We deconstruct a multi-step threat hunt for malicious OpenClaw AI agents, outlining our approach to identifying and mitigating risks posed by unauthorized AI skills.