mindmap-ppt-builder

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Mindmap PPT Builder

Mindmap PPT 构建工具

Goal

目标

Turn a user-provided source document into a presentation-ready
project/source.js
for this repo. The source document may be pasted text in the conversation or a local text/Markdown file path supplied by the user. The output is a preorder mind-map: concise two-line nodes, optional node images, and local assets that match the current light PPT style.
Read
references/project-format.md
when you need exact project file conventions or visual constraints.
将用户提供的源文档转换为此仓库中可直接用于演示的
project/source.js
文件。源文档可以是对话中粘贴的文本,也可以是用户提供的本地文本/Markdown文件路径。输出结果为前序遍历的思维导图:简洁的两行节点、可选的节点图片,以及符合当前浅色PPT风格的本地资源。
当需要明确项目文件规范或视觉约束时,请阅读
references/project-format.md

Workspace Requirement

工作区要求

Use this skill inside the
agegr/mindmap-ppt
repository root.
  • If the current directory already contains
    package.json
    ,
    index.html
    ,
    src/
    , and
    project/
    , treat it as the repo root.
  • If the repo is not present, clone
    https://github.com/agegr/mindmap-ppt
    into the current working directory as a folder named
    mindmap-ppt
    , then enter that folder.
  • If a
    mindmap-ppt
    path already exists but is not this repo, stop and ask the user where to place the clone.
  • Keep the application repository outside the skill folder. Do not copy or clone
    index.html
    ,
    src/
    , or
    project/
    into
    .agents/skills/mindmap-ppt-builder/
    .
  • Normal skill output should modify only
    project/source.js
    and local asset files under
    project/
    .
  • Do not delete existing project assets unless the user explicitly asks for cleanup.
  • Do not edit
    src/
    ,
    index.html
    , or application behavior unless the user explicitly asks for implementation changes.
请在
agegr/mindmap-ppt
仓库根目录内使用此技能。
  • 如果当前目录已包含
    package.json
    index.html
    src/
    project/
    ,则将其视为仓库根目录。
  • 如果仓库不存在,请将
    https://github.com/agegr/mindmap-ppt
    克隆到当前工作目录,命名为
    mindmap-ppt
    文件夹,然后进入该文件夹。
  • 如果
    mindmap-ppt
    路径已存在但并非此仓库,请停止操作并询问用户应将克隆放置在何处。
  • 请将应用仓库放在技能文件夹之外。不要将
    index.html
    src/
    project/
    复制或克隆到
    .agents/skills/mindmap-ppt-builder/
    中。
  • 常规技能输出应仅修改
    project/source.js
    project/
    下的本地资源文件。
  • 除非用户明确要求清理,否则请勿删除现有项目资源。
  • 除非用户明确要求修改实现,否则请勿编辑
    src/
    index.html
    或应用行为。

Workflow

工作流程

  1. Get the user's source document:
    • Use pasted text from the conversation when provided.
    • If the user gives a local file path, read only that explicitly provided document file and use its contents.
    • If neither pasted text nor a readable local file is available, ask the user for the document before generating
      project/source.js
      .
  2. Read the document and identify the presentation thesis.
    • Follow the source language by default: Chinese input -> Chinese output; English input -> English output.
    • For English output, still use an eyebrow/headline structure for two-line nodes when natural.
    • Do not silently correct facts. If the source has obvious contradictions or questionable claims, preserve the claim carefully or mention the conflict to the user.
    • For very long documents, preserve the original chapter structure first. If the material is too broad to reduce confidently, draft a high-level outline and ask the user to confirm priorities before finalizing.
  3. Build a clear logic tree:
    • root: document/source name or presentation topic
    • major branches: usually 2-4 sections, but follow the source logic when another structure is clearer
    • child nodes: use them for causes, consequences, evidence, examples, process steps, contrasts, or supplements
    • depth: add levels only when nesting makes the author's logic easier to understand
  4. Write each node as one unordered-list item plus an optional continuation line:
md
- 副标题
  主标题
Use the first line as a short category label and the second line as the main message. Keep each line under about 30 Chinese characters or 8 English words. Prefer two-line labels for all visible nodes; use a single-line node only when the label is already extremely short and clear.
  1. Choose image nodes sparingly:
    • Node images are optional.
    • A mind-map necessarily omits a lot of source detail; use images to preserve or explain the omitted detail on high-information nodes.
    • Pick 3-8 high-information nodes for a typical deck; short drafts may use 0-2 images.
    • Prefer nodes that summarize a process, architecture, comparison, timeline, metric, or conceptual model.
  2. Generate illustrations for chosen nodes with GPT Image 2 or the available image generation tool. Save them under
    project/
    or a subfolder of
    project/
    .
    • Prefer PNG for generated raster illustrations, SVG for simple diagram placeholders, and JPG only for photo-like assets.
    • If image generation is unavailable, either omit images or create simple SVG placeholder diagrams under
      project/
      using the same restrained palette. Use 16:10 composition, no text inside images, and descriptive kebab-case filenames such as
      project/demo-flow.svg
      .
  3. Reference images in Markdown metadata lines:
md
  @image process-overview.png
  1. Replace
    project/source.js
    with:
js
export const sourceMarkdown = `
- ...
`;
Escape backticks and
${...}
sequences before writing user-derived text inside the JavaScript template string.
  1. Run
    npm run check
    .
  2. Optional visual validation: run
    npm run dev
    and inspect
    http://127.0.0.1:5173/
    when browser inspection is available.
  1. 获取用户的源文档:
    • 如果对话中提供了粘贴的文本,则使用该文本。
    • 如果用户提供了本地文件路径,则仅读取该明确指定的文档文件并使用其内容。
    • 如果既没有粘贴文本也没有可读的本地文件,请在生成
      project/source.js
      前询问用户提供文档。
  2. 阅读文档并确定演示文稿的核心论点:
    • 默认遵循源语言:中文输入→中文输出;英文输入→英文输出。
    • 对于英文输出,在自然的情况下仍使用“副标题/主标题”的两行节点结构。
    • 请勿擅自修正事实。如果源文档存在明显矛盾或可疑主张,请谨慎保留该主张或向用户提及冲突点。
    • 对于篇幅极长的文档,首先保留原章节结构。如果内容过于宽泛而无法自信地精简,请草拟一份高层级大纲,并在定稿前请用户确认优先级。
  3. 构建清晰的逻辑树:
    • 根节点:文档/源名称或演示主题
    • 主要分支:通常为2-4个章节,但如果其他结构更清晰,则遵循源文档的逻辑
    • 子节点:用于表示原因、结果、证据、示例、流程步骤、对比或补充内容
    • 层级:仅当嵌套能更清晰地表达作者逻辑时才添加层级
  4. 将每个节点写为一个无序列表项加可选的续行:
md
- 副标题
  主标题
将第一行用作简短的类别标签,第二行用作核心信息。每行内容控制在约30个中文字符或8个英文单词以内。所有可见节点优先使用两行标签;仅当标签本身已极短且清晰时才使用单行节点。
  1. 谨慎选择需要添加图片的节点:
    • 节点图片为可选内容。
    • 思维导图必然会省略大量源文档细节;请在高信息量节点上使用图片来保留或解释这些被省略的细节。
    • 对于典型演示文稿,选择3-8个高信息量节点;短草稿可使用0-2张图片。
    • 优先选择总结流程、架构、对比、时间线、指标或概念模型的节点。
  2. 使用GPT Image 2或可用的图像生成工具为选中的节点生成插图。将图片保存到
    project/
    project/
    的子文件夹中。
    • 生成的栅格插图优先使用PNG格式,简单图表占位符优先使用SVG格式,仅在需要类照片资源时使用JPG格式。
    • 如果无法使用图像生成工具,则要么省略图片,要么使用相同的简约调色板在
      project/
      下创建简单的SVG占位图表。使用16:10的构图比例,图片内不包含文字,使用短横线分隔的描述性文件名,例如
      project/demo-flow.svg
  3. 在Markdown元数据行中引用图片:
md
  @image process-overview.png
  1. project/source.js
    替换为以下内容:
js
export const sourceMarkdown = `
- ...
`;
在JavaScript模板字符串中写入用户提供的文本前,请转义反引号和
${...}
序列。
  1. 运行
    npm run check
  2. 可选的视觉验证:当可以进行浏览器检查时,运行
    npm run dev
    并查看
    http://127.0.0.1:5173/

Mindmap Authoring Rules

思维导图创作规则

  • Do not force every deck into a strict
    root -> level 1 -> level 2 -> level 3
    taxonomy. That shape is only a useful default, not a rule.
  • Let the hierarchy express the author's logic structure. If cause A leads to result B, B can be a child node of A; if B further leads to result C, form an
    A -> B -> C
    subtree.
  • The only hard principle is clarity: a reader should understand why each child node belongs under its parent and what relationship is being expressed.
  • Follow the source order. This app reveals nodes in preorder: parent first, then all children. Do not move conclusions from later text into earlier parent labels.
  • Do not repeat the root topic in child nodes. If the root already states the problem or theme, children should advance the story.
  • Group nearby meanings under one parent. Keep backgrounds, criteria, risks, product/tool inventories, recommendations, and conclusions in their own coherent branches.
  • Main nodes carry judgments; child nodes carry evidence, reasons, examples, or supplements. If a node explains another node, make it a child, not a sibling.
  • Keep each parent to at most 5 children. If there are more, add grouping nodes.
  • Split tools/products only when the source analyzes them one by one. Merge them when the source merely lists options in passing.
  • Do not split sentence by sentence. One node should carry one complete small point.
  • Each node should correspond to about 10-80 Chinese characters of source material. Less than 10 is usually too fragmented; more than 80 usually needs splitting.
  • Node text may be slightly longer than a normal title, but one node should not contain multiple independent ideas.
  • Parent labels should summarize and navigate; child labels should reveal specifics. Avoid parent labels that spoil later details.
  • Put images on high-information nodes, such as framework, comparison, inventory, recommendation, or risk-model nodes. Avoid images on very small detail nodes.
  • 不要强行将所有演示文稿纳入严格的
    根节点→一级节点→二级节点→三级节点
    分类体系。这种结构只是有用的默认方式,而非规则。
  • 让层级结构表达作者的逻辑。如果原因A导致结果B,则B可以作为A的子节点;如果B进一步导致结果C,则形成
    A→B→C
    的子树。
  • 唯一的硬性原则是清晰:读者应能理解每个子节点为何属于其父节点,以及两者之间的关系。
  • 遵循源文档的顺序。此应用会以前序遍历的方式展示节点:先父节点,再所有子节点。请勿将后续文本中的结论移至前面的父节点标签中。
  • 子节点中不要重复根节点的主题。如果根节点已经说明了问题或主题,子节点应推进内容的讲述。
  • 将含义相近的内容归到同一个父节点下。将背景信息、标准、风险、产品/工具清单、建议和结论放在各自连贯的分支中。
  • 主节点承载判断;子节点承载证据、理由、示例或补充内容。如果一个节点是对另一个节点的解释,则将其设为子节点,而非同级节点。
  • 每个父节点最多包含5个子节点。如果超过5个,请添加分组节点。
  • 仅当源文档逐一分析工具/产品时才拆分它们;如果源文档只是顺带列出选项,则将它们合并。
  • 不要逐句拆分内容。一个节点应承载一个完整的小观点。
  • 每个节点应对应源文档中约10-80个中文字符的内容。少于10个字符通常过于零散;超过80个字符通常需要拆分。
  • 节点文本可以比普通标题稍长,但一个节点不应包含多个独立观点。
  • 父节点标签应起到总结和导航作用;子节点标签应展示具体细节。避免父节点标签提前透露后续细节。
  • 在高信息量节点上添加图片,例如框架、对比、清单、建议或风险模型节点。避免在非常细节的小节点上添加图片。

Markdown And Image Example

Markdown 与图片示例

Use
@image
as a metadata continuation line after the node's visible two-line label. The
@image
line is not displayed as node text.
js
export const sourceMarkdown = `
- 产品发布
  三分钟讲清楚新功能
  @image overview.png
    - 用户痛点
      当前流程成本很高
      @image image-asset-1/pain-points.jpg
    - 解决方案
      自动整理文稿和插图
    - 演示效果
      像 PPT 一样逐步展开
      @image diagrams/demo-flow.svg
`;
Image paths are relative to
project/
by default:
  • @image overview.png
    ->
    ./project/overview.png
  • @image image-asset-1/pain-points.jpg
    ->
    ./project/image-asset-1/pain-points.jpg
  • @image diagrams/demo-flow.svg
    ->
    ./project/diagrams/demo-flow.svg
在节点的可见两行标签之后,使用
@image
作为元数据续行。
@image
行不会作为节点文本显示。
js
export const sourceMarkdown = `
- 产品发布
  三分钟讲清楚新功能
  @image overview.png
    - 用户痛点
      当前流程成本很高
      @image image-asset-1/pain-points.jpg
    - 解决方案
      自动整理文稿和插图
    - 演示效果
      像 PPT 一样逐步展开
      @image diagrams/demo-flow.svg
`;
图片路径默认相对于
project/
  • @image overview.png
    ./project/overview.png
  • @image image-asset-1/pain-points.jpg
    ./project/image-asset-1/pain-points.jpg
  • @image diagrams/demo-flow.svg
    ./project/diagrams/demo-flow.svg

Image Prompt Pattern

图片提示词模板

Use this style prompt for GPT Image 2:
text
Create a clean presentation illustration for a light PPT mind-map node.
Subject: <node main idea>.
Include: <2-4 concrete visual elements from the source text>.
Style: restrained vector-like editorial illustration, warm off-white background, dark teal #183a4a, muted green #eef7f3, orange accent #d8894f, simple geometric shapes, thin shadows, small 8px-radius card-like forms, no photorealism, no text, no logos, no busy decorations.
Composition: centered, generous whitespace, readable at thumbnail size, aspect ratio 16:10.
If the source needs a real chart, diagram, or screenshot, create a simple diagrammatic illustration instead of inventing precise numbers. Do not put text in images; labels belong in nodes.
为GPT Image 2使用以下风格的提示词:
text
Create a clean presentation illustration for a light PPT mind-map node.
Subject: <node main idea>.
Include: <2-4 concrete visual elements from the source text>.
Style: restrained vector-like editorial illustration, warm off-white background, dark teal #183a4a, muted green #eef7f3, orange accent #d8894f, simple geometric shapes, thin shadows, small 8px-radius card-like forms, no photorealism, no text, no logos, no busy decorations.
Composition: centered, generous whitespace, readable at thumbnail size, aspect ratio 16:10.
如果源文档需要真实的图表、示意图或截图,请创建简单的示意性插图,而非编造精确数据。请勿在图片中添加文字;标签应放在节点中。

SVG Placeholder Pattern

SVG 占位图模板

When no image-generation tool is available but an illustration is still useful, a simple SVG placeholder is acceptable:
  • Size:
    1280x800
    .
  • Style: warm off-white background, dark teal
    #183a4a
    , muted green
    #eef7f3
    , orange accent
    #d8894f
    .
  • Content: abstract process blocks, arrows, cards, or timeline shapes based on the node idea.
  • No readable text, logos, dense decoration, or photorealism.
当无法使用图像生成工具但仍需要插图时,可使用简单的SVG占位图:
  • 尺寸:
    1280x800
  • 风格:暖灰白色背景,深青色
    #183a4a
    ,柔和绿色
    #eef7f3
    ,橙色强调色
    #d8894f
  • 内容:基于节点主题的抽象流程块、箭头、卡片或时间线形状。
  • 不包含可读文字、标志、密集装饰或写实风格元素。

Authoring Rules

创作规则

  • Preserve the user's argument. Do not flatten important causal relationships into generic slogans.
  • Keep the preorder reveal useful: each next node should add a clear idea.
  • Use
    @image
    only after the node's title line, before its children.
  • Use PNG, JPG/JPEG, or SVG assets.
  • Save image files under
    project/
    or a subfolder of
    project/
    .
  • Prefer short
    @image
    values relative to
    project/
    , e.g.
    @image user-journey.png
    ,
    @image diagrams/user-journey.png
    , or
    @image image-asset-1/a.jpg
    .
  • 保留用户的论证逻辑。请勿将重要的因果关系简化为通用口号。
  • 确保前序遍历的展示效果有用:每个后续节点都应添加明确的观点。
  • 仅在节点的标题行之后、子节点之前使用
    @image
  • 使用PNG、JPG/JPEG或SVG格式的资源。
  • 将图片文件保存到
    project/
    project/
    的子文件夹中。
  • 优先使用相对于
    project/
    的短
    @image
    路径,例如
    @image user-journey.png
    @image diagrams/user-journey.png
    @image image-asset-1/a.jpg

Validation Checklist

验证清单

  • project/source.js
    exports
    sourceMarkdown
    .
  • Asset files exist for every
    @image
    .
  • npm run check
    passes.
  • project/source.js
    导出
    sourceMarkdown
  • 每个
    @image
    对应的资源文件均存在。
  • npm run check
    执行通过。