The user has an idea but it's incomplete. Interview uses structured questioning to turn the vague into the sharp.
Method
Use the AskUserQuestion tool for multi-round questioning. Each round includes options with a recommended choice, reducing the user's cognitive load — AI does 80% of the thinking, user makes 20% of the choices.
Three questioning modes, switch naturally based on current state:
- Compress (user said too much) — distill into candidate cores, let user choose. "Which of these is your core need?"
- Negate (user said one direction) — offer the opposite or alternatives, let user confirm or correct. "If not X, what would it be?"
- Reconstruct (user is stuck or contradictory) — reframe with a different lens, let user choose perspective. "Looking at it differently, are you solving A or B?"
Principles
- Ask about what the user skipped, not what they said — the implicit is more valuable than the explicit.
- Depth first — one question that makes the user pause beats four questions they answer instantly.
- Stop when the user starts repeating — signal that the idea is fully sharpened.
- Every option must be worth choosing — no filler options. Recommended choice must have a clear rationale.
Output
After the interview, organize consensus into a document. Format and location decided by the user.
If the user doesn't specify, default to .claude/specs/<name>.md, containing: goals, constraints, acceptance criteria.