atomicbot-ai/atomic-chat/imagemagick

imagemagick

Transform still images with ImageMagick v7 (magick command). Pairs well with vision.describe — describe an image to decide what to do, then operate on it here. Always write to a new output file.

ImageMagick v7 uses magick in.png out.jpg. Legacy v6 uses convert; if only

convert exists, substitute it for magick in the examples below.

Setup health check (run first, every session)

Verify with one solo step:

[{ "tool": "os.shell.run", "args": { "cmd": "magick", "args": ["--version"] } }]

Outcome map:

• exit 0 + Version: ImageMagick 7… → ready, proceed.
• command not found: magick → try convert --version (v6); if also missing →

enter Setup playbook → "imagemagick missing".

Setup playbook (when prerequisites are missing)

imagemagick missing

Reply (solo reply step):

"ImageMagick is not installed. I can install it: brew install imagemagick. Install it?"

On yes:

[{ "tool": "os.shell.run", "args": { "cmd": "brew", "args": ["install", "imagemagick"] } }]

On Linux: apt-get install imagemagick.

On Windows, offer this exact winget command after confirmation:

[{ "tool": "os.shell.run", "args": { "cmd": "winget", "args": ["install", "--id", "ImageMagick.ImageMagick", "-e"] } }]

When to use

• "Resize / crop / rotate this image", "convert PNG to JPG / WebP".
• "Compress this image", "make a thumbnail", "combine images into a montage".
• "Add a text caption / watermark to this image".

When NOT to use

• Understanding image content — use vision.describe instead.
• Video frames / animation encoding — use the ffmpeg skill.
• Multi-page PDF assembly from images — use the pdf skill (magick can make a

PDF but page-level control lives in pdf).

Common operations

All examples invoke os.shell.run with cmd: "magick". The approval gate surfaces each write.

Rules

1. Always write to a NEW output path; never overwrite the source image.
2. Use magick identify first to learn dimensions/format before transforming.
3. After a transform, optionally vision.describe the result to confirm it

matches the user's intent for visual tasks.

1. Echo the output path and final dimensions after each operation.