I Stopped Taking Notes and Started Writing Books

The Problem

Digital notes are passive. I wanted to leverage AI to write them.

Programmers use Markdown to structure documentation for software projects: a table of contents, headers, lists, tables, images. It gives a subject structure and makes it navigable. Popular note-taking apps like Obsidian use this format too, but all you need is a code editor and familiarity with the Markdown format.

I used Notion for a while. It’s a great tool. But I already live in my code editor and think in markdown. My blog runs on Hugo, also markdown. Notion’s AI features aren’t free, and I wasn’t going to pay for two systems.

So I stopped taking notes and started building books with Claude Code.

How It Works

Each subject gets its own folder under books/. Each chapter is a file, prefixed by number.

books/
  kafka/
    1-introduction.md
    2-producers.md
    ...
  ccna/
  cka/

A chapter isn’t a summary of what I read. It’s a record of what I currently understand. When I find a better definition, I update it. When a video fills in a gap, I add it to the relevant chapter. When I hit a term I don’t recognize, I look it up and include it. The book grows with me.

This is research. Divide a topic into subtopics. Conquer each one in whatever order interest or confusion dictates. What I include is based on what I didn’t understand at first, or what I want to remember. The tangents aren’t distractions. They’re the point.

The Skill Stack

Claude Code is my daily driver. I run skills from the terminal inside my editor to generate and refine chapters. The skills below are available in ivorscott/cc-marketplace.

/book

Takes a PDF and summarizes it chapter by chapter into structured markdown. Each file is prefixed with its chapter number and opens with a table of contents. I run it once to start, then prompt next or a specific chapter to continue.

/book @kafka.pdf @1-introduction.md

For technical books covering a versioned tool or library, a fact-checker subagent runs in parallel. It looks up the current version and flags outdated sections inline. Outdated documentation is one of the more quietly frustrating parts of self-study, and the fact-checker handles it automatically.

/study

Generates a quiz or flashcard set in JSON from selected chapters, ready to run right after reading. The idea came from Google NotebookLM.

Arguments are optional. Without a chapter, all chapters are used. Type defaults to flashcard, difficulty to medium, count to 10.

/study ch2
/study ch2-4 quiz hard 20
/study flashcard easy 5

Sessions are saved to .stu/<slug>-<type>-<YYYYMMDD>.json. A vibe-coded CLI tool (stu) reads the file and renders an interactive terminal UI built with bubbletea. Quiz mode uses arrow keys or a-d to pick answers, h for hints, and r to retake. Flashcard mode uses spacebar to reveal, then c or x to mark correct or incorrect.

stu .stu/cka-quiz-20260331.json

/analogy and /diagram

Two smaller skills that make the books more engaging and easier to hold in memory. /analogy generates conceptual bridges for dense material. /diagram produces ASCII diagrams inline, useful for systems, flows, and anything spatial. Both operate on highlighted text: invoke the skill and the output appears directly below in the markdown.

Why It Works

When I’m having fun, I stay curious. I chase tangents. I dig into things I don’t understand instead of skipping past them. I built myself a game with its own rules and vocabulary, and the result is that I actually want to keep playing.

There’s no fixed curriculum. I control what goes in, in what order, and how deep. If something interests me, it gets a section. If I didn’t understand something the first time, it gets a definition, an analogy, a diagram. Whatever it takes. Each chapter reflects my current mental model, including the gaps.

AI generates the scaffolding. I decide what stays. The book becomes a record of what I actually understand, not just what I consumed. Every choice about what to include or cut is a small act of judgment. Over time, those choices compound. I recognize what matters faster. I get better at knowing what I don’t know.

That’s the real product of this system. Not the books. The books are a side effect. What I’m actually building is a better version of how I learn. The workflow is still evolving, and I’m curious to see where it goes next.