💭 cheap llm calls might be able to power content linting that's normally use extra markup, or provide that markup itself with invisible xml tags or something.