🧠 Mind Map Generator — Claude Skill

A Claude skill that generates interactive, visually polished mind maps from any content — text, documents, conversations, topics, brainstorms, or ideas.

Live demo → · Download .skill · MIT

What It Does

Feed it any content and it produces a React artifact with:

• Radial mind map with semantic grouping (related branches sit adjacent)
• Interactive controls: pan, zoom, collapse/expand, focus mode, fullscreen
• Content intelligence: contradiction detection (⚡), gap analysis (❓), cross-branch links, frequency weighting
• Progressive disclosure: toggle between keyword-only map view and full detail reading mode
• Hover tooltips with expanded context for compressed labels
• Source attribution with clickable 🔗 links on individual nodes and a footer reference list
• 4 color palettes switchable at runtime via 🎨 selector
• 7 export formats: SVG, PNG, PDF, Markdown, Mermaid, embeddable HTML
• Conversational editing: add, remove, move, rename, merge, split nodes through natural language — no full regeneration needed
• Adaptive layout: auto-selects radial, semi-circular, top-down tree, or left-to-right flow based on content shape — or override manually
• Knowledge Atlas: one private published page that stores every map in a database, links maps by shared concepts, and opens any map at #slug — Claude writes maps into it directly, so no rendering code is generated
• Markmap download: every map has a ⬇ .md button that exports a portable Markmap file, built in code with no extra model call
• Dual output: React artifact (default) for interactive exploration in Claude, or Markmap .md for portable use in VS Code, browsers, and markmap.js.org
• Deep mode: say --deep on complex documents — generates three competing structures, scores them on six dimensions, and synthesizes the strongest one

Installation

claude.ai — install the .skill file

1. Download mindmap.skill from the Releases page
2. Drag it into any Claude chat, or go to Settings → Profile → Custom Skills and upload it

Claude Code — install the plugin

/plugin marketplace add 0x-man/mindmap-skill
/plugin install mindmap@mindmap-marketplace

Manual install

Clone this repo and copy skills/mindmap/ into your Claude skills directory.

File Structure

.claude-plugin/
├── plugin.json                           # Claude Code plugin manifest
└── marketplace.json                      # Plugin marketplace catalog

skills/mindmap/                           # Single source of truth
├── SKILL.md                              # Core instructions + mandatory code
└── references/
    ├── template.md                       # Complete working React template
    ├── mindmap-best-practices.md         # Cognitive science, palettes, keyword compression
    ├── export-patterns.md                # SVG/PNG/PDF/Mermaid/Markdown/Embed export code
    ├── layout-engine.md                  # 4 layout algorithms (radial, semicircle, tree, flow)
    ├── atlas.html                        # Published Knowledge Atlas page (db-backed)
    ├── atlas-storage.md                  # Chat fallback: window.storage save and atlas
    └── judge-panel.md                    # Multi-agent deep structure mode

docs/                                     # GitHub Pages demo site

Both distribution channels read from skills/mindmap/. The .skill file attached to each release is built from this same folder — edit once, ship everywhere.

Usage

Just ask Claude to make a mind map:

• "Mind map this article" (paste or upload content)
• "Map out the key concepts of machine learning"
• "Turn this PDF into a visual summary"
• "Visualize the structure of this meeting transcript"

Then build your knowledge graph over time:

• "Save this map" — persists across sessions
• "Show my atlas" — see all your maps as a connected network
• "Open my Deep Work map" — reload a saved map
• "What connects to my crypto map?" — discover cross-map links

Or export as a portable Markmap file:

• "Mind map this article --md" — generates Markmap directly
• "Convert to markmap" — converts an existing React map
• "Make it portable" — same as above

For dense documents where structure really matters:

• "Mind map this paper --deep" — runs the judge panel for a better structure
• "Think harder about this one" — same as above

Features in Detail

Feature	What it does
Semantic gravity	Groups related branches adjacent on the circle
Contradiction detection	Flags conflicting claims with ⚡ and red dashed connectors
Inquiry nodes	Adds ❓ nodes where source material has genuine gaps
Cross-links	Dotted arcs between nodes in different branches
Focus mode	Click ◎ on a branch to isolate it, fading everything else
Fullscreen	⛶ button expands the map to fill the viewport; Escape to exit
Progressive disclosure	Toggle "⊕ Details" to show/hide expanded text under nodes
Palette selector	🎨 button with 4 named palettes (Bauhaus, Ocean Sunset, Nordic Forest, Pastel Garden)
Source linking	🔗 on hover opens source URL; footer shows numbered reference list
Embed export	</> button downloads a self-contained HTML file and copies an iframe snippet
Edge anchoring	Connectors start/end at pill edges, not node centers — no text overlap
Conversational editing	"Add X under Y", "move Z", "merge these two" — surgical updates without regenerating from scratch
Adaptive layout	Auto-selects radial, semi-circular, tree, or flow layout based on content. "Make it a tree" to override.
Knowledge Atlas	Where the session can publish artifacts, Claude saves each map as data in a private atlas page (owner-only access rule) and links you straight to it. In claude.ai chat, the 💾 button saves to browser-side storage instead.
⬇ .md	One-click Markmap export from the atlas or any chat map. Opens in VS Code, markmap.js.org, or npx markmap-cli.
Markmap output	Say "markmap" or "--md" to get portable Markdown that works in VS Code, markmap.js.org, and any browser via npx.
Deep mode	Say "--deep" for dense documents. Three proposers, six-dimension scoring, one synthesizer. Slower and pricier, noticeably better structure.

Token Usage Estimates

Mind map generation is a single-turn operation. Here's what to expect:

Component	First map	Follow-ups
Input: SKILL.md (always loaded)	~2,300	~2,300
Input: best-practices reference	~2,500	— (skipped)
Input: export-patterns reference	~1,400	— (skipped)
Input: layout-engine reference	~1,500	— (skipped)
Input: atlas-storage reference	~1,800	— (only when atlas requested)
Input: your content (text/PDF)	1,000–5,000	200–1,000
Input: system prompt overhead	~1,000	~1,000
Output: response + artifact	~5,500	~3,000
Total	~15,000–19,000	~7,000–8,000

In practice: The hybrid architecture (prose for content analysis, inline code for rendering) keeps SKILL.md lean at ~2,300 tokens. Reference files are loaded only on first generation, then skipped — saving ~55% on follow-ups. On Claude Pro, this is well within normal usage. On the API, expect ~$0.05–0.12 per first map and ~$0.02–0.04 per follow-up.

What affects cost:

• Longer input documents → more input tokens (but output stays ~5K regardless — the skill compresses aggressively)
• Follow-ups are significantly cheaper than the first generation
• Asking for structural changes ("add a branch", "merge these two") regenerates the full artifact (~5K output); cosmetic changes ("change palette") are lighter

Troubleshooting

Claude occasionally drops specific rendering features (save button, edge anchoring, opaque pills) when generating mind maps from scratch. This is a known limitation of LLM code generation — the content analysis and structure extraction work reliably, but visual implementation details can vary between generations.

Quick fix using inline editing (works every time):

Claude now supports inline artifact editing. If a feature is missing from your generated mind map:

1. Click the artifact to select it
2. Highlight the area where the feature should be (e.g., the toolbar region)
3. Type your instruction directly, for example:
   • "Add a 💾 Save button here that saves to window.storage"
   • "Make all pill backgrounds fully opaque with fillOpacity={1}"
   • "Fix connectors to start at pill edges instead of centers"

Claude will edit the existing artifact code — which it does reliably, since it's modifying code rather than generating from scratch.

Features that work reliably on first generation:

• Radial layout with semantic gravity
• 4–7 branch structure with keyword compression
• Content intelligence (contradictions, gaps, cross-links)
• Palette system and switching
• Pan, zoom, collapse/expand
• Hover tooltips

Features that may need an inline edit prompt:

• 💾 Save button
• Edge-anchored connectors (vs center-to-center)
• Fully opaque pills (vs semi-transparent)
• Fullscreen toggle

License

MIT