aipptskill.

Draw.io Architecture Diagram Skill

Generate Draw.io diagrams for ML models, network architectures, and algorithm flows. Two modes — generate from scratch or transfer-style from a reference image. By 小帅同学 (@xstongxue).

By 小帅同学 (@xstongxue)

Source: github.com

Install

npx degit xstongxue/best-skills/skills/drawio-diagram ~/.claude/skills/drawio-diagram
Draw.io Architecture Diagram Skill

drawio-diagram is the formal sibling of excalidraw-diagram. Output is standard .drawio XML — opens in app.diagrams.net, the VS Code extension, or Confluence’s Draw.io integration. Editable, not flattened.

Two modes

ModeTrigger
Generate from scratch”draw an architecture diagram”, “visualize this model”
Style migrationUser provides a reference image + content; output mimics the reference’s layout, palette, and style

What it gets right

  • Strict XML correctness — properly closed tags, escaped specials, sequential IDs
  • Standard mxfile structure with mxGraphModel, root, vertex/edge cells
  • ML-aware shapes — knows how to depict attention heads, conv layers, residual connections, normalization

Where it shines

  • Academic papers that need formal architecture diagrams
  • Tech blogs that need consistent visual styling across posts
  • Documentation where you’d be using Draw.io anyway

Style migration tip

If your team has a “house style” (specific palette, specific shape vocabulary), take a screenshot of one existing diagram, give it to the skill alongside the new content, and it’ll match. Useful for keeping a thesis or report’s diagrams visually consistent.

Pairs with

  • excalidraw-diagram — when you want hand-drawn vibe instead
  • pptgen-drawio — embed these diagrams into the deck
  • Confluence / Notion — both render Draw.io natively

Install on GitHub

Example prompt

Use drawio-diagram to visualize a Transformer encoder block:
Input → Embedding → Multi-Head Attention → FFN → LayerNorm → Output.
Show residual connections as dashed lines. Layered top-down layout.

Related skills