How to use the Draw.io Command-Line Diagrams agent skill

What does this skill make the agent do?

The skill teaches an AI agent to work with Draw.io diagrams by running a command-line program called cli-anything-drawio. Once installed, the agent can create a new diagram project, add shapes and connectors, move or restyle them, manage multiple pages, undo and redo changes, and export the result to PNG, PDF or SVG.

The program also has a machine-readable output mode that returns results as structured JSON, which is easier for software to read.

• Project: create, open, save, inspect and print the raw XML of a diagram
• Shape: add, remove, list, label, move, resize and style shapes
• Connect: add, remove, label and style connectors between shapes
• Page: add, remove, rename and list pages
• Export: render to a file and list available formats
• Session: check status, undo, redo and save state

Install the Draw.io skill in Claude

You can add the skill in one of two ways. Download the ZIP file from this page and then either unzip it into your personal skills folder at ~/.claude/skills/, or unzip it into a project's .claude/skills/ folder so only that project uses it.

If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

The skill's own instructions require the CLI to be installed with pip. Install the package with the command below, and make sure Draw.io (drawio) is installed on your system as well.

• Python 3.10 or newer is required
• Draw.io (drawio) must be installed on the system
• The pip package is named cli-anything-drawio

How do you trigger the skill with a natural request?

You do not need to type commands yourself. Describe the diagram you want in ordinary language and mention Draw.io, so the agent knows to use the skill. The agent then decides which cli-anything-drawio commands to run.

Be specific about the file, the shapes, and the output format. Asking for a PNG, PDF or SVG export tells the agent exactly what final file you expect.

• Create a Draw.io project called system-overview.json with a web app, a database and a queue, and export it to PNG.
• Open flow.json in Draw.io, rename page 1 to Login, and export the diagram to PDF.
• Add a connector from the API shape to the database shape, then undo that change if it looks wrong.
• Return the project information as JSON so my script can read it.

Run the CLI yourself to check what the agent does

You can run the same commands by hand to see or verify the agent's work. Show the help text first, then create a project file and export it. The export command takes the project file with the --project option, the output file name, and an optional --overwrite flag.

Use --json before the command name when you want structured output. The skill's guidance says to use absolute paths for file operations and to confirm that output files exist after an export.

• Run cli-anything-drawio --help to see the available commands
• Run cli-anything-drawio with no subcommand to enter the interactive REPL, which supports tab-completion and history
• Run cli-anything-drawio project new -o myproject.json to create a blank diagram
• Run cli-anything-drawio --project myproject.json export render output.pdf --overwrite to export it

What are the safety notes for this skill?

This skill runs scripts. The listing states that it runs scripts and does not use the network. The agent will execute the cli-anything-drawio program on your machine, so only install it in environments where running that software and installing Python packages is allowed.

The skill file does not say whether diagram data leaves the machine. If your diagrams are sensitive, review the package documentation before you use it with confidential content.

• Use absolute paths so the agent writes to the files you intended
• Check return codes: 0 means success, non-zero means an error
• Parse stderr for error messages when a command fails
• Verify that exported files exist after each export

Tips for getting reliable diagrams

Keep your diagram in a project file so you can reopen, change and re-export it later. Use undo and redo while you work, noting that the session keeps up to 50 levels of history. For scripted use, request JSON output so the results can be parsed reliably.

The package includes further documentation in README.md, test coverage in TEST.md and methodology in HARNESS.md under the cli-anything-plugin, so check those files when you need detail beyond this guide.

• Save the project after important changes
• Use the presets command to list available page size presets
• Use the formats command to list available export formats
• Use the styles command to list edge styles for connectors