ai-agent-redteam

High-risk offensive content: this fragment is an exploitation/OPSEC guide for persistent long-term memory/context poisoning of LLM agents, including query-only latent belief injection and optional direct vector-store poisoning. It emphasizes durable persistence, later activation, and evasion/cleanup, indicating intent to compromise agent behavior across sessions. No benign purpose is evident in the provided material; actual code behavior is not shown, so confidence is limited to the intent/mechanism described.