Document Processing Toolkit

What it solves

When feeding office documents into LLM workflows, the friction is rarely the model but the files: mixed formats, oversized context, and Chinese text encoding. document-skills turns PDF, DOCX, XLSX, PPTX, HTML, CSV, JSON, and TXT files into inspectable text, and keeps a repeatable path for search, summarization, and comparison.

How it works

• Text extraction: use scripts/extract_text.py; for large files, write output to disk with -o and read portions selectively.
• Format conversion: scripts/convert_format.py supports PDF → txt/md, DOCX → txt/md/html, XLSX → csv/json/txt, PPTX → txt/md, HTML → txt, CSV → json/xlsx, JSON → csv, and TXT → pdf/docx.
• Search and summarization: scripts/search_doc.py supports regex, case-insensitive matching with -i, and context lines with -C N; summarization typically extracts text first, then analyzes sections or page ranges.
• Large-document strategy: PDF can be chunked with --pages, XLSX can be processed sheet by sheet with --sheet, and other files can be exported before partial reads.

Boundaries

This skill is best for preprocessing structured or semi-structured office files for agent pipelines, not OCR, layout reconstruction, or cross-format editing. For Chinese documents, try --encoding gbk or --encoding gb18030; if dependencies are missing or files are corrupted, resolve the library or file issue before expecting stable output.