amplifylisted

# Shape a goal A goal is a pointer, not a plan. The computer board owns priority, entity pointers, claims, owners, and resume. The entity's local `PLAN.md` owns the outcome, decisions, checkpoints, acceptance, proof, and anti-sprawl controls. ## Compile 1. Read the board, owning plan, active claims, Git state, and named surfaces. 2. If the plan is not executable, update it with the missing authority, contradictions, acceptance, and reachable checkpoints, then continue into implementation. Stop after planning only when explicitly asked. 3. Preserve the whole requested outcome. Quantify only when a number changes a decision; task, agent, commit, test, and artifact counts are receipts rather than success. Choose proof that would fail a plausible shallow result. 4. Leave the one standing Shadow goal unchanged and skill-free. For a specific goal, emit at most 80 words across four one-line sentences: ```text Outcome: <finite end state and what the person stops supervising>. Resume: shadow status --by <seat> + PLAN.md at committed ref; continue owned claims, then the highest-value reachable checkpoint; delegate bulk to the cheaper roster when a delegate harness is installed, else run native and record the fallback. Proof: <mechanical finish line>; keep source, merge, install/deploy, and live distinct. Skills: <one to four canonical invocation names from the current session's available skill catalog>. ``` `Resume` always names Shadow by its command and always carr