next

When the user invokes /vaner:next, surface the most-ready next candidates Vaner has prepared:

1. Call mcp__vaner__suggest with a small limit (default: 3, cap at 5) to fetch the top intent suggestions. If the user passed arguments to /vaner:next, use them as a focus hint (e.g., /vaner:next auth → pass focus: "auth").

2. If a top suggestion looks actionable, also call mcp__vaner__resolve for it to pull the prepared context package (capture the resolution_id for later mcp__vaner__feedback).

3. Render the results as numbered candidate cards, one per suggestion, with three pieces:

   • Label — short, imperative phrasing of the candidate (e.g., "Trace the auth middleware chain", "Review the cockpit scenario cluster").
   • Why now — one line on what makes this candidate relevant in the current state (recent edits, open scenarios, unresolved decisions, signals Vaner is tracking).
   • Readiness / confidence — one line on how prepared Vaner is for it: confidence score, cache tier, or evidence count if mcp__vaner__resolve was called; otherwise the vaner.suggest score alone.

   Example format:

   1. **Trace the auth middleware chain**
      Why now: edits in src/auth/ today; Vaner has a prepared scenario covering session-token flow.
      Readiness: confidence 0.82, 7 evidence items, cache tier: warm.
      Pick: `/vaner:next 1` or ask "help me with #1".
   

4. After rendering, ask the user to pick a number (or describe a different task). If they pick, use the captured resolution_id to continue — do not re-call vaner.resolve for the same candidate.

Do not render raw prediction dumps or invent readiness information Vaner didn't supply. If mcp__vaner__suggest returns no candidates, say so plainly and suggest running vaner up or checking mcp__vaner__status for daemon readiness rather than fabricating suggestions.

This skill is a structured presenter for predictions; it is not a planner. Skip it entirely if the user is already mid-task or has given a concrete instruction.