monitor-launcher

monitor-launcher — mon <name> + the monitors.toml registry

mon <name> [args] launches any registered monitor in a reusable, named Windows-Terminal window (with a plain cmd /c start console fallback). It is the generalized successor to the old wfmon-only launcher; wfmon and agentmon are thin aliases that delegate to mon.

All the real logic — parsing monitors.toml, matching a monitor's trigger_tool to a tool, and building the launch argv — lives in the 100%-gated monitorcore.monitors core (vendored inside the sibling workflow-monitor skill). The .cmd files here and the monitor-autolaunch hook are thin shims over it.

Files

• mon.cmd — puts the vendored monitorcore on PYTHONPATH and runs python -m monitorcore.monitors %*.
• agentmon.cmd — mon agentmon alias (per-agent tail terminals).
• monitors.toml — the registry. One [table] per monitor: dir (deployed skill dir), entry (launchable script), trigger_tool (the Claude tool that auto-opens it via the hook; omit for launch-by-hand-only).

Run

mon wfmon           :: whole-fleet workflow cockpit (skills/workflow-monitor/workflow_monitor.py)
mon agentmon        :: one live tail terminal per active subagent
mon codexmon        :: live board of OpenAI Codex threads (relay streams + ~/.codex rollouts)

PATH / deploy

scripts/sync-skills.ps1 copies this whole dir to ~/.agents/skills/monitor-launcher/ (a file copy — the monitors.toml sync mirrors hooks/mcp-userscope-sync/heal.py, never a TOML rewrite). Add ~/.agents/skills/monitor-launcher to PATH so mon / agentmon resolve; mon.cmd finds monitorcore via the sibling workflow-monitor skill, and finds the registry at ~/.agents/skills/monitor-launcher/monitors.toml (override with MONITORCORE_MONITORS_TOML).

Add a monitor

Add a [table] to monitors.toml with dir / entry / trigger_tool, and (if it should auto-open) ensure the outer PreToolUse matcher in hooks/_dispatch/manifest.json covers the new trigger_tool. No code change — the loader/matcher/builder are generic.