Guides agents through using large language models and retrieval-augmented generation safely on health content — controlling hallucination, grounding claims in sources, preventing PHI leakage and prompt injection, evaluating outputs for faithfulness, and keeping a human accountable. Use when an LLM will summarize, answer, extract from, or generate any health-related content, when building a RAG system over clinical or medical documents, or when an LLM sits inside an agentic tool with access to data or actions. Use even when the output "sounds authoritative" — fluent and correct are different properties, and in health the gap between them is harm.
