Scope to Ship — one agent, aligned first, dogfooded to done

Run a feature from fuzzy idea to dogfooded, shipped code with ONE agent and a front-loaded alignment phase — instead of fanning work out across parallel sub-agents. The loop: discuss + scope (surface a couple of design options, lock the simplest-yet-most-correct-long-term direction, answer ≤5 high-impact questions), crystallize the decision into a tiny numbered ADR under docs/adr, execute to completion against a Master PRD, then dogfood the result end-to-end before reporting it done. Use this when the user wants to scope or design a feature before building, align on architecture, write or follow an Architectural Decision Record, build a non-trivial feature end-to-end in o…