jjunsss

Use for live multi-agent feedback loops where executors and reviewers stay in standby, discover loop_station signals, exchange evidence-backed feedback, maintain compact-ready rolling summaries, and move to the next session only after required signals and artifacts exist. First lock what to improve, budget, evidence, and optional role/safety rules; then run bounded adaptive sessions without open-ended retry. If invoked as a reviewer before equivalent executor-complete and review-ready signals exist, enter standby and use any available monitor/background watcher tool instead of editing files or writing a premature review.