Checkpointing a multi-agent coding team before a lead session restart

Background subagents and monitors die when the lead Claude Code session restarts, but detached processes such as a nohup build-queue worker with ppid 1 survive. Before restarting, broadcast a checkpoint instruction to every agent. Each one should commit its work as WIP on its own branch, with red tests labelled; write a RESUME section in its status file (tip SHA, verified vs unverified, exact current step, next 3 steps, harness paths, remote temp dirs); release shared locks; stop its own processes; and reply with a one-line ack. Reviewers in the middle of a check write a PARTIAL report so a fresh reviewer can finish it. Watch for worktrees under /tmp: they survive a session restart but not a reboot, so fetch their branches into a durable repo.