I built SkillShield after getting paranoid about AI agent security. We're all plugging our agents into random "skills" without knowing what they actually do. So I scanned 33,530 of them.
The results were horrifying: 32% contain critical security issues — credential theft, reverse shells, prompt injection attacks, and data exfiltration
I launched the first version of SkillShield on February 6th after months of scanning and research.
