What problem does it solve? Half-formed proposals, plans, and ideas often contain unstated assumptions, unfalsifiable success criteria, and contradictions that only surface after work has started. This Skill interrogates the user's own proposal one question per turn until it can be stated in terms an outsider could act on, producing a hardened statement of intent or an honest account of where the interrogation stopped. ## Core Features & Use Cases - One-question-per-turn interrogation: Asks exactly one answerable question per turn, classifies each answer (grounded, asserted, deferred, contradiction, unfalsifiable, scope-drift), and adapts follow-ups from that record. - Scope classification with a one-way ratchet: Classifies proposals as probe, bounded, or structural before the first question, and upgrades mid-session when complexity surfaces. - Contamination tracking: Records every assistant-supplied answer the user adopted, including answers picked from assistant-authored option sets, and discloses them in the final report. - Use Case: Before writing a work item for a new API integration, run the interrogation to expose that the acceptance criterion "it works" names no check, that two stated constraints contradict each other, and that the scope quietly grew — then approve a hardened statement with testable criteria. ## Quick Start Ask the AI to grill your proposal or plan, for example: "Grill this plan for migrating our billing service before I write the work item."
