Have you ever tried sending an e-book of more than 500 pages to an AI?
Correspondingly, most postgraduate entrance examinations, final review, and scientific research studies often require you to review some fragmented knowledge points in books.
However, most electronic books do not have a table of contents at all.
It is definitely unrealistic to turn page by page by yourself.
So you thought of using AI to help you locate and interpret knowledge points
But I sent him a PDF file directly. The file was too large.
He cannot and cannot eat it in one sitting
The correct approach is to convert the PDF into a file format suitable for AI reading (markdown)
Then divide it into chapters and automatically divide the chapter level according to the subject classification in the book.
If you use MinerU to parse, you will get a lot of garbled formatting.
It is also very difficult to adjust the content manually.
You will definitely want to
Is there a way to abstract this process into a standardized, repetitive work?
Yes, there is such a SKILL that can solve this problem.
What can this SKILL do?
He can distill any data into structured text that is readable and searchable by AI.
Organized into categories just like books organized into different chapters.
Load on demand when needed, knowledge base similar to RAG and LLM wiki
However, the form is changed to SKILL
In the future, if you ask a professional question, AI will automatically call this skill to retrieve the answer from the books and materials you provided.
The formula derivation is based on the original text, the content reference is derived from the original text, and the views and opinions are all derived from the original text.
If it really doesn't work, then we can deduce it based on the Agent's own model capabilities.
Project address:
Install with one command
After installation, use it directly on your Agent tool and send it the books you need it to learn and distill.
Supports PDF scanned files, docx and other file types, and the content type is not limited: books, papers, fragmentary notes, drafts, company documents, specifications, contracts, etc.
For example:
I sent the physical chemistry book to the Codex in Pi
Halfway through, I asked him to switch to using MinerU API to parse PDF.
Because the Docling provided by SKILL does not support scanned PDFs
After completion, a skill will be generated, which can be called later with the tilt stick.
I asked him to check what the first chapter said
As you can see, it is indeed completely consistent with the content in the book.
Then you can continue to ask for knowledge points
Yes, I think this is very necessary for final review, postgraduate entrance examination preparation, etc.
If books are replaced by papers, they can be used as a literature knowledge base
In short, this book-to-skill skill is very useful, especially in scientific research work.
Currently there are 27.7kstar on GitHub, I think it is worth a try
This is a tool that has a knowledge base after karpathy wiki
- Author:Sanmu
- URL:https://luffysite.top/article/2026-09-02
- Copyright:All articles in this blog, except for special statements, adopt BY-NC-SA agreement. Please indicate the source!
Relate Posts
