What drawio-skill does
drawio-skill converts descriptions and technical source files into editable draw.io architecture models. It supports diagram generation from natural language, Mermaid, whiteboard images, application code, infrastructure definitions, schemas, API specifications, event contracts, CI pipelines, and live infrastructure snapshots.
The resulting models can be reviewed as architecture diagrams, queried for dependencies and ownership, compared across revisions, or projected into several views. Supported view types include executive, system, deployment, data-flow, and security perspectives. The model separates semantic information, source provenance, and geometry so that later updates can preserve deliberate layout work.
How it works
The workflow is centered on the diagramctl command-line interface. Its commands cover building diagrams, synchronizing changes, generating views, querying relationships, running architecture tests, reviewing differences, simulating failure propagation, creating walkthroughs, publishing, and transforming output. Core workflows use the standard library and can run offline.
For generated diagrams, the skill writes draw.io XML, renders a PNG for inspection, and can correct overlaps, clipped labels, and stacked edges through repeated checks. Incremental synchronization updates changed nodes and relationships while retaining adjusted coordinates, styles, and annotations. Removed elements remain reviewable by default.
Source import can use Graphviz placement, transitive reduction, nested module containers, official cloud and Kubernetes icons, and deterministic layout engines. Mermaid text can be converted into native, editable draw.io diagrams when using draw.io 30 or later. Whiteboard photos and screenshots can be processed into editable diagrams while retaining the detected original arrangement.
Setup and configuration
Install the draw.io desktop CLI before using workflows that render or convert diagrams. On macOS, the project documents Homebrew installation; Windows uses the draw.io installer; Linux uses a Debian or RPM package and may require xvfb for headless operation. Verify the executable with drawio --version, or drawio.exe --version on Windows. Draw.io version 30 or later is recommended for Mermaid conversion and the ELK layout pass.
Install the skill with npx skills add Agents365-ai/drawio-skill -g, or clone the repository and place or link its skills/drawio-skill/ directory into the agent’s skills directory. The host entry must contain SKILL.md directly. The project also documents installation paths for Claude Code, OpenClaw, Codex, Autohand Code, and other Agent Skills-compatible hosts.
The repository includes an optional MCP server. It exposes the CLI workflows to MCP hosts including Claude Desktop, Cursor, VS Code, and Codex. The skill can also be used directly by compatible Agent Skills hosts without a daemon.
Tools and capabilities
- Generate editable diagrams from prompts using presets for ERD, UML class, sequence, C4, architecture, machine learning, flowchart, SysML, BPMN, network topology, and swimlane diagrams.
- Import Python, JavaScript/TypeScript, Go, and Rust codebases, including Python class hierarchies.
- Convert Terraform, Kubernetes, Docker Compose,
terraform show -json, docker inspect, and kubectl get -o json data into infrastructure diagrams.
- Build diagrams from SQL DDL, OpenAPI or Swagger, AsyncAPI, Protocol Buffers, GraphQL SDL, GitHub Actions, and GitLab CI definitions.
- Synchronize source changes while preserving manual geometry, styles, and annotations.
- Check architecture rules such as forbidden access paths, cycles, orphaned components, trust-boundary violations, and contrast issues.
- Query dependencies, owners, paths, articulation points, and coupling; compare diagrams or live snapshots; and simulate failure propagation.
- Publish interactive HTML, PowerPoint, animated data-flow SVG, Mermaid, Markdown, Story walkthroughs, and click-through runbooks.
- Apply style presets, bilingual labels, heat maps, and alternate visual treatments; resolve thousands of official draw.io shapes and AI or LLM logos.
Limitations and notes
PowerPoint output uses raster page images, so its shapes are not individually editable. Mermaid-to-draw.io conversion and the ELK layout pass require draw.io 30 or later. Headless Linux use may require xvfb. The project’s core workflows are designed to run offline, while live infrastructure diagrams depend on the corresponding local snapshots or command output.