What problem does it solve? Feature ideas often get turned into specifications drafted by an AI and rubber-stamped by a person, so the document encodes guesses instead of real requirements. This Skill prevents that by interrogating the requester in staged rounds, using Gherkin's grammar to decide which questions must be asked and when the specification is structurally complete. ## Core Features & Use Cases - Question-driven elicitation: Seven staged rounds (restate, ability header, rules, scenarios, steps, decision gate, emit) extract knowledge from the person rather than proposing content for approval. - Structural completeness checking: The spec_holes.py script scans the .feature file for rules without examples and examples without steps, exiting non-zero until every gap is closed or explicitly deferred with a chosen fallback. - Dual output artifacts: Produces a .feature behavioural contract and a spec.json manifest (scope, constraints, gate verdicts, deferred decisions) that the task skill consumes for planning, plus an optional HTML rendering via render_spec.py. - Use Case: A product owner says "spec out the export feature". The Skill asks who the actor is, what rules must always hold, and what scenarios prove each rule, records every answer, forces rulings on open points, and emits a verified specification ready for implementation planning. ## Quick Start Ask the agent to turn your feature idea into a task specification, for example: "Create a task spec for adding CSV export to the reports page."
