How to use the Inkscape Vector Editing CLI agent skill

What the skill makes the agent do

Normally you would draw in the Inkscape app by clicking through menus. With this skill, the agent runs commands of a tool called cli-anything-inkscape instead. The tool directly edits SVG (XML) documents and stores its state in a JSON project file, so work can be tracked and undone.

The command groups cover document setup, shapes (rectangle, circle, ellipse, line, polygon, path, star), text, styles (fill, stroke, opacity), transforms (translate, rotate, scale, skew), layers, path boolean operations, gradients, export to PNG, SVG or PDF, and session undo and redo. Progress is saved in a JSON project file, with up to 50 levels of undo history.

Install the skill and its requirements

Before installing the skill, make sure you have Python 3.10 or newer and Inkscape installed on your system. The CLI itself is installed from the pip package named cli-anything-inkscape. The skill will not work without Inkscape and these Python packages.

To add the skill, download the ZIP from this page and unzip it into one of these locations: ~/.claude/skills/ for personal use, or a project's .claude/skills/ folder to share it with one project. You can also upload the skill in the Claude apps under Settings → Capabilities → Skills.

• Python 3.10 or newer
• Inkscape installed on the system
• pip package: cli-anything-inkscape

Trigger the skill with natural requests

You do not need to type commands yourself. Describe the graphic you want in plain language and mention Inkscape or the vector work, and the agent can choose the matching commands. Give concrete details such as canvas size, text, colours, and the export format and file name.

Ask for a result you can check. For example, you can request a title card at a given size with a wrapped headline, then ask for a PNG export. The agent can also inspect the project with JSON output to report what it changed.

• Make a 1280x720 title card with a wrapped headline in a dark blue fill, then export it to PNG.
• Add three circles to a new layer named Icons and export the project to SVG.
• Undo the last change and show me the project's shape list.

Examples of commands the agent runs

You can run the same commands yourself to see what the agent does. Create a project file, then export it. Use the --overwrite flag only when you are sure the output file can be replaced.

For title cards and panels with long copy, the source recommends wrapped text boxes rather than manual line breaks. Set the box width, box height and line height when adding text, and adjust them later with text set.

Safety notes: scripts and files

This skill runs scripts on your computer, and according to its listing it does not use the network. The agent writes and overwrites project and export files on your disk, so keep backups of any sensitive or production files before you let it work on them.

Ask the agent to use absolute paths for file operations, as the skill's own guidance recommends, so it is clear which files are changed. Check the output files exist after an export.

• Keep backups of important project and export files.
• Review the file paths the agent plans to write before it runs.
• Use --json output so the agent can read errors; non-zero return codes signal failure.

Tips for better results

Expect several rounds of adjustment. The commands are structured operations, so the agent may need a few iterations to reach the look you want. Give specific values for positions, sizes and colours to reduce this.

If you want to inspect or fine-tune by hand, you can start the interactive mode, which has tab-completion, history, and undo and redo support. The skill's README and TEST.md files in the package hold more detail.

• Use wrapped text boxes for long headlines so they do not run off the canvas.
• Use undo or history to step back after an unwanted change.
• Save the project file regularly with the save command.