local-ocr: Offline Local OCR Skill

`local-ocr` is an offline local OCR plugin designed specifically for DeepSeek Harness to process images or PDFs. It prioritizes the native Windows OCR engine, which offers high Chinese recognition accuracy, and supports RapidOCR and Tesseract as fallback options, ensuring free access to structured text in offline environments. Installation requires Python 3.11. The plugin is invoked via CLI and supports selecting an engine, preprocessing, and self-testing. Output is in JSON format: image results include text with line-level scores, while PDFs are processed page by page. Note that the native Windows OCR engine does not provide true confidence scores.

Read More