Catalog›Agent skills›Files & documents›pdf-word-reader-zhPWpdf-word-reader-zh20041002liu-cloud/pdf-word-reader-zh · skills.sh Open site Project Open in the app JSON README (API)Skill publicada por 20041002liu-cloud/pdf-word-reader-zh no skills.sh. Instale com: npx skills add 20041002liu-cloud/pdf-word-reader-zh@pdf-word-reader-zhREADMELoading documentation…# pdf-word-reader-zh ```bash $ npx skills add https://github.com/20041002liu-cloud/pdf-word-reader-zh ``` SKILL.md This skill provides end-to-end document understanding for PDF, DOCX, and PPTX files. It turns a file path into structured extraction, chunked evidence, and an analysis-ready report. The user provides a document path and asks to read, summarize, extract, analyze, or fully understand the content. ## What This Skill Does - Read `.pdf` files with text extraction and OCR fallback - Read `.docx` files with paragraph/table extraction - Convert `.pptx` to PDF automatically, then process as PDF - Split long text into chunked evidence (`C001`, `C002`, ...) - Generate a report scaffold for evidence-based conclusions ## Workflow 1. Detect input type: PDF, DOCX, or PPTX. 2. If PPTX, convert to PDF (LibreOffice or PowerPoint COM on Windows). 3. Extract structured content (text, pages, tables when available). 4. Apply OCR fallback on low-text PDF pages. 5. Build evidence chunks with stable chunk IDs. 6. Output analysis scaffold with chunk evidence index. ## Command ```bash python scripts/prepare_document_context.py "<input-file>" --output-dir "output/document-understanding" ``` ## Output - `01_extracted.json`: structured extraction result - `02_chunks.json`: chunked evidence units - `03_understanding_report.md`: analysis scaffold and evidence index ## Common Options - `--disable-ocr`: disable OCR fallback - `--max-pages N`: quick run on first N pages - `--fail-on-empty`: fail if no text is extracted ## Dependencies ```bash python -m pip install -r requirements.txt ``` Recommended system tools: - `tesseract` with Chinese language pack (`chi_sim`) - `pdftoppm` (Poppler) - For PPTX conversion: `soffice` (LibreOffice) or Microsoft PowerPoint (Windows) ## Quality Rule For final analysis output: - Read all chunks before concluding. - Cite chunk IDs for key claims, e.g. `[C003][C011]`. - Separate facts from assumptions. - List missing evidence explicitly.DetailsTypeAgent skillsTopicFiles & documentsPublisher20041002liu-cloudSourceskillsshCategoryToolsStars5Last update2026-03-14T06:04:38ZProject statusActiveLanguagePythonAdded2026-10-07 05:24:39Updated2026-10-07 05:24:39Source id20041002liu-cloud/pdf-word-reader-zh/pdf-word-reader-zhMoreMore agent skills on Files & documentsAll from 20041002liu-cloudA–Z index
