Image Reader

Extract text from images using OCR (Optical Character Recognition). Use this skill when you need to read text content from images, screenshots, photos, or an...

Security Scan

The skill's requirements, code, and runtime instructions align with an OCR reader: it runs local OCR using RapidOCR, asks no unrelated credentials, and does not appear to perform unexpected file access or network exfiltration.

Purpose & Capability

Name/description match the included code and SKILL.md. The script performs image OCR using RapidOCR and Pillow, which is consistent with an 'Image Reader' skill. No unrelated binaries, config paths, or credentials are requested.

Instruction Scope

SKILL.md and the script only operate on the provided image path and return structured JSON. The README notes that the first run will download OCR model files (~50MB); that implies network access to fetch model weights, which is expected for many ML packages but worth being aware of (the code itself does not upload images to external APIs).

Install Mechanism

There is no packaged install spec; SKILL.md instructs users to pip install rapidocr, onnxruntime, and pillow. Installing from PyPI is normal for Python tools, but third-party packages may download additional runtime assets (the models). Verify the upstream package sources and signatures if you require higher assurance.

Credentials

The skill requires no environment variables, credentials, or config paths, and the code does not read environment variables or access unrelated system files.

Persistence & Privilege

The skill is not configured as always-on and does not request persistent agent-wide configuration or elevated privileges. It only runs when invoked.

Assessment

This skill appears to be what it claims: a local OCR reader. Before installing, consider: (1) the pip packages (rapidocr, onnxruntime) will be installed and may download model files from the internet on first run—review the package source and where it pulls models from if you need to trust the host; (2) run the tool in a virtual environment or isolated environment if you are concerned about third-party packages; (3) the script processes images locally and does not itself transmit images to external services, but verify network activity of the rapidocr package if you need to ensure full offline operation; (4) avoid feeding sensitive images until you confirm where model assets come from and whether any telemetry is enabled by the upstream packages.

Image Reader - OCR Text Extraction

A high-performance OCR skill for extracting text from images. Powered by RapidOCR with PP-OCRv4 models, supporting Chinese and English text recognition.

Features

• Multi-language: Chinese (simplified/traditional), English, and mixed text
• High accuracy: PP-OCRv4 model with >95% accuracy on typical screenshots
• Structured output: Text with confidence scores and bounding boxes
• Image info: Dimensions, format, and color mode included
• Fast: CPU-only, no GPU required

Quick Start

python scripts/read_image.py /path/to/image.jpg

Usage Examples

Extract text from a screenshot

python scripts/read_image.py screenshot.png

JSON Output

The script outputs structured JSON:

{
  "success": true,
  "text": "Full extracted text",
  "lines": [
    {
      "text": "Individual line",
      "confidence": 0.98,
      "box": [[x1,y1], [x2,y2], [x3,y3], [x4,y4]]
    }
  ],
  "line_count": 5,
  "image_info": {
    "format": "PNG",
    "size": [1920, 1080],
    "mode": "RGB"
  }
}

Requirements

pip install rapidocr onnxruntime pillow

First run will download OCR models (~50MB) automatically.

Common Use Cases

• UI Screenshots: Extract text from app/website screenshots
• Document Photos: Read text from photographed documents
• Diagrams: Extract labels and annotations
• Receipts: Parse receipt/invoice data

Output Fields

Field	Type	Description
success	bool	Whether OCR succeeded
text	string	All extracted text
lines	array	Individual text lines with metadata
line_count	int	Number of text lines detected
image_info	object	Image metadata

Technical Details

• Engine: RapidOCR (ONNX Runtime backend)
• Models: PP-OCRv4 (detection + recognition)
• Languages: Chinese, English (auto-detected)
• Performance: ~1-2 seconds per image on CPU

License

MIT License

Third-party dependencies:

• RapidOCR - Apache 2.0 License
• ONNX Runtime - MIT License