deadclaw_skill

Overview

This skill provides an emergency kill switch for OpenClaw agents that immediately halts all agent activity, pauses scheduled jobs, kills active sessions, and logs a full incident report. It supports message triggers, a WebChat kill button, and a phone home-screen shortcut, plus a background watchdog that auto-kills on dangerous conditions. Designed to work with native and Docker-based OpenClaw installs and to confirm actions back to the trigger surface.

How this skill works

When activated, the skill runs a core kill script that stops agent processes, stops or restarts Docker containers, pauses cron jobs, kills sessions, and writes a timestamped entry to deadclaw.log. Activation can be manual (trigger word or UI button) or automatic via a watchdog daemon that monitors runtime, token spend, unauthorized network calls, and out-of-bounds file writes. Status and restore scripts provide health reports and controlled recovery after a kill.

When to use it

• Immediate halt of all OpenClaw activity during an incident
• If you detect runaway agents or unexpected high token usage
• When you suspect unauthorized outbound network access from an agent
• As a remote panic button accessible from phone or chat
• Before performing emergency maintenance or containment

Best practices

• Set up multiple activation surfaces (chat trigger, WebChat button, phone shortcut) for redundancy
• Review and customize watchdog thresholds to match your deployment and risk tolerance
• Keep a current network whitelist and workspace path in environment variables
• Test scripts with --dry-run in a non-production environment before relying on them
• Log and review deadclaw.log after every activation to refine rules and restore procedures

Example use cases

• Nighttime autonomous agents enter a runaway loop and must be immediately stopped
• An agent begins making outbound calls to unauthorized domains — watchdog auto-triggers a kill
• Remote operator presses the WebChat red button to halt all agents during an incident response
• Developer accidentally deploys a faulty skill that rapidly burns tokens; send the chat trigger to stop it
• On-call engineer uses a phone home-screen shortcut to stop agents while offsite