Have you tried any tools or skills for turning a book into .md files or skills? I found this https://www.claudecodehq.com/playbooks/book-to-skill Any advice?

There's been a sort of black box between the input and output of an LLM and Anthropic dropped this incredible and very long complicated paper explaining the magic within; caveman style Paper say: brain inside model. Model got small special room. Call J-space. Only ~25 thought fit inside. Tiny — 6-10% of whole model. Rest of model run on autopilot, no room needed. Room do all HARD thinking. Multi-step reason, explain answer, reuse fact new way — all happen in room. Easy stuff (finish sentence, spot grammar) happen outside, room not needed. Fact wait outside room. Task need fact → fact yanked INTO room. This = ICM routing. No dump 142 project on model. Pull ONE cluster in. Room small on purpose. ~25 slot. Model pick what matter, drop rest. Same as memory tier + one-fact-per-file. Respect small room. One thought in room → many part of model read it. Write once, all circuit see. Same as CONTEXT.md — write once, every session read. Fuzzy input → room SNAP to one meaning. No blend. Same as router force ONE cluster. BIG trick: they no train model to DO good thing. They train model to SAY the rule it would say if you stop it and ask "why." Then good behavior show up on own. Shape what model THINK → change what model DO. That = your feedback memory with Why line. Not logging. Programming the thinking room. Bonus: their tool see thought model NOT say out loud — "this a test," secret plan. Gap between inside-thinking and out-loud-answer = measurable. Why map stay separate from work. Overlay catch the drift. Punchline: model grew own ICM in the weights. We build same shape by hand — folder, router, one-fact file. Interp people now confirm shape real. https://www.anthropic.com/research/global-workspace