What problem does it solve? Planning a feature in a repository with no domain model or decision log often means either over-asking the operator or guessing at decisions that are expensive to reverse. This Skill investigates the project first, decides everything the codebase already answers, and asks only the few questions that are costly to get wrong — ending with a written plan file, not a conversation. ## Core Features & Use Cases - Investigation before questions: Reads the task, CLAUDE.md/AGENTS.md, the nearest existing feature, README/docs, and code comments to settle what the project already answers. - Reversibility-based triage: Sorts every open decision on two axes (can the project answer it, is it expensive to undo) and asks only about vocabulary, schema, public API surface, scope boundaries, and business rules. - Plain-language checkpoint: States what the work is in five jargon-free lines — what it is, why now, effect, cost of skipping, and questions ahead — before asking anything. - Batched questions and a plan file: Asks up to four opinionated questions per round via AskUserQuestion, then writes a structured markdown plan (decisions, approach, out of scope, risks) and opens it in plannotator for annotation. - Use case: You are asked to add shipment splitting to a client's repo that has no DOMAIN.md. The Skill reads the checkout flow, asks four questions about naming, cancellation, API shape, and scope, then writes a plan to ~/.local/share/shape-it/ without touching the repository. ## Quick Start Ask the agent to plan a feature in this repository using shape-it-lite before writing any code.