prd-taskbreakerlisted

# PRD Taskbreaker Raw idea → spec-driven PRD → numbered tasks an AI agent can build **without over-engineering**. The binding rule: **every task has exactly one primary accepted requirement; any other requirement or quality IDs are constraints.** A task with no primary requirement is YAGNI — cut it. Requirement acceptance criteria, task completion, test procedures, and observed evidence remain separate (see Verification vocabulary). ## Flow ``` [Idea] → [0. Inspect authority, repository, and existing flow] → [1. Clarify — GATE, resolve material ambiguity] → [2. PRD.md — context + goals + non-goals + numbered requirements] → [3. UX/design handoff — only when browser-visible] → [4. PLAN.md — ONLY if it touches architecture/DB/integration] → [5. Review contract and real alternatives] → [6. TASKS.md — each task → one primary requirement, constraints, deps, Done-when] → (per task) implement → run its TEST → record observed EVID → commit additively ``` ## Modes - **`prd`** — PRD only, from a description - **`plan`** — technical PLAN from an existing PRD (architectural features) - **`tasks`** — break an existing PRD/PLAN into tasks - **`full`** *(default)* — clarify → PRD → (plan if needed) → tasks in one run - **`update`** — update existing artifacts; preserve every accepted requirement and task ID (never renumber or reuse it); append new IDs and explicitly supersede changed accepted items ### Suite-pack mode If a `developmen