Agent skill wrapper for the Claude command
.
When the original command text references
,
, or named arguments, map them from the user's current request.
Systematically resolve underspecified areas in a feature spec through structured questioning. Run before
.
-
Locate the spec:
- If input names a feature or file path, use it.
- Otherwise check ; if multiple, ask which to clarify.
- Fall back to in the project root.
-
Load context: Read
and
.specify/memory/constitution.md
if present.
-
Scan for ambiguities across nine taxonomy categories:
- Functional scope, Data model, UX flows, Non-functional attributes, Integration points, Edge cases, Constraints, Terminology, Completion signals.
-
Select questions: Choose the 5 highest-impact ambiguities. Prioritize by: scope > security > UX > technical detail. Each question must materially affect architecture, data modeling, testing, UX, or compliance.
-
Ask one question at a time:
- Present the question with 2-5 multiple-choice options (or note that a short phrase is acceptable).
- Mark the recommended answer with reasoning.
- Wait for the user's answer before asking the next question.
-
Update spec incrementally: After each answer, incorporate the clarification naturally into the relevant spec section. Do not append a Q&A log — merge it into the prose.
-
Final validation: Check for duplicates, contradictions, and terminology drift across the updated spec.
-
Report: Coverage map across all nine ambiguity categories (resolved / deferred / clear / outstanding).
Next step: Run
to generate the technical implementation plan.