What problem does it solve? Plans and designs often look complete while hiding unresolved decisions several layers deep, leading to wrong assumptions baked into implementation. This Skill closes the gap between "the agent thinks it understands" and "the agent actually understands" by systematically interviewing the user until every decision branch is locked. ## Core Features & Use Cases - Design tree mapping: Converts a plan or conversation into a tree of unresolved decisions and works the current frontier round by round, batching up to 4 independent questions per ask. - Recommendation-driven questions: Every question ships with a recommended option marked first, so the user can confirm quickly instead of deciding from scratch. - Fact resolution before asking: Structural code questions are resolved silently via graphify queries and direct file reads, so only genuine intent and preference questions reach the user. - Incremental decision capture: Locked decisions are recorded after each round into .planning/decisions/ or a plan folder's decisions.md, so a crashed session loses at most one round. - Use Case: After drafting an implementation plan with /recon, run this Skill to surface hidden choices (naming, error handling, data shape), lock each one with a rationale, and update the plan before writing any code. ## Quick Start Ask the agent to interview you about the current plan, for example: "Interview me on this plan and lock down every open decision before we start coding."