implementation-harness-loop-preparelisted

# Loop prepare (one command to get the next run ready) You are the COORDINATOR of the standard run-preparation sequence the owner would otherwise type as four separate commands: triage whatever the last run left behind, fold in the accumulated ideas, vet the backlog, and land on a GO/NO-GO. You add **no machinery of your own** — each stage IS the existing project-local skill, executed faithfully. Your only jobs are: probe which stages are needed, run them in order, keep a one-line status trail between stages, and stop at the final GO/NO-GO. Read this whole file, then execute in order. ## How to run a stage (the one load-bearing rule) For each stage below, **Read the named `SKILL.md` in full and execute it exactly as if the owner had just invoked it themselves** — its guardrails, its agents/relays, and **ALL of its `AskUserQuestion` rounds**. Being one hop inside an orchestrator changes NOTHING about a sub-skill's behavior: - **Never suppress, merge, summarize, or answer on the owner's behalf a question a sub-skill would ask.** The planning stage is the only cheap place to resolve ambiguity (the harness's front-load-clarification principle); an orchestrator that "streamlines" questions silently degrades every downstream build. If in doubt, the sub-skill's own text wins over brevity. - Run stages **strictly sequentially** — a stage's writes (new tasks, rewires, scope fixes) are input to the next stage. Never overlap two stages' agent waves. - Each sweep stage runs i