine they return to when they need the output again.
Name it thesis-thinking and something different happens. The gerund signals ongoing development. They bring a question. They watch how the agent works through it — what it asks first, what it checks, how it builds the argument. Over weeks, they start doing that sequence themselves before they open the tool. The skill taught them the workflow.
The name is the first lesson. It decides whether the user is a consumer or a learner before they ever see the instructions.
This is why design systems use three tiers: primitives (thesis-thinking), semantics (thesis-before-calls), components (review-proposal-with-rigor). Most skills libraries stop at the component tier — task-specific, artifact-bound, useful but not memorable. The primitive and semantic tiers are where behavior change lives. Skills named after mental models generalize. Skills named after outputs don’t.
One audit you can run today: for each skill in your library, ask whether the name makes someone a practitioner or a user. The practitioner builds capability over time and eventually applies the frame without the tool. The user returns only when they need the output again.
Your skills library is building one or the other. The name is where that choice gets made.