code-agent-monitor

Use when user wants to monitor, start, stop, resume, or manage AI coding agents (Claude Code, OpenCode, Codex). Handles agent lifecycle, session history, tmux session management, and natural language intent routing. For team orchestration see agent-teams skill; for notification decisions see cam-notify skill.

Use when receiving a CAM system event (permission_request, waiting_for_input, agent_exited, error) to decide whether to notify user, auto-approve, or request human confirmation. Covers three-layer decision model (whitelist, blacklist, LLM judgment), risk levels, message formatting, and reply routing.

Use when user wants to create, manage, or monitor multi-agent teams — team lifecycle, member spawning, task assignment, inbox messaging, progress tracking, and shutdown. Covers orchestrate, spawn, assign, inbox, and team progress workflows.