drawio-skill Claude skill

How drawio-skill routes a request

The SKILL.md opens with a routing table, and the route decides which reference file the agent reads first. The main entrypoint is scripts/diagramctl.py, which handles generation, incremental sync, multi-view projection, queries, tests, reviews, failure analysis and publishing over a shared Diagram IR.

For a prompt-driven diagram, the README describes this loop:

1. Describe the system, for example: "Draw a C4 container diagram for a web app, an API service and a Postgres database."
2. The skill plans the layout, writes the XML and exports a PNG.
3. It checks its own PNG and fixes overlaps, clipped labels and stacked edges, running up to 2 rounds of fixes.
4. You review the result and ask for changes. It accepts up to 5 rounds of feedback.

When the source changes later, diagramctl sync updates the nodes and relations that changed and keeps your tuned coordinates, styles and annotations. Removed nodes are not dropped silently. They stay reviewable by default.

Supported diagram types, sources and exports

The skill ships 11 diagram type presets: ERD, UML Class, Sequence, C4, Architecture, ML/Deep Learning, Flowchart, SysML, BPMN, Network Topology and Cross-Functional Swimlane. On draw.io 30 or newer, you can also write any of 28 standard Mermaid types (mindmap, gantt, timeline, journey, pie, sankey, kanban and others) and have the CLI turn them into a laid-out, editable .drawio.

Supported source inputs:

• Import graphs for Python, JS/TS, Go and Rust, plus Python class hierarchies, laid out with Graphviz and grouped into nested module containers
• Terraform, Kubernetes and docker-compose configs rendered with official AWS, Azure, GCP and K8s icons
• Snapshots of what is actually deployed, taken from terraform show -json, docker inspect or kubectl get -o json
• SQL DDL to ER diagrams, OpenAPI/Swagger to API diagrams colored by HTTP method, GraphQL SDL to entity type diagrams
• GitHub Actions and GitLab CI configs rendered as pipeline DAGs

For vendor icons, scripts/shapesearch.py searches an index of 10,446 real draw.io palette shapes, each with its exact style and size. The skill's shape reference tells the agent to prefer a searched style over a hand-written one.

Output options include an interactive HTML viewer with pan, zoom and search, a PowerPoint deck, animated data-flow SVG, Mermaid or Markdown export, a click-through runbook, and an exec-summary compression. Built-in style presets such as dark and corporate restyle a diagram, and bilingual label variants leave the layout untouched.

Setup gotchas

After installing the draw.io desktop app, run drawio --version to confirm the CLI is on your path. Version 29 and earlier lack both Mermaid conversion and the ELK --layout pass. On WSL2, the CLI is the Windows desktop executable, so plan for that path rather than a Linux binary.

The SKILL.md frontmatter sets disable-model-invocation: true. On hosts that honor that field, the agent will not load the skill on its own, so you need to call it explicitly. The frontmatter also requests Bash, Read, Write and WebFetch tools.

On macOS, the metadata lists optional Homebrew formulas for drawio (native exports) and graphviz (automatic layout). Without them, the core IR, XML, sync, query, test, review and Story workflows still run on Python 3 alone.

According to the install docs, the recommended installer has its own update command. SkillsMP is a listing only and does not install or update the skill, so a listing there does not mean you have the skill installed.