What problem does it solve? When a product team delivers only an HTML prototype plus backend API documentation (no complete PRD), the gap between these two sources hides implicit business behavior—state transitions, field dependencies, orphan references, display-vs-submit mismatches—where most bugs live. AI coding agents tend to fill these gaps with plausible-looking assumptions instead of asking. This Skill forces every gap to become an explicit human decision with confidence labels, producing a single source of truth before coding starts. ## Core Features & Use Cases - Five-pattern gap scanning: Systematically probes entry differences (create vs edit mode), dynamic field candidates, orphan reference cleanup, display-value vs submit-value mismatches, and session draft retention, using a built-in checklist and real-case reference doc. - Confidence-labeled spec template: Ships a ready-to-copy Feature Intake Spec template where every conclusion is tagged [产品确认]/[接口确认]/[原型确认]/[实现现状]/[协议TODO], plus a field-by-state matrix and dependency chains as direct coding input. - Human-decision workflow: Separates business gaps from contract-level TODOs, routes ambiguous items to explicit human ruling, and defines clear boundaries for what AI may implement autonomously. - Use Case: You receive a prototype HTML and an API doc for a new frontend feature with no PRD. Run this Skill to produce a complete intake spec listing every inconsistency, each either resolved with evidence or marked as a specific searchable TODO for the backend or product owner. ## Quick Start Tell the agent: I have the prototype HTML and backend API docs for this new feature—run feature intake and produce the spec with all gaps listed for my review.
