Design Patterns for Securing LLM Agents against Prompt Injections

26 points by simonw