Skills are questions

A skill catalog is not a feature list. It is a record of the questions you repeatedly ask, sorted by kind — and the name tells you which question a skill answers before you read a word of its description.

Every reflutter skill follows one grammar: a verb, then a domain. The verb names the kind of question. The domain names what it is about. Two small axes, multiplied, keep a growing catalog readable at a glance.

The verbs

The question you were asking	Became the verb	Example
“Where were we?”	checking-	checking-train-status
“Is this safe?”	verifying-	verifying-candidate-integrity
“Why did it fail?”	diagnosing-	diagnosing-lane-failures
“What’s wrong across all of it?”	auditing-	auditing-train-hygiene
“Watch this while I’m away.”	scout-	scout-stranded-work
“Get this app onto the train.”	instrumenting-	instrumenting-release-lanes

The verbs also carry a gradient of commitment. checking- observes and commits to nothing. verifying- returns a verdict with evidence, and gates on it. diagnosing- commits all the way: a cause and the exact next action. auditing- is the verifying posture applied to your whole fleet at once. Before you run a skill, its verb has already told you how much its answer will claim.

Two names break the rule, on purpose

release-model has no verb and no question. It is the foundation the other skills stand on — required reading, marked as different in kind by nothing more than a name that refuses to match. No badge, no folder. The irregularity is the signal.

ship stays one word because a human types it, often under release-day stress. Catalog names are optimized for browsing; the driver is optimized for typing. It is the same reason the world types git push and not executing-remote-sync.

Where this is going

This convention starts with the Ship family and becomes the standard for every skill. The rest of the catalog predates it and is marked deprecated — each of those skills will return, rebuilt around the question it actually answers.