openspec-clarify

Clarify ambiguities in change artifacts through structured questioning. This is a passive skill — invoke it when you detect unclear requirements, vague language, or contradictions while working on a change.

Trigger: When you encounter ambiguities while creating artifacts (during /openspec-continue or /openspec-ff), or after analyze flags ambiguity/gap issues.

Input: The change name (from context) and optionally which artifact has the ambiguity.

Steps

1. Identify ambiguities in the current artifacts (up to 5, ranked by importance):

   High importance (ask first):

   • Requirements without scenarios (### Requirement: with no #### Scenario:)
   • Contradictions between artifacts
   • Explicit markers: TBD, TODO, NEEDS CLARIFICATION
   • Missing scope boundaries (no Non-Goals)

   Medium importance:

   • Vague language in specs: "should", "may", "might", "consider", "as needed"
   • Undefined terms or concepts
   • Design decisions without rationale

   Lower importance:

   • Missing error/edge case scenarios
   • Implicit assumptions

   If no ambiguities: skip this skill silently — do not mention it to the user.

2. For each ambiguity, ask ONE question at a time using AskUserQuestion:

   • Clear question about the specific ambiguity
   • 2-3 options including AI recommended answer (marked "(Recommended)")
   • Description on each option explaining impact on artifacts

   Example:

   Question: "Auth spec says users 'should' be logged out after password change. Mandatory or optional?"
   Options:
   - "Mandatory (MUST)" (Recommended) — Safer, forces logout on all sessions
   - "Optional (MAY)" — More flexible, let implementation decide
   - "Configurable" — Add a setting, most flexible but adds complexity
   

3. After each answer, immediately update the artifact with clarified wording. Show briefly: "Updated specs/auth/spec.md: 'should' → 'SHALL'"

4. After all questions (or user says stop), show a brief summary:

   Clarified N ambiguities:
   - specs/auth/spec.md: session logout → mandatory (SHALL)
   - design.md: added Redis rationale
   

Guardrails

• Maximum 5 questions per invocation
• One question at a time
• Always provide a recommended answer
• Respect user's choice — never re-ask
• Minimal targeted edits when updating artifacts
• Do NOT create new artifacts — only update existing ones
• If user seems impatient or says to skip, stop immediately
• Keep it lightweight — this is a mid-flow check, not a separate workflow