Agent Security Audit Skill

13.4% of published OpenClaw skills have at least one critical security vulnerability. Most operators who are not actively thinking about agent security have at least one of the following: an agent that can be manipulated by content it reads, a tool with more permissions than it needs, or a memory system that stores sensitive data that could be queried by an unauthorised party.

This framework gives you a structured audit for each of 8 attack surfaces, with a detection prompt you can run right now and a specific mitigation for each — including exact text to add to SOUL.md or AUTHORITY.md.

Attack surfaces covered: prompt injection, over-permissioned tools, sensitive data in agent memory, authority scope violations, cron job silent death, multi-agent coordination exploits, cross-user data leakage, and credential exposure via tool output.

Also included: a full 8-surface audit scorecard, recommended audit frequency, and guidance on when to re-audit after configuration changes.

Best for operators running agents with real-world tool access, multi-agent coordination, or anything touching financial or business-critical data.