Ask HN: Does piping LLM output into a RAG stack sound like a good idea?

I would like to know if this can improve agent memory or adherence to coding guidelines and policies in large codebases over time.

I'm also interested in this as a general concept.