Об этом Skill
A specialized workflow for scans without a usable text layer: split the book, run MinerU OCR, map chapter titles, build DOCX, then check suspicious text against page images. The included converter contains Yijing-specific correction rules and a Windows OCR path; adapt these for another book or platform instead of treating it as a universal one-click converter.
Почему мы его выбрали
Useful for scanned Chinese and classical texts: it treats OCR correction and chapter reconstruction as reviewable steps, with explicit adaptation requirements.
Для каких задач подходит
- Recover editable Word text from scanned Chinese or classical books.
- Rebuild missing chapter headings and preserve a reviewed correction list.
- Reuse existing MinerU Markdown while refining the DOCX output.
Workflow
Split and OCR
Work on manageable chapter-sized files and retain page-index metadata from MinerU.
Map titles and build Word
Create the title map and document-specific corrections, then generate DOCX from the OCR Markdown.
Proofread against scans
Find suspicious markers, inspect the corresponding original image, regenerate fixes and deliver unresolved questions with the document.
Installation and runtime
Install the Skill instructions, then provision the documented runtime dependencies before running a task.
| Команда | Для чего полезна |
|---|---|
python scripts/convert.py part.pdf -o output.docx --md ocr.md --toc toc.py | Reuse reviewed OCR Markdown and a title map. |
python scripts/check.py output.docx | Inspect the generated document for known error markers. |
Scope and practical limits
The workflow targets scanned Chinese books; PDFs with an existing clean text layer are better extracted directly. Its sample-specific corrections require adaptation.
License and source
The project is distributed under MIT. Agent subscriptions, external services and optional integrations have their own requirements.
Перед использованием
- Python 3.10–3.12 with MinerU, python-docx and PyMuPDF; OCR model downloads and runtime can be substantial.
- The documented OCR environment uses Windows and PowerShell. Other platforms need path and font adjustments; --md can reuse existing OCR output.
- Adapt the example corrections, title map and domain-specific symbols to the actual book, and review uncertain passages manually.
Поддерживаемые агенты
Источники и материалы
Ссылки на оригинальные материалы приведены ниже. Практические идеи предложены редакцией SkillHey.
