I agree with you, the linting pass seems valuable and it's something I'm thinking about adding - it's a great idea.

What I'm pushing back on specifically is the insistence that the core loop - retrieving the most relevant pieces of knowledge for wiki synthesis - is not RAG. In order for the LLM to do a good job at this, it needs some way to retrieve the most relevant info. Whether that's via vector DB queries or a structured index/filesystem approach, that fundamental problem - retrieving the best data for the LLM's context - is RAG. It's a problem that has been studied and evaluated for years now.

thanks for checking it out