hunt-cache-poison

This skill is not overt malware and shows no credential harvesting, hidden exfiltration, or suspicious installer chain in the provided text. However, it is a high-risk offensive security skill for AI agents: its stated purpose is to probe and potentially exploit cache-poisoning vulnerabilities on real targets, with workflows that can cause broad real-world impact if misused. The authorization gate and lab-focused caveats reduce concern somewhat, but they do not change that the capability is inherently dangerous and beyond a normal benign developer-assistance skill.