What problem does it solve? Founders get encouragement instead of honest feedback when they ask an AI about a business idea, so weak ideas get built and strong ones go untested. This Skill runs an adversarial evidence diagnostic that grades what the founder can actually prove, separates that from market research, and ends with a verdict, the narrowest testable wedge, and one real-world assignment. ## Core Features & Use Cases - Forcing questions by stage: Asks one question at a time from a stage-matched set (pre-product, users-not-paying, paying customers), pushing past rehearsed first answers. - Evidence grading and verdict: Grades each dimension as evidenced, asserted, or absent, then renders a Validated, Unproven, or Contradicted verdict with quoted founder answers. - Optional market scan: Researches incumbents, pricing, and customer complaints via web search, kept strictly separate from the founder-evidence grade. - Use Case: A founder says "should I build this invoicing tool for freelancers?" The Skill asks about demand evidence, names the actual buyer, checks competitors and pricing online, then delivers an Unproven verdict with the assignment to charge one named freelancer real money this week. ## Quick Start Ask the agent to validate my startup idea and pressure-test whether anyone would actually pay for it.