Efficient Document Processing Multi-Agent Team

Problem to solve

When you have multiple bilingual PDF, Word, Excel, or image documents and need to read, translate, classify, summarize, and report them in one pass, manual extraction is brittle and hard to reproduce. doc-processing-team targets batch document processing, not quick single-document edits or original content generation.

How it works

The skill runs a five-agent pipeline: Reader extracts text, tables, and metadata; Translator converts English content into Chinese; Classifier groups documents by themes such as technical specs, commercial quotations, or solutions; Summarizer produces key points and metrics; ReportWriter outputs a Markdown report.

A typical run starts with confirming the file list and target language, choosing --demo or --files, installing optional dependencies, then executing the pipeline. Results are written to scripts/document_report.md. The engine includes blackboard communication, checkpoint resume, circuit breakers, and watchdog timeouts; it can also learn terminology or category rules via --teach and persist local state under state/.

Boundaries and caveats

This skill runs offline and does not perform live web retrieval. Real file parsing requires pymupdf, openpyxl, and python-docx. --demo uses mock data; translation relies on terminology and phrase tables rather than a large model; classification uses keyword rules. Keep --files batches under about 10 documents. For important contracts, niche domains, or very long files, review output manually or split the input.