Ground a review in verified facts

Ask an agent to imagine how a design would be used and it will produce the median answer about that job title: fluent, specific-sounding, and untethered. Ask twelve agents and you get twelve variations of the same invention, which reads like corroboration and is not.

The fix is not better prompting. It is giving them facts somebody checked, so their reasoning has something to be wrong about.

This skill produces that briefing. It also does the harder thing: it refuses to certify one when too little can be verified, because a disciplined-looking review built on an empty briefing is worse than no review — it wears the costume of evidence.

Composition: delphi-imagine consumes the briefing this skill produces, and its insufficient verdict is what triggers that skill's refusal rule.

Quickstart