XMind Mind Map Skill

Create, read, and analyze .xmind mind map files. Supports natural Chinese and English interaction.

Natural language routing

This skill supports both Chinese and English. When the user makes a request, identify the intent and route to the correct script and action. Respond in the same language the user uses.

Creation intents → create_xmind.mjs

Read/analysis intents → read_xmind.mjs

Routing tips

• When the user provides a .xmind file path and asks to "look at" / "check" / "看看" / "打开" it, use read action
• When the user says "这个脑图里有什么关于xxx的" (what does this map say about xxx), use search_nodes
• When the user asks about task progress ("进度怎么样" / "完成了多少"), use search_nodes with searchIn: ["tasks"]
• When the user mentions multiple files, use read_multiple
• When the request is vague about location ("找到那个关于后端的节点"), use extract_node for fuzzy matching
• Always respond in the language the user used — if they speak Chinese, reply in Chinese

How to create an XMind file

1. Build a JSON object with path and sheets fields (see format below)
2. Write it to a temp file, then run:

node <skill-dir>/scripts/create_xmind.mjs < /tmp/xmind_input.json

Where <skill-dir> is the directory containing this SKILL.md file.

Updating an existing file (theme/style preservation)

If path points to a file that already exists (e.g. you're adding/editing a branch in a mind map someone already styled), create_xmind.mjs automatically preserves that file's visual theme, layout extensions, and embedded resources (like Thumbnails/thumbnail.png) — matched per-sheet by title. You do NOT need to read the old theme, back up the file, or manually merge anything yourself; just include the unchanged sheets/branches in your input JSON (with the same sheet titles) alongside the new/edited content, and the script carries the styling forward.

This only applies to the modern zen format. If you rewrite a sheet under a different title than it had before, it's treated as a brand-new sheet and gets a fresh default theme — keep sheet titles stable across updates when you want styling preserved.

Format compatibility

Both scripts handle two XMind format families:

Reading: Auto-detects format — no configuration needed. Checks for content.json first (zen), then content.xml (legacy). All 8 actions work identically on both formats.

Creating: Add "format" to the input JSON. Accepts many version aliases:

{
  "path": "/tmp/old_format.xmind",
  "format": "xmind8",
  "sheets": [...]
}

Heuristic for numeric values: version ≤ 9 → legacy (XMind 3–8), version ≥ 10 → zen (XMind 10+), year ≤ 2019 → legacy, year ≥ 2020 → zen.

Checking format: Use format_info action to inspect an existing file:

echo '{"action": "format_info", "path": "/path/to/file.xmind"}' | node <skill-dir>/scripts/read_xmind.mjs

When the user says "兼容旧版" / "XMind 8格式" / "旧版本格式" / "old format" / "XMind 8 compatible" / "兼容其他编辑器", use "format": "legacy". Otherwise use the default zen format.

JSON Input Format

{
  "path": "/Users/user/Desktop/my_mindmap.xmind",
  "sheets": [
    {
      "title": "Sheet 1",
      "rootTopic": {
        "title": "Central Topic",
        "children": [
          {
            "title": "Branch 1",
            "notes": "Plain text note",
            "children": [
              { "title": "Sub-topic A" },
              { "title": "Sub-topic B" }
            ]
          }
        ]
      },
      "relationships": [
        { "sourceTitle": "Sub-topic A", "targetTitle": "Sub-topic B", "title": "related" }
      ]
    }
  ]
}

Topic Properties

Each topic object supports:

Topic shapes

• org.xmind.topicShape.roundedRect — rounded rectangle (default)
• org.xmind.topicShape.diamond — diamond (use for conditions/decisions)
• org.xmind.topicShape.ellipserect — ellipse (use for start/end)
• org.xmind.topicShape.rect — rectangle
• org.xmind.topicShape.underline — underline only
• org.xmind.topicShape.circle — circle
• org.xmind.topicShape.parallelogram — parallelogram (use for I/O)

Layout structures

• org.xmind.ui.map.clockwise — balanced map
• org.xmind.ui.map.unbalanced — unbalanced map
• org.xmind.ui.logic.right — logic chart (right)
• org.xmind.ui.org-chart.down — org chart (down)
• org.xmind.ui.tree.right — tree (right)
• org.xmind.ui.fishbone.leftHeaded — fishbone
• org.xmind.ui.timeline.horizontal — timeline

Task properties

Simple checkbox (no dates needed):

• taskStatus: "todo" or "done"

Planned tasks (for Gantt/timeline view in XMind):

Two approaches for planned tasks:

1. Relative (preferred): Use durationDays + dependencies. XMind auto-calculates dates.
2. Absolute: Use startDate + dueDate for fixed dates.

When the user mentions "planning", "schedule", "timeline", "Gantt", "project", "phases", "排期", "甘特图", "项目计划", "时间线", "里程碑", use RELATIVE planned tasks unless specific dates are given.

Sheet properties

Logic / Flow diagrams

For flowcharts, logic diagrams, or algorithmic diagrams, use free positioning with detached topics and straight relationships:

{
  "path": "/tmp/flowchart.xmind",
  "sheets": [{
    "title": "Algorithm",
    "freePositioning": true,
    "rootTopic": {
      "title": "START",
      "shape": "org.xmind.topicShape.ellipserect",
      "structureClass": "org.xmind.ui.map.clockwise"
    },
    "detachedTopics": [
      {"title": "IS X > 0?", "position": {"x": 0, "y": 130}, "shape": "org.xmind.topicShape.diamond"},
      {"title": "PRINT YES", "position": {"x": 200, "y": 130}},
      {"title": "PRINT NO", "position": {"x": -200, "y": 130}},
      {"title": "END", "position": {"x": 0, "y": 260}, "shape": "org.xmind.topicShape.ellipserect"}
    ],
    "relationships": [
      {"sourceTitle": "START", "targetTitle": "IS X > 0?", "shape": "org.xmind.relationshipShape.straight"},
      {"sourceTitle": "IS X > 0?", "targetTitle": "PRINT YES", "title": "YES", "shape": "org.xmind.relationshipShape.straight"},
      {"sourceTitle": "IS X > 0?", "targetTitle": "PRINT NO", "title": "NO", "shape": "org.xmind.relationshipShape.straight"},
      {"sourceTitle": "PRINT YES", "targetTitle": "END", "shape": "org.xmind.relationshipShape.straight"},
      {"sourceTitle": "PRINT NO", "targetTitle": "END", "shape": "org.xmind.relationshipShape.straight"}
    ]
  }]
}

Conventions: Use ellipse for start/end, diamond for conditions, rectangle (default) for actions, parallelogram for I/O. Use "org.xmind.relationshipShape.straight" for all connectors. Position topics on a grid (y increments of ~130px, x offsets of ~200px for branches).

When the user mentions "flowchart", "algorithm", "logic diagram", "流程图", "逻辑图", "算法图", "程序流程", use this pattern.

Working with large files

When reading a PDF or other large file fails (e.g. "PDF too large"), extract text using CLI tools before building the mind map:

# Preferred: pdftotext (install: apt install poppler-utils)
pdftotext input.pdf /tmp/extracted.txt

# Fallback if pdftotext unavailable:
python3 -c "
import subprocess, pathlib, sys
p = sys.argv[1]
try:
    subprocess.run(['pdftotext', p, '/tmp/extracted.txt'], check=True)
except FileNotFoundError:
    subprocess.run(['pip', 'install', 'pymupdf'], check=True, capture_output=True)
    import importlib; fitz = importlib.import_module('fitz')
    doc = fitz.open(p)
    pathlib.Path('/tmp/extracted.txt').write_text('\n'.join(page.get_text() for page in doc))
" input.pdf

Then read /tmp/extracted.txt to build the mind map.

How to read and analyze XMind files

Pipe a JSON object with an action field to the read script:

echo '{"action": "read", "path": "/path/to/file.xmind"}' | node <skill-dir>/scripts/read_xmind.mjs

Available actions

Action details

read — Parse a complete mind map:

echo '{"action": "read", "path": "/path/to/file.xmind"}' | node <skill-dir>/scripts/read_xmind.mjs

Returns an array of sheets, each with a root topic containing the full tree structure including notes, labels, markers, task status, relationships, etc.

list — Find .xmind files in a directory:

echo '{"action": "list", "directory": "/Users/user/Documents"}' | node <skill-dir>/scripts/read_xmind.mjs

search_files — Search by filename or content:

echo '{"action": "search_files", "pattern": "project", "directory": "/Users/user"}' | node <skill-dir>/scripts/read_xmind.mjs

Filename matches are returned first, then content matches.

extract_node — Fuzzy search when you don't know the exact path:

echo '{"action": "extract_node", "path": "/path/to/file.xmind", "searchQuery": "Backend API"}' | node <skill-dir>/scripts/read_xmind.mjs

Returns top 5 matches with confidence scores. Use when exploring complex maps.

extractnodeby_id — Direct ID lookup (fastest):

echo '{"action": "extract_node_by_id", "path": "/path/to/file.xmind", "nodeId": "abc123def456"}' | node <skill-dir>/scripts/read_xmind.mjs

search_nodes — Advanced search with filters:

echo '{"action": "search_nodes", "path": "/path/to/file.xmind", "query": "auth", "searchIn": ["title", "notes"], "taskStatus": "todo"}' | node <skill-dir>/scripts/read_xmind.mjs

searchIn options: title, notes, labels, callouts, tasks. Default: all fields.

When to use each action

• Browsing/understanding a map: Use read to get the full structure
• Finding .xmind files: Use list or search_files
• Navigating to a specific topic: Use extractnode (fuzzy) or extractnodebyid (exact)
• Finding content across a map: Use search_nodes with appropriate searchIn filters
• Filtering tasks: Use search_nodes with taskStatus: "todo" or "done"
• Comparing maps: Use read_multiple to load several files at once

Important rules

• NEVER use unzip, zipinfo, or any external ZIP tool on .xmind files. Both scripts handle ZIP internally — createxmind.mjs writes ZIP with built-in code, readxmind.mjs reads ZIP with built-in code. Just pipe JSON to the script and it handles everything. There is no need to extract, decompress, or inspect the .xmind file yourself.
• When editing/adding to an existing .xmind file, always include every sheet with its original title — even sheets you aren't changing — so create_xmind.mjs can match them up and preserve their theme/styling/thumbnail (see "Updating an existing file" above). Only omit or retitle a sheet when you intentionally want it reset to default styling.
• The output path MUST end with .xmind
• Always write the file where the user requests (e.g. ~/Downloads, ~/Desktop)
• IDs are generated automatically
• Topic references in relationships and dependencies are resolved by title
• HTML notes: only <strong>, <u>, <ul>, <ol>, <li>, <br> are supported. <code> is NOT supported by XMind.
• Internal links (linkToTopic) work across sheets
• Notes should be substantial and detailed — don't just repeat the topic title. Use notes to add explanations, context, definitions, examples, key points, or reasoning. Aim for 2-5 sentences minimum per note. Use HTML notes with <strong>, <ul>/<li>, <br> for well-structured content. Most topics should have notes unless they are self-explanatory leaf nodes.
• The read script outputs JSON to stdout — pipe or redirect as needed
• All read actions require the action field in the input JSON