crop-tool

SKILL.md

Crop Tool

When to Use

  • Reading small text, legends, or axis labels in charts
  • Extracting values from dense tables or diagrams
  • Any region where initial analysis feels uncertain

How to Use

Simply ask naturally about the image. Claude decides which regions to examine and handles the cropping automatically — no coordinates or technical input needed from the user.

"Read the values in the legend."
"What does the small text in the bottom section say?"
"Compare the metrics across all three columns."

Skill Mechanics

See Giving Claude a crop tool for better image analysis for a detailed understanding of the mechanics.

In addition to cropping, this skill enhances the extracted region further:

  • 2x upscale — using LANCZOS resampling for maximum quality
  • 1.4x contrast boost — makes text and edges pop
  • 1.3x sharpness boost — reduces blur introduced by upscaling

How Claude Must Invoke the Tool

IMPORTANT: Always use the skill's crop_tool.py CLI via Bash — do NOT write your own PIL/image-processing code.

python3 /Users/jdl/.claude/skills/crop-tool/crop_tool.py \
  <image_path> <x1> <y1> <x2> <y2> [--output <out_path>]
  • Coordinates are normalized (0–1), where (0,0) is top-left and (1,1) is bottom-right
  • The script prints the output file path; read that file with the Read tool to view the cropped region
  • Iterate across regions as needed until the question is fully answered

Example:

python3 /Users/jdl/.claude/skills/crop-tool/crop_tool.py \
  chart.png 0.5 0.2 1.0 0.65 --output /tmp/legend_crop.png

For full documentation see README.md.

Weekly Installs
20
First Seen
Feb 27, 2026
Installed on
opencode20
gemini-cli20
claude-code20
github-copilot20
codex20
kimi-cli20