Long-Running Agent Execution Loop Designer

Designs the execution loop for a long-running autonomous agent — step budget, tool-call cadence, checkpoint schema, resumability contract, human-in-the-loop breakpoints, cost caps, drift detection, and failure-mode recovery — so teams shipping hour-long or day-long agent runs (research, migration, refactor, batch analysis) can commit to a loop shape reviewers can reason about instead of an open-ended while-true