🧠 mindmap

Understand anything at a glance — turn a file, a URL, or a topic into a zoomable mindmap without leaving your AI coding agent.

English · 中文

mindmap is a plugin for Claude Code and GitHub Copilot. Point it at a dense report, a long article, or just a topic, and it distills the key ideas into a clean, zoomable Markmap mindmap — right from your terminal.

Two languages in one plugin: /mindmap (English) and /mindmap-zh (中文).

From any source to a map — in one command

flowchart LR
    IN["Input<br/>file / URL / text / topic"] --> CL{Classify}
    CL -->|URL| FE[Fetch page]
    CL -->|file| RD[Read file]
    CL -->|prose| TX[Use as text]
    CL -->|topic| KN[Model knowledge]
    FE --> ST["Build hierarchy<br/>4-7 branches, depth 3-4"]
    RD --> ST
    TX --> ST
    KN --> ST
    ST --> WR["Write Markmap .md"]
    WR --> RN{--render?}
    RN -->|yes| HT["Standalone .html"]
    RN -->|no| MD[".md deliverable"]

---

👀 See it

One command turns a long GitHub blog post into a mindmap:

/mindmap https://github.blog/ai-and-ml/how-we-made-github-copilot-cli-more-selective-about-delegation/ --render

A ~1,500-word article becomes a structure you can read in seconds:

# Smarter Subagent Delegation
├── The Problem
│   ├── Delegation is powerful but not free
│   └── Unnecessary handoffs, overlapping searches, waiting
├── The Approach
│   ├── Analyze → find the delegation bottleneck
│   ├── Change → handle focused work directly
│   └── Validate → offline, then online, then ship
├── Results
│   ├── Tool failures per session −23%
│   └── Wait time −5% P95, no quality regression
└── What's Next

That's real output — see examples/ for the full Markmap .md plus the rendered interactive .html (open in any browser to zoom and collapse branches).

▶ Open the live interactive mindmap — no install, just click.

---

✨ Why use it

• Grasp dense material fast — collapse a 3,000-word report into 5–7 branches you can scan at a glance, instead of reading top to bottom.
• One command, any source — a file, a URL, pasted notes, or just a topic. No copy-pasting into a separate web tool.
• Stays in your workflow — runs inside Claude Code / GitHub Copilot; the map lands as a file right next to your work.
• Smart structure, not a text dump — mirrors a structured doc's outline, or distills loose prose into 4–7 concise branches. Nodes are short phrases, not sentences.
• Portable, interactive output — standard Markmap .md that opens anywhere, plus an optional standalone .html you can share.

Good for: skimming research papers and reports · digesting blog posts and docs · outlining a topic before you write · turning meeting notes into a shareable map.

---

📦 Install

The same repo is a valid plugin for both Claude Code and GitHub Copilot — they share the plugin/marketplace format.

Claude Code

/plugin marketplace add https://github.com/galiacheng/mindmap-skills.git
/plugin install mindmap@mindmap-marketplace
/reload-plugins

The explicit https:// URL avoids an SSH clone. The galiacheng/mindmap-skills shorthand also works if you have GitHub SSH keys configured; otherwise it fails with Permission denied (publickey).

GitHub Copilot

copilot plugin marketplace add galiacheng/mindmap-skills
copilot plugin install mindmap@mindmap-marketplace

Then run /mindmap in your session. On GitHub Copilot the skill uses the same workflow; tool names map automatically (see skills/mindmap/references/copilot-tools.md).

The marketplace manifest lives at .claude-plugin/marketplace.json.

Manual (local, no marketplace)

Copy the skill into your project's (or user-level) skills directory:

# project-local
mkdir -p .claude/skills
cp -r skills/mindmap .claude/skills/

# or user-level (available in every project)
mkdir -p ~/.claude/skills
cp -r skills/mindmap ~/.claude/skills/

Then run /reload-skills (or restart Claude Code). Confirm with /help that /mindmap is listed.

---

⚡ Usage

/mindmap <input> [--panel] [--render] [--output <path>]

Input	Example	Behavior
File	/mindmap report.md	Reads the file, mirrors its structure, condenses nodes. Writes report.mindmap.md.
URL	/mindmap https://example.com/article	Fetches the page (WebFetch), maps its content. Writes <page-slug>.mindmap.md.
Pasted text	/mindmap "notes about X, Y, Z..."	Extracts concepts into branches. Writes <title-slug>.mindmap.md.
Topic	/mindmap "vector databases"	Generates a map from the model's knowledge. Writes vector-databases.mindmap.md.

Flags

• --panel — design the structure with a multi-agent judge panel; best for complex or important sources (see The judge panel below).
• --render — after writing the .md, also produce a standalone interactive .html (needs Node.js / npx).
• --output <path> — write the .md to a specific path instead of the default.

---

🧑‍⚖️ The judge panel (--panel)

For complex or high-stakes sources — papers, long reports — add --panel. Instead of designing the structure in a single pass, the skill runs a multi-agent panel: 3 proposers → 3 judges → 1 synthesizer. They compete on structure and pick the best format for every node.

flowchart TB
    SRC["Source content + skill rules"]
    subgraph PROPOSE["1 - Propose (3 lenses)"]
        P1["Proposer A<br/>narrative-first"]
        P2["Proposer B<br/>data-first"]
        P3["Proposer C<br/>audience-first"]
    end
    subgraph JUDGE["2 - Judge (score 10 dimensions)"]
        J1["Judge 1"]
        J2["Judge 2"]
        J3["Judge 3"]
    end
    subgraph SYNTH["3 - Synthesize"]
        SY["Top proposal = spine<br/>+ graft judges' best ideas"]
    end
    SRC --> P1 & P2 & P3
    P1 & P2 & P3 --> J1 & J2 & J3
    J1 & J2 & J3 --> SY
    SY --> OUT["Final Markmap .md"]

1 · Propose — three agents each design a full structure through a different lens:

Lens	Designs the map around...
Narrative-first	the source's own arc and section order
Data-first	the headline numbers and evidence, every metric bolded
Audience-first	the punchline first, every leaf legible on a slide

Each proposer also tags every node with a format and a render tier:

Format	Tier	Best for
bullet list	core	parallel facts
bold inline	core	headline metric, emphasis
link	core	references, repo, arXiv
table	rich	comparisons, benchmark numbers
code block	rich	formulas, reward functions, code
checkbox	rich	tasks, limitations, checklists

2 · Judge — three judges independently score every proposal 1–5 on ten dimensions (branch count, depth, phrasing, legibility, source fidelity, punchline-first, headline numbers, leaf legibility, visual balance, format fit) and name each proposal's single best idea.

3 · Synthesize — one agent takes the top-scored proposal as the spine, grafts in the best ideas the judges flagged from the other two, and emits the final Markmap .md.

Cost: the panel spawns ~7 agents and is token-intensive — reserve it for sources worth the spend. Without --panel, the skill does a fast single-pass build. The exact prompts and schemas live in skills/mindmap/references/judge-panel.md.

Both real examples in examples/ were generated with --panel.

---

🗂️ Output format

The skill writes Markmap-flavored markdown — a single # root, ## branches, and nested bullets:

---
title: Retrieval-Augmented Generation
markmap:
  colorFreezeLevel: 2
  maxWidth: 300
---

# Retrieval-Augmented Generation

## Indexing
- Chunk documents
- Embed chunks
- Store vectors

## Retrieval
- Embed the query
- Find nearest chunks

## Generation
- Inject context into prompt

Viewing the .md:

• Paste it at markmap.js.org, or
• Open it in VS Code with the Markmap extension, or
• Use --render to generate an .html you can open in any browser.

---

🌐 Rendering to HTML

/mindmap "transformer attention" --render

This runs npx markmap-cli <file>.md -o <file>.html --no-open under the hood (via skills/mindmap/scripts/render.sh).

• The .md is always the guaranteed deliverable. Rendering is best-effort.
• If npx / Node.js isn't installed, the skill still writes the .md, reports that rendering was skipped, and prints the exact command to run manually — nothing is lost.

Requirement: Node.js (provides npx). No global install needed — npx fetches markmap-cli on demand.

---

🛠️ Contributing

Curious how the skill is wired, or want to run the tests? See CONTRIBUTING.md for the project layout, the prompt-driven design, and the bash test suite.

---

📄 License

MIT © 2026 Haixia Cheng