dx-distill

Establish the context later skills would otherwise guess at. Help the user make the decision in front of them, then record its basis — scope, rules, open risks — as context/foundation/briefs/<slug>.md, where every claim carries its source. dx-brainstorm decides whether one idea is worth building; this establishes or refreshes the ground it stands on.

Writes that file and stops. It creates no container and promotes nothing — dx-new stays the router.

Never invent research, user behaviour, quotes, or measurements. A missing fact stays unknown or becomes a named hypothesis. A human's choice to build on an untested assumption doesn't convert it into evidence; a decision with no recorded confirmer stays proposal. Material the current decision actually needs goes on the collection plan — an irrelevant empty section needs no research task.

Guards. context/foundation/ absent → say to run /dx-init, stop. No user available → stop; this is a conversation, and the appetite, constraints and experience it exists to ask about are exactly what you must not invent. Every path named must resolve; an unresolvable one is asked about, never guessed at. An existing <slug>.md is a refresh: keep its ids and its history — a changed decision gets a superseding row, it doesn't overwrite the old one.

1 — Frame the decision, and the emphasis with it

Read enough to state what this session should help decide. Unclear → ask what the brief is for; a brief written toward no decision is a filing exercise, and its sections will be padded to look complete.

In the same exchange, name the emphasis you infer from what the material actually is — the way type already activates plan characteristics without the user naming a mode: