capitalparser/notebooklm-wiki-pipeline ? reverse-engineered prompt

Build me a Python tool that takes a Google Drive PDF and turns it into a clean Obsidian note, but without dumping the full PDF into the model context.

I want the flow to use NotebookLM as the reader, then have the agent pull back only the structured summary it needs to make a reusable wiki style note. It should support reusing one NotebookLM notebook per topic, then when I add a new PDF, it should query only that selected source so the note stays grounded in the right document.

Please include a simple command line way to run it from a Drive link, a small topic registry for routing to the right notebook, and a verification step so it can catch when Drive search grabs the wrong PDF. The output should be Markdown with Obsidian wiki links and metadata like source, notebook, file id, topic, and query scope.

If you need to, look up current docs online for NotebookLM or the MCP setup.