菜单
Produce the book's figures — establish the graphics style guide, spec each figure, and generate them. Use after chapters are drafted, in parallel with review.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/danielrosehill/Claude-Book-Writing-Plugin --skill book-graphics命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Run the definition stage of a book project — record the brief verbatim, interview for the gaps, and settle the spec. Use at the start of a book, before any planning or outlining.
Dispatch chapter-writing subagents in parallel batches to draft the book, and handle research where a chapter's sources are not yet gathered. Use once briefs exist. Enforces the batching rule that keeps parallel writers from colliding.
Run the planning stage — write the style guide, design the structure, and produce one brief per chapter. Use after the spec is settled and before any drafting.
正在显示 SKILL.md
| name | book-graphics |
| description | Produce the book's figures — establish the graphics style guide, spec each figure, and generate them. Use after chapters are drafted, in parallel with review. |
For technical and instructional books, figures are content. This stage exists so that thirty figures made in thirty separate invocations look like one book's figures — the prose-voice problem in another medium, with the same solution.
graphics-director → figure-maker (per figure, parallel)
guide + specs sources + assets
graphics-director writes guides/graphics-style-guide.md first, before
any figure exists. Retrofitting a visual language across drawn figures does
not work.graphics/specs/, adding paths to each chapter's figures list.figure-maker produces figures from specs. Parallelises freely — each
reads its own spec and the style guide, and nothing else.Read [production] in book.toml. If the interior prints greyscale, no
distinction may be carried by hue alone. This cannot be retrofitted across a set
of finished figures, and colour interiors change cost per page substantially.
By showing a mechanism prose would take a page to describe and still leave ambiguous. If prose does it in two sentences, prose wins. Decorative images and "visual interest" are out.
Image models produce diagrams with invented labels, near-miss arrows and plausible structure that is subtly wrong, and none of it survives print scrutiny. Diagrams are Typst or SVG source, committed, so a wrong label is fixable.
Generated imagery is fine for illustration that carries no information. Each spec says which kind it is.
Figure-makers invent plausible labels when they are not given exact ones, and invented labels print. Every string appearing in a figure is written in its spec.
python3 scripts/book.py lint fails on a ::: figure <id> directive with no
spec at graphics/specs/<id>.md.
It cannot check that the figure shows what the spec describes. Look at each figure at final printed size — a diagram that is clear on screen and illegible at 3.5 inches wide has failed.