spec-interview

# `spec interview` — deep requirement extraction CLI at `scripts/specctl`. Extract complete requirements through structured Q&A. Produces PRD-quality `REQ-*.md`. Input is an idea text, a doc path, or an existing `REQ-id` to refine. ## Step 1: Load context - If input is a `REQ-id`: run `scripts/specctl show <REQ-id>` and read the file. Interview to refine. - If REQ-id not found: tell the user "REQ-<id> does not exist. Run `spec-status` to list existing requirements." Stop. - If input is a file path: read it. Use content as starting context. - If the file is unreadable: tell the user the path and ask them to re-check it. Stop. - If input is idea text: start fresh. - If empty: ask "What feature or requirement would you like to explore?" ### Before questions: load durable project context Read when present: - `CONTEXT.md`, `CONTEXT-MAP.md` — domain language - `docs/adr/*.md` — architectural decisions - `.out-of-scope/*.md` — rejected enhancements If the new requirement resembles an `.out-of-scope/` record, surface it: "This resembles a previously rejected concept because <X>. Reconsider, narrow scope, or stop?" Use the domain vocabulary from `CONTEXT.md`. If the user uses an overloaded term, resolve it before writing the REQ. ## Step 2: Interview process Ask one question at a time using the runtime's multi-choice / structured-question mechanism. Do not output questions as plain prose paragraphs — use the structured question facility so answers are clean to parse. Pl...