K2pdfopt Skill — Development skill for Claude Code

🇪🇸 [Leer en español](README.es.md)

k2pdfopt

A Claude Code skill that converts scanned/native PDF books into Kindle- and Kobo-ready PDFs with [k2pdfopt](https://www.willus.com/k2pdfopt/), then verifies the result by sampling rendered pages instead of trusting the conversion blind — or rendering all of them.

 *Real conversion, same book: 2.9 MB / 140-page scan → 51.3 MB / 727-page grayscale Kindle output (`-fs 18`) and 97.8 MB / 717-page color Kobo Libra Colour output (`-fs 22`). Cover photo and publisher logos blurred here for copyright reasons — the source book isn't ours to redistribute; the point is the grayscale/color contrast, which survives the blur.*

The problem

A PDF built for print or scanned from a physical book does not read well on a 6-7" e-ink panel: tiny justified text, margins sized for 8.5x11 paper, and sometimes JPEG2000-encoded scan images that not every reader decodes. A single `k2pdfopt` invocation isn't enough on its own, either — you have to pick the right mode/device profile/DPI for the *target device*, and on a 700+ page reflowed output you have no practical way to confirm nothing broke without opening the whole book.

What the skill does

1. Resolves the k2pdfopt binary — no Homebrew formula exists for it (see below), so this is a real step, not a formality.
2. Inspects the source PDF — page count, whether it's a real text layer or a pure scan, whether it embeds JPEG2000 images.
3. Pre-flattens JPEG2000 scans to JPEG when needed — some k2pdfopt builds don't decode JP2 reliably; poppler does, so it's used as a pre-pass instead of gambling on silent corruption.
4. Picks target device(s) — Kindle, Kobo Libra, or both.
5. Calibrates large-print settings, only if asked — font size in points is tuned per device (see below for why that's not a constant), with an optional visual size