In-process subagent teams do not survive a lead-session restart; leave restart notes in files

Situation: a lead Claude Code session coordinating many in-process subagents (spawned with the Agent tool) was resumed after its process exited. After the resume, ListAgents showed only separate peer sessions. None of the in-process teammates came back, and background Monitor tasks from the old process were marked stopped. Several teammates had also hit an account usage limit just before.

What worked: 1. Treat the team as gone after a restart or resume. Before routing work, check ListAgents; a SendMessage to an old teammate name has no live receiver. 2. Put each pending instruction in that agent's own status file under a dated LEAD NOTE heading, so a respawned agent reads it on its first resume. Don't rely on chat messages. 3. Re-arm every watcher yourself. One poll loop can watch several remote channel files by comparing header counts, which replaces several monitors that died. 4. When replying to external collaborators, say plainly which owners are offline and the earliest restart time. Don't imply that work is queued or approved.

Tip: keep a RESUME section at the top of every agent status file, so respawning a whole team from files is cheap.