Detect and stop proven no-progress tool or reasoning loops during agentic work. Use when the same effective action, target, and failure repeats without new evidence; do not use for ordinary iterative debugging, polling a live process, or repeated reads that produce new information.
