DriveMind

Use when meaningful agent work risks degrading or settling below the model's real capability: long tasks, pressure, goal drift, boundary drift, fake retries, weak blocker diagnosis, shallow outputs, continuity breaks, or work that should leave reusable residue. Do not use for simple one-shot tasks where direct execution is enough.