This wraps Surya, a GPU-accelerated OCR engine that handles 90+ languages and claims 2x better accuracy than Tesseract. You'd reach for it when processing multilingual documents, extracting text from screenshots, or dealing with complex layouts and tables. The skill documentation is thorough about gotchas: it walks through OneDrive path issues on Windows, API breaking changes between versions (the langs argument got removed in 0.17.x), and transformers 5.x compatibility problems. If you're on a machine with a GPU and need reliable text extraction beyond what lightweight OCR provides, this is worth trying. Just note Surya itself requires a commercial license if your company does over $2M in revenue.
npx skills add https://github.com/aktsmm/agent-skills --skill ocr-super-surya