How to use the GIMP CLI agent skill (cli-anything-gimp)

What is an agent skill and what does this one do?

An agent skill is a folder with a SKILL.md file that tells an AI agent how to use a specific tool. This skill covers a command-line tool called cli-anything-gimp, which is designed for AI agents and power users who need to create and manipulate images without a graphical interface.

With the skill loaded, the agent can create project files, add and arrange layers, resize, scale or crop the canvas, add text and rectangles, check media files, and render the result to a PNG or other image. The tool also keeps undo and redo history so changes can be reverted.

• Create and save a project file (for example project.json)
• Add blank layers or layers from image files, then list, move, duplicate, merge or flatten them
• Set layer name, opacity, visibility, blend mode, and offsets (negative offsets are accepted)
• Resize the canvas without scaling content, or scale canvas and content proportionally
• Crop, set colour mode, and set DPI
• Add text and rectangles (applied at render time)
• Apply and adjust filters on layers
• Export with a render command, with JSON output for programmatic use

How do you install the cli-anything-gimp skill?

You need Python 3.10 or newer. The command-line tool is installed as part of the cli-anything-gimp package, which you can install with pip.

To add the skill to your agent, download the ZIP from this page and unzip it into ~/.claude/skills/ for all your projects, or into a project's .claude/skills/ folder to limit it to one project. If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

The skill does not replace the Python package. The agent needs the cli-anything-gimp command available on your system, so install the package as well.

• Requires Python 3.10 or newer
• Install the package: pip install cli-anything-gimp
• Option A: unzip the downloaded ZIP into ~/.claude/skills/
• Option B: unzip it into a project's .claude/skills/ folder
• Option C: upload it in the Claude apps under Settings → Capabilities → Skills

What do you need installed before it works?

Python 3.10 or newer is required. GIMP is optional but recommended for native batch rendering. If GIMP is not installed, or if the project has deferred draw operations, layered compositions render through the built-in Pillow path.

The source says that for projects with deferred draw operations, the Pillow render path is the stable default even when GIMP is installed.

• Python 3.10 or newer (required)
• pip package cli-anything-gimp (required)
• GIMP (optional, recommended for native batch rendering)

How do you check that the command works?

Run the help command first to confirm the tool is on your path. Then create a test project. Use the --json flag when an agent or script will read the output, because the default output is formatted for people.

You can also start interactive mode by running the command with no subcommand. In that session you can use tab-completion and history, type help to list commands, and use undo and redo.

• Show help: cli-anything-gimp --help
• Create a project: cli-anything-gimp project new -o project.json
• Read project info as JSON: cli-anything-gimp --json project info -p project.json
• Start interactive mode: cli-anything-gimp

How do you ask the agent for image work?

Describe the result you want and the file names involved. The agent can then choose the matching commands. Good requests name the input image, the size or position you want, and the output file. Ask for a PNG if you need a specific format.

The source recommends that the agent use absolute paths for all file operations and verify that outputs exist after export. You can ask for these checks directly.

• Example request: Make a 1200 by 628 poster on a new canvas, add my logo.png as a layer at the top left, put the title text in a large size, and render it to poster.png.
• Example request: Take every PNG in this folder, resize the canvas to 1080 by 1080 without scaling content, and render each to a new file.
• Example request: Check that all media referenced in project.json exist, then render the project to final.png and confirm the file is there.
• Example request: Undo the last change and show me the undo history.

What does a layered poster workflow look like?

For compositions with overlays, the source shows a workflow in which the agent sets layer offsets, draws text on a layer, and renders the output. Negative offsets are accepted in both command mode and REPL mode.

Text and rectangles are render-time operations. They are stored as draw operations and applied when you render, so review the exported image rather than only the project JSON.

• Set layer offsets: layer set 0 offset_x -48, then layer set 0 offset_y 24
• Add text to layer 0 with draw text, giving text, position, size and colour
• Render the result with export render output.png --overwrite

What safety and review steps should you take?

The skill runs scripts, so read the SKILL.md and the package before giving an agent broad access to your files. The skill does not use the network, according to this listing. Keep the agent working in a folder you control, and use absolute paths so it writes where you expect.

Treat the exported image as the thing to review. Check return codes (0 means success), parse stderr when a command fails, and confirm the output file exists. Use undo to revert a change you do not want.

• The skill runs scripts; review them before enabling it
• Network access: none listed for this skill
• Check return codes: 0 for success, non-zero for errors
• Read stderr for error messages
• Verify output files exist after export
• Review rendered images after offset, draw, blend-mode or filter changes