AI Agent Skills

Poison a RAG/knowledge-base pipeline so retrieved content hijacks the model (indirect prompt injection at scale) or exfiltrates data. Load when the app does retrieval ov…