agent-looplisted

# Agent Loop ```text assemble context → act/tool → observe → verify → update state → stop? ``` Required stops: success gate, max turns, no-progress, human escalation. Feed structured errors back into the next act. Do not restart from zero each turn.