subagent-overseer
oleh globalcaos
Web & Frontend DevelopmentRingkasan
Monitor sub-agent health and progress via a pull-based bash daemon.
Deskripsi
Apa yang Dilakukan Skill Ini
A bash daemon that monitors sub-agent health without consuming AI tokens. It reads /proc filesystem data and calls openclaw sessions list on a configurable interval, writing results atomically to a JSON status file. The main agent reads that file on each heartbeat instead of making live tool calls.
Replaces in-agent polling loops with a zero-token OS-level daemon, so heartbeat cost stays flat regardless of how many sub-agents are running.
Kapan Menggunakannya
- Detecting a stuck sub-agent in a parallel coding pipeline
- Monitoring filesystem activity across multiple concurrent agents
- Getting local voice alerts when an agent stops making progress
- Checking gateway process health without spending context tokens
- Auto-stopping the monitor once all sub-agents have finished
Contoh Alur Kerja
User asks: watch my 4 sub-agents and alert me if any get stuck
- 1
Start overseer.sh in the background with --workdir, --interval 180, and --max-stale 4
- 2
Overseer polls /proc for gateway health, runs openclaw sessions list, and checks find -newer each cycle
- 3
Overseer writes completed JSON atomically to /tmp/overseer/status.json
- 4
On each heartbeat trigger, agent reads status.json with a single cat call
- 5
If any agent shows stale count >= 4, report that label as stuck and offer to kill it
status.json shows subagent 'worker-2' as stuck after 12 minutes of no filesystem changes
Persyaratan
- jarvis command (local TTS via sherpa-onnx) for --voice alerts
Instalasi
clawhub install globalcaos/subagent-overseerJalankan perintah CLI-nya, atau tempel prompt ke Claude Code, Cursor, Codex, atau coding agent apa pun yang Anda pakai.