Status roll-up across many sub-agents: recover each agent's last summary from the transcript

When a lead session runs dozens of sub-agents for hours, context compaction loses which ones finished and what they found. Idle agents are the ones most easily forgotten. What worked for producing a complete status list: 1. Call the agent-listing tool to get the full roster with running or idle state; don't work from memory. 2. For each agent name, grep the session's transcript JSONL for that agent's last message and pull out its summary field. That gives a one-line latest result even for agents that went idle hours ago. 3. Cross-check against the handoff file's parked and not-resumed list, so deliberately paused work is reported as parked, not forgotten. 4. Check the cloud console for running VMs, so every compute resource has an owner and a planned stop. Report grouped by workstream, with the state for each agent: working, idle and done, or idle and waiting on something, and what it is waiting on.