RapidOCR

ClawHub skill for local image OCR with RapidOCR.

When to use

• The user wants text extracted from a local image file.
• The input is a local png, jpg, jpeg, webp, bmp, tif, or tiff file.
• The user wants either plain text output or structured JSON output.

Do not use

• Remote image URLs.
• PDF OCR.
• Relative paths when an absolute path is available.

Execution

1. Pass the user's original request directly to the wrapper script.
2. Run node "{baseDir}/run_rapidocr.js" "{{input}}".
3. If the host does not substitute {baseDir}, resolve the directory containing this SKILL.md and run the sibling file run_rapidocr.js from there.
4. For local testing, RAPIDOCR_PYTHON=/path/to/python can be used to force a specific interpreter.
5. Otherwise the wrapper auto-discovers python3, python, or py.
6. If the user asks for JSON, preserve the script's JSON output exactly.
7. If dependencies are missing, tell the user to run <python> -m pip install rapidocr onnxruntime with the interpreter they intend to use.
8. Publish this folder to ClawHub with slug rapidocr.

Behavior

• The wrapper extracts a local image path from natural language, JSON-like input, or a direct path.
• The wrapper only reads existing local image files.
• Default output is plain text, one recognized line per line.
• JSON mode returns text, lines, boxes, scores, and source.