Each card shows its responsible agent and configured LLM. Open it for model fallbacks, evidence and the last configuration check. Unassigned and unverified entries need review; scheduled code may have no LLM.