菜单
Publish a new monthly edition of a deep-dive book in books/<NNN>-<slug>/. Use when the user says "refresh the <book> book", "new edition", "publish an update", or mentions monthly/quarterly refresh of a book in this repo. Handles research, per-chapter updates, consistency review, metadata bumps, and build.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Create a brand new company deep-dive book from scratch. Use when the user says "create a book about <company>", "new book", "write about <company>", or wants to start a deep-dive analysis of a company that does not yet have a book in this repo. Handles research with a team of agents, chapter planning, parallel chapter writing, consistency review, about page generation, and build.
Prepare a book for Leanpub publication after a refresh. Use when the user says "publish", "release notes", "notify readers", "send to Leanpub", or wants to prepare the email notification for a new book edition. Updates the about page fields to reflect the latest manuscript content, then generates concise email-ready release notes to paste into Leanpub's "Notify Readers" form.
基于 SOC 职业分类
| name | book-refresh |
| description | Publish a new monthly edition of a deep-dive book in books/<NNN>-<slug>/. Use when the user says "refresh the <book> book", "new edition", "publish an update", or mentions monthly/quarterly refresh of a book in this repo. Handles research, per-chapter updates, consistency review, metadata bumps, and build. |
| argument-hint | <slug> |
Use this skill when the user asks to publish a new edition of any deep-dive book in books/<NNN>-<slug>/. Triggers include "refresh the Palantir book", "new edition of Rocket Lab", "publish an update", or any phrasing that implies rolling the book forward to a new monthly (or quarterly) cut. The skill is book-agnostic: it takes a slug like palantir or rocketlab as input and operates on the corresponding books/<NNN>-<slug>/ directory.
Editions are date-based, not numbered. A new release is "April 2026 edition", never "Third Edition". The book identity (title, subtitle) does not encode an edition number and must never be changed to add one.
Two metadata.yaml fields anchor the edition:
date: "<Month Year>": human-readable edition label (for example, "April 2026"). Updated every refresh.last_published: "<YYYY-MM-DD>": ISO date of the previous release. This is the delta window anchor that the research step uses to decide what to look for. Updated every refresh, at the end of the cycle, to the release date of the new edition.Leanpub's built-in version history archives prior editions for readers, so nothing in the book's title, subtitle, frontmatter, or about page needs to reference "First Edition", "Second Edition", or earlier release dates. If you find that language, strip it.
Default cadence is monthly. The skill works for any cadence: it reads last_published from metadata and treats everything since that date as the delta window, so quarterly or ad-hoc refreshes work identically.
Run these steps in order for each refresh. Each numbered file under .claude/skills/book-refresh/ is a self-contained prompt; invoke them by reading the file and following its instructions with the book slug filled in.
01-research-delta.md): one consolidated web-research pass for everything material about the book's subject since last_published. Output goes to books/<NNN>-<slug>/.research/<YYYY-MM>-delta.md.02-fan-out-findings.md): turn the research file into a per-chapter change list at books/<NNN>-<slug>/.research/<YYYY-MM>-changes.md.03-update-chapter.md): apply the change list to one chapter at a time. Determine chapter ordering dynamically (see below).04-consistency-review.md): cross-chapter pass for contradictions (numbers, dates, contract values, thesis language) and fix any drift.05-metadata-framing.md): update metadata.yaml, about.md, and 00-frontmatter.md to reflect the new edition date. Strip any edition-numbering language.06-release-notes.md): regenerate the "What's new in this edition" section in about.md from the research file, the changes file, and the git diff of the manuscript.07-build.md): run python scripts/build.py <slug> from the repo root. Verify the output.Do not hardcode chapter numbers. Instead, read each chapter's H1 title and determine ordering by volatility:
Read the chapter titles at the start of step 3 and order accordingly. For a typical 10-chapter company deep dive, the order is usually: financials, market, use cases, ethics, risk, outlook, history, technology, introduction, conclusion. Adapt for books with different chapter structures.
This skill takes a book slug as input. Everywhere a path appears, substitute the folder number and slug:
books/<NNN>-<slug>/manuscript/<NN>-<name>.mdbooks/<NNN>-<slug>/metadata.yamlbooks/<NNN>-<slug>/about.mdbooks/<NNN>-<slug>/manuscript/00-frontmatter.mdbooks/<NNN>-<slug>/.research/ (create if missing)books/<NNN>-<slug>/output/Nothing in this skill is hardcoded to Palantir or any specific book. If a prompt needs the book subject (company name, focus area), read it from books/<NNN>-<slug>/metadata.yaml at the start of the step.
Respect writing conventions in project CLAUDE.md: no em dashes, sentence-case headings, acronym rules, chapter files start with # Title, frontmatter has no H1.