xmind-skill（x 思维技能）

xmind 🧠

Generate and read XMind .xmind files using the published MCP server xmind-generator-mcp (npm).

This skill is chat-first:

• “Generate an XMind for this (optionally: save to …)” → generates a .xmind and sends it back.
• “Read this XMind and tell me what it’s about” → explains the map first, then offers a Markdown export.

When to use

• User wants: “generate an XMind file from this outline / plan / PRD / test plan”.
• Output should be a real .xmind file that opens in XMind.

Input format (for generation)

The MCP tool generate-mind-map accepts Schema A JSON:

{
  "title": "Root topic",
  "filename": "mindmap-name-no-date",
  "topics": [
    {
      "title": "Topic title (keep core info in the title)",
      "note": "Optional: use sparingly; only when the title would be too long",
      "labels": ["optional"],
      "markers": ["Arrow.refresh"],
      "children": [{"title": "Child topic"}]
    }
  ]
}

How the assistant should call the MCP (stdio via mcporter)

Generate XMind

Use npx xmind-generator-mcp@0.1.2 as the MCP server command:

mcporter call --stdio "npx -y xmind-generator-mcp@0.1.2" generate-mind-map --args '{...}'

Read XMind (Markdown export)

mcporter call --stdio "npx -y xmind-generator-mcp@0.1.2" read-mind-map --args '{"inputPath":"/path/to/file.xmind","style":"A"}'

Read behavior in chat (UX requirement)

When a user sends an .xmind file and asks to “read/understand” it: 1) First, explain in detail what the mind map is about (summary + structure + key points + actionable items + gaps). 2) Only after that, ask whether the user wants an export. 3) If the user wants an export, default to Markdown.

• generate --output can be a directory (recommended) or a full .xmind path.
• If outputPath is omitted, the MCP server defaults to its configured outputPath environment variable (see below).

Chat-first workflow (what the assistant should do)

Trigger phrases:

• “Generate an XMind for this” / “Make an XMind from this”
• “Read this XMind” / “Summarize this XMind”

Language rule (important):

• If the user does not explicitly specify the language of the generated mind map, match the language of the user’s request.

- Example: user asks in Chinese → generate Chinese topic titles. - Example: user asks in English → generate English topic titles.

• Only switch languages if the user explicitly asks (e.g. “generate in English”).

Filename rule (important):

• If the user provides a filename/path, use it.
• If the user does not provide a filename:

- Default filename should follow the user’s request language. - For Chinese: use short hyphen style, no date (e.g. one-day-trip-detailed in Chinese characters, but keep it short). - For English: use a short slug, no date (e.g. hong-kong-1-day-itinerary).

• Always sanitize invalid filename characters: \\ / : * ? \" < > | → replace with -.

Steps: 1) Parse whether the user specified a save location/path. - If specified, use that. - If not specified, default to ~/Desktop. 2) Determine output language: - If user specified language → follow it. - Else → match the user’s message language. 3) Convert the user’s content into Schema A JSON (in the chosen language). - Keep it reasonably sized (avoid thousands of nodes in one go). - Use XMind enrichment elements when helpful (but keep titles informative): - note: use sparingly. Only put content in note when it would make the topic title too long or noisy. If the content is core to the plan, keep it in the title instead of hiding it in note. - labels: add lightweight categorization (e.g. must-do/optional/rainy-day/budget/family-friendly). - relationships: sparingly (mode A) — only for truly cross-branch links that add clarity. 4) Write the JSON to a temp file, e.g. /tmp/xmind-<ts>.json. 5) Call the MCP tool generate-mind-map via mcporter stdio (npx xmind-generator-mcp@0.1.2). Capture/compute the output file path. 6) Send the generated .xmind back in chat as an attachment. 7) Optionally also tell the user where it was saved on disk.