Pratik shr

PhD Candidate (Microarchitecture Security)

[Pratik shr @pratik_shr]

Building SkillTrap with @TheLaxmanKC
AI agent skills can run with powerful access, but many are never tested. SkillTrap runs them in sealed sandbox, watches what they do, and flags risky behavior with clear evidence. No keyword guessing. Just real behavior. Open sourcing it soon