SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsbook-illustration-workflow

book-illustration-workflow

Design & creation

View repository

Updated 10-15 20:07

About this skill

Book Illustration Workflow

This skill handles the complete workflow of "chapter screenshots from planning to finalizing the text and then syncing to Feishu."

Goal

Break down the illustration work in a chapter draft into a stable process:

  1. Clarify the screenshots and final images actually needed in the chapter
  2. Give step by step the real prompts that can be executed in Claude Code
  3. Constrain screenshot filenames, figure numbers, and body text positions to correspond one-to-one
  4. Put images back into the correct positions in the local Markdown
  5. Remove author notes that readers should not see
  6. Sync the reader-facing chapter and images to Feishu

Trigger conditions

Use when the user has any of the following intentions:

  • Ask you to help him do the screenshot process for a chapter of a book draft
  • Ask you to guide Claude Code step by step to take screenshots or generate images
  • Ask you to organize "figure number -> prompt -> filename -> original text position"
  • Ask you to insert images back into the local article
  • Ask you to sync the chapter to a Feishu document
  • Ask you to "continue with the same process as last time"

Core principles

1. The prompts in the screenshots must look like real work

In the prompts given to Claude Code, the following words are forbidden:

  • 书
  • 章节
  • 案例
  • 演示
  • 截图
  • 适合展示
  • chapter7-demo
  • 为了写书

The user input in the screenshots must look like a real office task.

2. Data preparation stays outside the screenshots

Do not put "create test data," "create directories for screenshot use," or "prepare samples for demonstration" into the Claude Code conversation.

Correct approach:

  • Prepare data outside the screenshots
  • Inside the screenshots, show only real usage

3. Advance only one step at a time

If the user asks "let's go step by step," each time give only:

  1. The content to send to Claude Code
  2. Which screen to capture
  3. The screenshot save name
  4. The corresponding original text position
  5. Whether there is already a placeholder in the original text

After the user completes it, give the next step.

4. Image positions follow the semantics of the body text, not insertion as a whole block based on placeholder blocks

If editing traces like this exist in the article:

  • 素材占位
  • 插图占位
  • A group of figure numbers gathered at the beginning of a section

Do not mechanically stuff all the images into that block.

Correct approach:

  • Find the sentence in the body text that actually introduces the figure
  • Place the image after that sentence or paragraph
  • When syncing to Feishu, upload in this order as well

5. The body text shown to readers must not retain author notes

These contents should not appear in the reader-facing body text or the Feishu document:

  • 素材占位
  • 插图占位
  • 建议拍法
  • 建议做法
  • 插图说明文案
  • 来源提示
  • 如果这里配一张……
  • 如果你在书里配图……
  • 你在书里可以……

Handling principles:

  • Rewrite valuable information into reader-facing body text
  • Delete pure editorial notes directly
  • For screenshots already completed, replace directly with the actual images

Standard process

Phase A: Sort out the chapter

  1. Find the target chapter Markdown file
  2. Find all figure numbers, code blocks, table placeholders, and "material placeholder" paragraphs
  3. Determine which figures are:
    • Screenshots of raw data
    • Claude Code conversation screenshot
    • Finished chart/report
    • External official page screenshot
    • Post-production layout image
  4. Give the minimal executable screenshot sequence

The usual priority order is:

  1. Installation or environment confirmation
  2. Raw data chart
  3. Core task invocation chart
  4. Result chart
  5. Comparison chart
  6. Extended application chart

If you need to quickly establish a screenshot plan for the entire chapter, prioritize using the references/chapter-illustration-template.md master table template in . It is suitable for listing "figure number, purpose, prompt, filename, original text location, whether completed" all at once before formal screenshots.

Phase B: Step-by-step screenshot guidance

If the user wants to execute step by step, output according to this template:

第N步：<动作名称>

发给 Claude Code 的内容：
```text
<真实提示词>

Which screen are you capturing:

  • <Clearly state whether to keep input / output / result chart>

Screenshot save name:

  • <filename.png>

Corresponding original text location:

  • <location in the chapter or figure number description>

Is there currently a placeholder in the original text:

  • Yes / No

要求：

- 不抢跑后续步骤
- 不一次给太多
- 文件名直接可落盘

### 阶段 C：回填本地文章

先检查本地 Markdown 是否属于哪一种：

1. `读者版正文 + 独立图片占位`
2. `作者版正文 + 大块素材占位`
3. `正文里夹着大量编辑备注`

处理方式：

- 如果还是作者版，先清理成读者版
- 如果图片已经有了，直接在正文中插入图片 Markdown
- 如果暂时没有图片，只保留简洁的读者可见表述，不保留编辑备注

图片插入格式优先：

```md
![图7-10 趋势图成品示例](第七章图片/图7-10_趋势图成品示例.png)

Phase D: Sync to Feishu

When syncing to Feishu, do not blindly make large-scale corrections to misaligned images in the old document. The preferred approach:

  1. First fix the local Markdown into a reader version
  2. Split in the order of "markdown paragraph -> corresponding image -> next paragraph"
  3. Create a new Feishu document and upload
  4. Make the new document the main document

Only patch the old document when the user explicitly requests to keep the original document.

Phase E: Feishu upload rules

When using lark-cli :

  • docs +create create document
  • docs +update --mode append append markdown
  • docs +media-insert insert image

Note:

  • Images must be inserted in body text order
  • Do not insert by "figure number sorting"; they must follow the order of appearance in the article
  • If filenames have prefix conflicts, for example 图7-1 will mistakenly match 图7-10, exact matching must be used

Exact matching rules:

  • Can only match ^(图号)(_|\\.|$) this kind of boundary
  • Cannot use simple prefix inclusion

Naming rules

Screenshot filenames use by default:

  • 图号_简短说明.png

For example:

  • 图7-2A_添加官方Skills市场.png
  • 图7-8A_调用xlsx清洗数据界面.png
  • 图7-18_报表首页示例.png

Requirements:

  • The figure number must be at the front of the filename
  • Keep the Chinese description as short as possible
  • Subsequent text insertion and Feishu upload both depend on this naming

Attached template

Quality checklist

Before declaring completion, you must self-check:

  • Whether each image matches the corresponding figure number
  • Whether the prompts look like real office tasks
  • Whether author notes still remain
  • Whether the image order in the local Markdown is correct
  • Whether images in Feishu follow the correct paragraphs
  • Whether there is 图7-1 mistakenly matching 图7-10 this kind of problem

Default response style

If the user says "follow this process," execute by default as follows:

  1. First check the target chapter file and image directory
  2. If screenshot guidance is needed, enter step-by-step mode
  3. If the screenshot is already done, fix the local Markdown first
  4. After confirming locally, then sync to Feishu

Things not to do

  • Do not leave editing notes written for the author himself in the reader's version of the main text
  • Do not reveal in the screenshot prompt that "this is for writing a book"
  • Do not blindly insert images by placeholder based on large blocks of material
  • Do not repeatedly make messy fixes on the old Feishu document, causing more misalignment
  • Do not use vague file-matching rules to batch-find images