I’m normally selecting snippets small enough that they don’t invite spoilers. I think the plugin gives the book author/title but I don’t know the full prompt (I should look). So far I’ve not hit any spoilers. For all I know the prompt says something about avoiding spoilers.

I’ve thought for a while about writing something that only gives the model access to the book as far as you are. Maybe some king of RAG but with page numbers associated so you can filter “< $currentPage” and being able to ask questions “who is this again?”, “how are they related to X?”. The “world knowledge” is the main thing I’d have to be wary of. It’s going to know Harry is a wizard before you get to that part in the books (albeit that’s early, just making a point).

For the Horatio Hornblower series it was mostly older English phrases that had the feel of a “saying” but I couldn’t be sure of the meaning. Getting a response like “in the late 1700’s this phrase was used to mean…” was really cool, better than having to pull out my phone or just guess.

This [0] may or may not interest you, yes it’s vibe coded but I wanted a way to hook BookOrbit into my Omni-Claude [1] instance.

[0] https://github.com/joshstrange/bookorbit-mcp

[1] This isn’t a “thing”, just Claude code on the CLI that has access to my obsidian notes, home assistant, Unifi, bookorbit, etc. My own, very limited “Claw”. “Speaking” home assistant automations into existence is a whole other topic I could on and on about.