← all repositories
Agents365-ai/drawio-skill

Diagrams that critique themselves before you do

An agent skill that generates draw.io diagrams from plain English, then reads its own PNG output to fix overlaps and clipped labels.

★10.4k stars Python Coding Assistants
drawio-skill
Velocity · 7d
+19
★ / day
Trend
↘cooling
star history

What it does

drawio-skill is an agent-compatible skill (SKILL.md format) that turns natural language into .drawio XML and exports to PNG, SVG, PDF, or JPG via the draw.io desktop CLI. It also reverse-engineers existing codebases — Python, JS/TS, Go, Rust — into auto-laid-out structure diagrams using Graphviz and transitive reduction.

The interesting bit

The self-check loop is the unusual part: it renders a draft PNG, reads it back, and auto-fixes up to six visual defects (overlaps, clipped labels, stacked edges) across two rounds before showing you anything. If you still want changes, it runs a targeted refinement loop for up to five more rounds. Most diagram generators ship the first draft and hope; this one treats layout quality as a testable artifact.

Key highlights

  • Six diagram presets: ERD, UML class/sequence, architecture, ML/deep learning, flowchart
  • 10,000+ official draw.io shapes searchable by name — no more guessing mxgraph.* style strings for AWS or Kubernetes icons
  • 321 AI/LLM brand logos (OpenAI, Claude, Ollama, LangChain, etc.) via aiicons.py, since draw.io ships none natively
  • Codebase extractors for import graphs and Python class hierarchies, with nested module containers and deterministic validation
  • Style presets learned from existing .drawio files or flat images; built-in default, corporate, handdrawn
  • Runs across Claude Code, Cursor, Copilot, OpenClaw, Codex, Hermes — no MCP server, no background daemon

Caveats

  • Requires the draw.io desktop CLI installed separately; headless Linux needs xvfb
  • Optional Graphviz dependency for codebase auto-layout; everything else works without it
  • AI brand logos default to CDN references (unpkg), so offline rendering needs the --embed flag

Verdict

Worth a look if you generate architecture or ML diagrams regularly and are tired of hand-tweaking node positions. Less compelling if you already live in Mermaid for README embeds or need real-time collaborative whiteboarding — the related excalidraw-skill and mermaid-skill cover those.

Frequently asked

What is Agents365-ai/drawio-skill?
An agent skill that generates draw.io diagrams from plain English, then reads its own PNG output to fix overlaps and clipped labels.
Is drawio-skill open source?
Yes — Agents365-ai/drawio-skill is open source, released under the MIT license.
What language is drawio-skill written in?
Agents365-ai/drawio-skill is primarily written in Python.
How popular is drawio-skill?
Agents365-ai/drawio-skill has 10.4k stars on GitHub and is currently cooling off.
Where can I find drawio-skill?
Agents365-ai/drawio-skill is on GitHub at https://github.com/Agents365-ai/drawio-skill.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.