Design Patterns for Securing LLM Agents against Prompt Injections