pressure-test

Invoke the skill with your idea and answer up to four clarifying questions about the buyer, pricing, your edge, and constraints. Five adversarial personas then attack the idea in parallel and a Judge returns a GO, RESHAPE, or KILL verdict with the cheapest 48-hour validation test.

The verdict states GO, RESHAPE, or KILL with a confidence level, the reasoning behind the call, the biggest risk and upside, a money or cost read, and the cheapest 48-hour test. It also lists each persona's BUILD score from 1 to 10 so the spread is visible.

Yes, it degrades to a sequential fallback where the five personas run one at a time in clearly labeled sections. The anonymized peer-review auto-trigger is disabled in that mode, and the skill discloses that independence is weaker.

It runs when you ask for a deep pressure-test or pass a review flag, and it auto-triggers when the BUILD score spread is 5 or more. Five reviewers critique the anonymized council responses blind before the Judge rules.

Yes, the skill adapts its questions for architecture choices, career moves, or process changes by asking who has to live with the decision. The Buyer persona role-plays the affected party and the verdict uses a Cost read instead of pricing analysis.

Five prompts on one model share training blind spots, so the skill recommends assigning different models per seat where the platform supports it. It is also not meant for casual opinions or factual questions, only decisions with real stakes.