53,577 Skills, 946 Threats: The First Large-Scale Security Audit of the AI Agent Ecosystem

The largest AI agent security scan ever conducted. We scanned 53,577 skills across OpenClaw (50,485) and Skills.sh (3,115) using 108 ATR detection rules. 946 skills flagged (1.77%), 875 critical, 52 high, 19 medium. 0% false positive rate. Tool description poisoning accounts for 71% of detections. Average scan latency 5.39ms per skill. Open source: github.com/Agent-Threat-Rule/agent-threat-rules