A Run Ledger Makes Long-Running AI Work Easier to Resume

A long-running agent task can fail without crashing. A scheduled run may never start, a browser action may wait for approval, or a public action may finish while the session loses its place. A transcr