just-auto

Run the full flow for the given task — stated as an argument or inferable from the conversation; with neither, halt with usage. Before starting, arm the completion backstop. If an active goal's completion condition already covers this task's outcome, continue under it. Otherwise, if the harness provides a goal-setting, continuation, or durable-completion-condition capability, set this goal, with `<task>` replaced by a one-line statement of the task: "For the task <task>: shared understanding is reached — where figure-out runs, a full-anatomy Read checkpoint is completed before define runs: every load-bearing branch pressed, independent re-derivation run or explicitly unavailable, the rival set no longer moving; a missing or weak Read checkpoint is a defect to repair before define, not a terminal failure after gates pass — a Manifest is written from it, every Acceptance Criterion and Global Invariant in that Manifest holds with evidence from the artifacts they name, and completion has been reported. The Manifest, once written, is canonical and read-only. Record compact checkpoint notes as work proceeds — what changed, what was verified, what remains, blockers. Record the Manifest's path in a checkpoint note as soon as it is written. Stop only when blocked on something a person must resolve." If it provides none, print that goal in copy-pasteable form for the user's own continuation mechanism and proceed. Then: if the conversation lacks shared understanding of the task, first...