AI agent skills pass every scanner. 87% still degrade agent safety(faberlens.ai)
faberlens.ai
AI agent skills pass every scanner. 87% still degrade agent safety
https://faberlens.ai/blog/skill-safety-problem
4 comments
have been looking into a skill repo that is "safe". If I want to drop one of these hardened skills into production, what's the failure mode? are you also exploring if the default guardrails ever block " legitimate" workflows?