Show HN: Vaultak – Security for AI agents (built before the breaches started)

This is really good work and the problem is real. I read Synk's toxicskills report that scanned 3900+ public skills and found that 13.4% had critical security issues, 8 were confirmed malicious and publicly available, 36% contained prompt injection.

I'm more curious about the false positives. How do you guys handle that?