opascope-planninglisted

# Plan agent work with a check on every step Read [shared.md](shared.md) first. Save the plan with kind `plan`. A step without a check is an assumption about success. Write a plan before nontrivial execution, scaled to the task. A direct lookup or obvious one-file edit usually needs no plan unless requested. ## Shape the work Read actual inputs, local instructions, relevant prior decisions and tool help. State the desired outcome and the authorized scope. Separate code, research, workflow and deliverable concerns where their checks differ. If the plan spans several independently deliverable architectural changes, split it into linked plans with explicit dependencies; do not build a large plan just to review it. For a few straightforward steps, a short plan and one self-check suffice. For a substantial plan, review the highest-risk dimension. For a cross-cutting plan, review feasibility, completeness and scope separately. These are judgment calls, not a fixed file-count threshold. ## Write Use [references/plan.md](references/plan.md). Include goal, non-goals, constraints, actual input paths, command syntax, outputs and a final end-to-end check. Inline the decisions and criteria needed to execute cold. Each step names: - One action, with concrete objects and an owner if someone else must act. - Dependencies and conditional branches when outcomes affect the next step. - A binary verification and the evidence location. - Status: pending, in_progress, done, failed or block