Show HN: A policy enforcement layer for LLM outputs (why prompts weren't enough) | Dark Hacker News