oreilly-reader

Claude Code skill that reads and synthesizes O'Reilly Learning books.

Features • Installation • Usage • Modes • How It Works

---

oreilly-reader is a Claude Code skill. It navigates an O'Reilly Learning book chapter by chapter through Playwright browser automation, extracts the full text from each chapter, and produces a technical synthesis tuned to the reader's background.

O'Reilly renders chapter content with JavaScript. A standard HTTP fetch returns a shell. The skill drives the Playwright MCP browser, evaluates document.querySelector('main').innerText on each chapter page, and pulls the actual content. The skill avoids browser_snapshot for chapter extraction: real chapters exceed 170K characters and the snapshot path truncates.

Features

• Two modes: single book (URL or title to synthesis) and research goal (engineering goal to cross-book synthesis)
• Playlist URL handling: extracts the book list, picks the target book
• TOC parsing via browser_evaluate, robust across publisher URL patterns (Packt, Apress, O'Reilly native)
• Per-chapter text extraction direct from the rendered DOM
• Sub-agent synthesis with tunable depth
• No Python dependencies. Skill bundle ships as a single .skill archive
• Lives alongside the rest of the NuClide research toolchain

Installation

Requires Claude Code with the Playwright MCP plugin enabled and an active O'Reilly Learning subscription (logged in via the Playwright browser).

claude mcp install oreilly-reader.skill

Or download oreilly-reader.skill from the releases page and install through Claude Code settings.

Usage

Drop an O'Reilly book URL into Claude Code:

https://learning.oreilly.com/library/view/build-a-large/9781633437166/

Or a playlist URL:

https://learning.oreilly.com/playlists/d1def9e4-815c-4a24-9db5-db7da9ebc16d

The skill handles playlist to book-list extraction, TOC parsing, chapter-by-chapter text extraction, and synthesis through a sub-agent at tunable depth.

Modes

Single book. User provides a URL or a title. The skill reads targeted chapters and synthesizes. Best for "what's the meat of this book" sessions.

Research goal. User has an engineering goal ("find books for X"). The skill searches O'Reilly, evaluates TOCs, reads targeted chapters across multiple books, then produces a cross-book synthesis. This is the more powerful mode, designed for read-apply-repeat cycles on a specific engineering target.

How it works

O'Reilly renders chapter content via JavaScript. Static HTTP fetch returns a shell. The only reliable extraction path is document.querySelector('main').innerText via browser evaluation after the page loads.

browser_snapshot is never used for chapter text. Snapshots blow past Claude Code's token limit on any real chapter (171K characters trigger truncation). The skill calls browser_evaluate with the filename parameter for every chapter read.

TOC discovery uses browser_evaluate against a[href*="/library/view/"] and filters by ISBN. The same approach holds across publishers; only the per-chapter URL pattern varies (Packt uses /Text/Chapter_N.xhtml or /BN_NN.xhtml, O'Reilly native uses /c01.xhtml, Apress uses /XHTML/B978...).

Auth

If Playwright redirects to the login page, log into O'Reilly in your browser. The Playwright session needs a live O'Reilly cookie. Once logged in, navigate back to the book URL.

Skill bundle

The skill ships as a Claude Code skill bundle. The bundle contains:

oreilly-reader/
  SKILL.md          # skill definition: name, description, allowed-tools, workflow
oreilly-reader.skill  # the packaged bundle

The allowed-tools field gates the skill to the Playwright MCP browser tools plus Bash and Agent for the synthesis sub-agent.

Our other projects

• aimap — fingerprint scanner for AI and ML infrastructure
• pharos — autonomous offensive research agent
• sentinel — CVE-reactive exposure pipeline
• scanner — fast active-banner stage between Shodan and aimap
• visorlog — finding ledger

License

MIT. Part of the NuClide toolchain. Contact: nuclide-research.com