Read book

Turn a book or long PDF into structured, grep-able notes. Same content-consumption pattern every time: ingest → chunk → extract → write a dated clip to the taste-library. The clip is the deliverable — a knowledge asset that /content-strategy and /thought-leadership pull from for evidence, angles, and named frameworks.

Triggers

"read book", "read this book / PDF", "book notes", "extract notes from this PDF", "summarize this ebook", "what's in this book", "pull quotes from this", "extract frameworks from [book]".

What it accepts

Input	How it's read
PDF	The Read tool reads PDFs natively (~10 pages per call — chunk longer books). The pdf skill handles extraction edge cases.
Markdown /.txt	Read directly. No conversion.
Pasted text	Use what was pasted. Treat a short paste as one chunk.
URL (public-domain text)	WebFetch first (free). Firecrawl (firecrawl_scrape) only if the page is JS-heavy or blocked. Project Gutenberg / archive.org .txt URLs are cleanest.
EPUB / MOBI	Convert to PDF or markdown first (ebook-convert, calibre), then read as PDF. If no converter is installed, deferred — give the user the one-line install.

Detect type from the file extension or URL. If ambiguous, ask.

Modes

Invocation	Mode	What you get
read-book <input>	notes (default)	Chapter-by-chapter: TL;DR + key concepts + quotes + action items + frameworks
read-book <input> summary	summary	Whole-book TL;DR (1 paragraph) + 3–5 takeaways + who-it's-for
read-book <input> quotes	quotes	Pull-quote highlights only, with chapter + page refs
read-book <input> study	study	Notes + 10–20 spaced-repetition Q&A cards

Full per-mode templates: the premium reference. A long book (>200 pages) with no mode given defaults to notes — warn it'll take many read passes.

Process

Step 1 — Parse input and mode

Detect the format (table above) and the mode (default notes). Confirm the target book if the input is ambiguous.

Step 2 — Get the text and plan the chunks

Per-source ingestion: the premium reference. Chunk in priority order:

1. By chapter when a TOC exists (PDF bookmarks, or converted EPUB chapter headers).
2. By 50-page block for PDFs with no TOC.
3. By 30,000-char block (~7,500 words) for markdown / text.

Keep a short chunking plan (source, title, author, type, total pages, strategy, chunk list) in the scratchpad — not in the repo.

Step 3 — Read each chunk and extract

Loop: read chunk N → extract per the chosen mode (the premium reference) → append the partial to the scratchpad. For PDFs, read chunks individually — never the whole book in one call (the PDF Read cap is ~10 pages). If a chunk is front-matter or diagrams with nothing to extract, log the skip and continue.

Step 4 — Aggregate into the full notes

Combine the chunk partials into the mode's full-book shape: TL;DR, key takeaways (3–7 — the ones that survive forgetting), who-it's-for, verdict, chapter notes, and cross-reference suggestions.

Step 5 — Write the taste-library clip

The clip is the deliverable. Write to:

projects/research/taste-library/resources/{MMYY}-{slug}.md

{MMYY} = current month+year (0726 = July 2026). {slug} = author + short-title in kebab-case (0726-hormozi-100m-offers.md). Frontmatter matches the taste-library clip convention (below). Confirm the tags against the blessed vocabulary in projects/research/taste-library/CLAUDE.md, and write a real why — a clip without a why is a bookmark, not a taste signal.

Step 6 — Report

In chat: one-line headline (<title> · <author> · <pages or words> · <mode> · <chunks>), the clip path, the TL;DR, and the top 3 takeaways (top 3 quotes for quotes mode).

Evidence discipline

Bound by .claude/rules/evidence-bound-outputs.md. Quotes are the highest-value output and the easiest to corrupt:

• Quote verbatim. Copy the exact words. Never paraphrase into quotation marks.
• Cite the locator. Page number (p. 84) when available; chapter + paragraph (Ch 3, ¶12) for EPUB-converted books with no stable pages.
• Attribute. When the author quotes someone else, name them after the quote. When it's the author, no attribution is needed.
• Never invent. If the book doesn't say it, it doesn't go in the notes. Flag a gap rather than fill it.

Quality notes

• Don't flatten. A 30-page chapter earns 8–15 lines, not 3. Compression is good; flattening loses the specifics that make the clip worth keeping.
• Preserve specifics. Names, numbers, dates, examples — keep them. The point is that later you can grep "what did X say about Y" and find it.
• Action items are explicit. When the book makes you think "I should do X," flag it. These are the highest-leverage lines for GTM work.
• Name the frameworks. When the author names a framework, call it out by name — a framework you remember outperforms ten ideas you forget.
• Push back. In notes and study mode, surface disagreements. Reading critically beats reading reverently.

Error handling

Failure	Response
EPUB/MOBI, no converter	brew install calibre (ebook-convert) or brew install pandoc, then convert to PDF/markdown first.
Scanned PDF (no text layer)	OCR first: brew install ocrmypdf && ocrmypdf <in.pdf> <out.pdf>.
PDF has no TOC	Fall back to 50-page chunks; note it in the clip.
Book >500 pages	Warn on time and read passes; offer summary mode instead of full notes.
Chunk extracts nothing	Skip, log, continue — don't fail the whole run.

Composes with

• /content-strategy — book takeaways and frameworks feed cluster planning and content angles.
• /thought-leadership — quotes, frameworks, and pushback become evidence and counterpoints in long-form.
• /deep-research — when a research question surfaces a book, this is the next step; notes feed back into the brief.
• /storytelling and /gtme-pulse — book stories and stats become narrative and newsletter material.

Attribution

Adapts coreyhaines31/makerskills/read-book (MIT, © 2026 Corey Haines), accessed 2026-07-08. Re-pointed at our taste-library as the notes sink.