mindmap-ppt-builder

Mindmap PPT Builder

Goal

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.

Workspace Requirement

This skill operates only within the current working directory tree (cwd itself or a mindmap-ppt/ subfolder under it). Do not search parent, sibling, or home directories for an existing checkout.

• Clone https://github.com/agegr/mindmap-ppt into ./mindmap-ppt/ under the current working directory and cd in.
• If ./mindmap-ppt/ already exists, stop and ask the user how to proceed. Do not cd .., find, or otherwise probe the filesystem to locate or create a checkout elsewhere.
• 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.

Workflow

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:

- 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.

5. 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.
6. 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 diagrams or placeholders, and JPG 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, minimal short text only when useful, and descriptive kebab-case filenames such as project/demo-flow.svg.
7. Reference images in Markdown metadata lines:

  @image process-overview.png

8. Replace project/source.js with:

export const sourceMarkdown = `
- ...
`;

Escape backticks and ${...} sequences before writing user-derived text inside the JavaScript template string.

9. Run npm run check.
10. Optional visual validation: run npm run dev and inspect the URL printed by Vite (default http://127.0.0.1:5173/, may differ if the port is taken) when browser inspection is available.

Mindmap Authoring Rules

• 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 (this is the source span a node covers, not its visible label length). 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.

Markdown And Image Example

Use @image as a metadata continuation line after the node's visible two-line label. The @image line is not displayed as node text.

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

Image Prompt Pattern

Use this style prompt for GPT Image 2:

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, minimal short text only when useful, 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. Prefer no text in images; when text improves comprehension, use only 1-3 short labels or keywords and keep detailed labels in nodes.

SVG Placeholder Pattern

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.
• Minimal short text only when useful; no logos, dense decoration, or photorealism.

Image And Asset Rules

• 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.

Validation Checklist

• project/source.js exports sourceMarkdown.
• Asset files exist for every @image.
• npm run check passes.