According to @openclaw, scans of 67,453 ClawHub skills show 0.31% malicious and under 8.5% agreement across tools, signaling risk taxonomy gaps.