What problem does it solve? Restarting a fleet of long-running Claude Code agent sessions after a reboot, plugin update, or account switch is error-prone: hand-rolled tmux commands silently drop board identity variables, Discord state scoping, and startup dialog handling, leaving peers that look healthy but cannot write or connect. ## Core Features & Use Cases - Four respawn modes: missing fills gaps after a reboot, plugin cycles sessions lacking the fleet plugin, all resets the whole registry, and running cycles every live session at its own cwd after a /login account switch. - Identity and state preservation: passes CW_AGENT_NAME/FEEDBACK_AGENT_NAME and per-peer DISCORD_STATE_DIR via tmux -e flags, resumes transcripts with --continue, and auto-dismisses startup dialogs. - Team-lead self-respawn: self-respawn.sh cycles the team-lead's own session using a detached, sentinel-armed revival job that aborts safely if the respawn path fails to arm. - Use Case: After switching Claude accounts with /login, run python3 respawn.py --mode running --execute to cycle every live session onto the new account's MCP handshakes without losing worktree transcripts. ## Quick Start Ask the team lead to respawn your sessions, for example say "respawn all agents" or run a dry-run of respawn.py first and then rerun it with --execute for the mode you need.
