• PDF: file path (Claude reads PDFs natively in chunks via

  Read pages:"X-Y"

  )
• EPUB / MOBI: file path (needs

  pandoc

  or

  ebook-convert

  to extract — see

  references/sources.md

  )
• Markdown / .txt: file path (read directly)
• Pasted text: just use what was pasted
• URL to public-domain text:

  WebFetch

  (Project Gutenberg, archive.org, etc.)

references/sources.md

1. By chapter if a TOC exists (PDF with bookmarks, EPUB/MOBI converted via pandoc preserves chapter headers)
   • Use

     pdfinfo <pdf> | grep "Pages"

     for PDFs
   • Use

     pdftotext -layout <pdf> | grep -i "^chapter\|^part"

     for chapter detection, or read TOC from page 1–5
   • EPUB: after

     pandoc <epub> -o tmp.md

     , chunks are between

     # Chapter X

     headers
2. By page count for PDFs without TOC: 50 pages per chunk
3. By character count for text/markdown: 30,000 chars per chunk (~7,500 words)

~/Documents/books/<author>-<title-slug>-<YYYY-MM-DD>/chunks.json

{
  "source": "<path>",
  "title": "<book title>",
  "author": "<author>",
  "type": "pdf",
  "total_pages": 287,
  "chunking": "by-chapter",
  "chunks": [
    {"i": 0, "label": "Introduction", "pages": "1-12"},
    {"i": 1, "label": "Chapter 1: The Problem", "pages": "13-32"},
    ...
  ]
}

1. Read chunk N (

   Read

   tool with

   pages:

   for PDF, full file for text/MD)
2. Extract per the chosen mode (see

   references/output-modes.md

   for templates)
3. Append the chunk's notes to

   ~/Documents/books/<workdir>/notes-<NNN>-<label-slug>.md

1. ...

• Could connect to [[Longevity Biomarkers]] (per Chapter 3 discussion of biomarkers)
• Could connect to [[Productivity & Systems]] (per Chapter 7 framework)

The cross-reference suggestions are advisory — they're suggestions for `/sb compile` to act on, not auto-applied. Keep responsibilities separated.

"Want to capture this to second-brain? I'll write it to

${SECOND_BRAIN_VAULT:-$HOME/Documents/SecondBrain}/raw/highlights-<slug>.md

matching your vault's

highlights-

type prefix."

1. Copy the final

   notes.md

   (with the second-brain-compatible frontmatter at top) to

   ${SECOND_BRAIN_VAULT:-$HOME/Documents/SecondBrain}/raw/highlights-<slug>.md

2. Tell the user the path
3. Suggest: "Run

   /sb compile

   later to merge this into wiki pages — the cross-reference suggestions in the footer are starting points."

• One-line headline:

  <title> · <author> · <total_pages or word count> · <mode> · <chunks processed>

• Workdir path
• The TL;DR section
• For

  notes

  /

  study

  modes: brief list of top 3 takeaways
• For

  quotes

  mode: top 3 quotes
• If captured to second-brain: that path too

• second-brain

  — primary integration: writes

  highlights-<slug>.md

  to

  raw/

  . Then

  /sb compile

  merges into wiki pages.

• deep-research

  — when a research question turns up a book,

  /read-book

  is the next step. Notes feed back into the research brief.

• business-brainstorm

  — when scoring an idea (e.g., business books on similar models), read-book provides the structured evidence.

• decide

  — when a decision hinges on what an authority has written (e.g., "should I take VC money?" → read Naval / Jason Cohen),

  read-book

  extracts the relevant chapter.

• slide-deck

  — book takeaways → talk material (book talk pattern).

• watch-video

  — sibling skill, same content-consumption pattern. Audiobook? Use

  watch-video transcript

  mode.

• nonfictionskills

  /

  fictionskills

  — when researching to write a book, this skill reads the comp titles.

• Don't summarize beyond recognition. A 30-page chapter should produce 8–15 lines of notes, not 3. Compression is good; flattening is bad.
• Preserve specifics. Names, numbers, dates, quotes — keep them. The whole point is later-the user can grep "what did Andy Wilkinson say about X" and find it.
• Quotes are sacred. When you flag a quote, copy it verbatim. Note the page if possible.
• Action items are explicit. If the book makes you think "I should do X," flag it explicitly. These are the highest-leverage outputs.
• Frameworks deserve their own bullets. When the author names a framework (e.g., "the 9-dimension filter," "Save the Cat beats"), call it out by name in the notes.