DOCX

Document Processing Workflow

Process, create, convert, and analyze documents by combining PDF tools, Word document generation, spreadsheet processing, and web research.

Sub-Skills

• anthropics/pdf — Full PDF toolkit: text extraction, table extraction, creation, merging, splitting, OCR for scanned docs, watermarking, page rotation, form filling, password protection. Uses pypdf, pdfplumber, reportlab.
• anthropics/docx — Word document creation and editing: formatting, tracked changes, comments, two-column layouts, text extraction, conversion to images. Uses docx-js and OOXML.
• anthropics/xlsx — Excel spreadsheet creation and analysis: formulas, formatting, data visualization, financial models, pivot-style analysis. Uses openpyxl and pandas.
• rebyteai/internet-search — Research for content generation, template references, or data to populate documents.

Workflow

Step 1: Identify the Operation

Parse what the user wants. Document tasks fall into these categories:

Identify:

• Input — Is there an uploaded file? What format? (PDF, DOCX, XLSX, CSV, scanned image)
• Operation — Extract, create, transform, analyze, or edit?
• Output format — PDF, Word doc, Excel, online spreadsheet, JSON, plain text?

Step 2: Process Input (if file uploaded)

When the user uploads a file:

PDF files:

• Use pdf for text extraction, table extraction, form data, or image extraction
• For scanned PDFs, use OCR capabilities (pdftotext, Tesseract)
• For tables, pdfplumber gives structured row/column data

Spreadsheet files (XLSX, CSV):

• Use xlsx (openpyxl + pandas) for analysis, formula work, and formatting
• pandas for statistical analysis (describe(), groupby(), pivot tables)
• Preserve existing formulas when editing

Word files (DOCX):

• Use docx for reading content, extracting text, modifying tracked changes
• Pandoc for text extraction when formatting doesn't matter

If no file is uploaded and the task is "create from scratch," skip to Step 3.

Step 3: Execute the Operation

Creating documents:

Word documents: Use docx with html2pptx-style workflow:

1. Build content structure (headings, paragraphs, tables, lists)
2. Apply professional formatting (styles, fonts, colors)
3. Add structural elements (TOC, headers/footers, page numbers)

PDF documents: Use pdf with reportlab for creation:

1. Define page layout and styles
2. Add content with proper typography
3. Include images, tables, charts as needed

Spreadsheets:

• Use xlsx for Excel files with formulas, formatting, and data visualization

Templates: For reusable templates (mail merge, invoices, forms):

1. Create the document structure
2. Add placeholder fields or form fields
3. Include instructions for use

Transforming documents:

Merge PDFs: Use pdf to combine multiple files in order, add page numbers and bookmarks Split PDFs: Detect chapter/section boundaries, create separate files Convert formats: PDF → text (pdftotext), DOCX → text (pandoc), CSV → XLSX (pandas + openpyxl) Restructure: Split sheets, combine data, reorganize sections

Analyzing documents:

1. Extract the data (text, tables, numbers)
2. Process with appropriate tools (pandas for stats, formulas for calculations)
3. Generate insights (summaries, outliers, trends)
4. Present results (new document, spreadsheet, or inline summary)

Step 4: Research (if needed)

Use internet-search when creating documents that need external content:

• Industry benchmarks for comparison documents
• Template best practices (e.g., invoice standards, contract clauses)
• Data to populate templates (current rates, regulations, standards)

Skip this step for pure file processing (merge, split, extract, OCR).

Step 5: Deliver

Based on what was created:

Downloadable files (PDF, DOCX, XLSX):

• Upload to Artifact Store
• Note the format and any special instructions (e.g., "enable macros for formulas")

Extracted data:

• For structured data: output as spreadsheet or JSON
• For text: output as formatted markdown or plain text
• For tables: output as spreadsheet with proper column headers

Reports/analysis:

• Combine findings into a clear summary
• Include both the raw data and the analysis

Decision Points

• "Which PDF tool?" — anthropics/pdf covers all PDF operations. It's the only PDF skill needed.

• "Scanned PDF?" — If the PDF has no extractable text (scanned/photographed), use OCR. The pdf skill includes pdftotext and Tesseract for this.

• "Should I use internet-search?" — Only when creating documents that need external content. Not needed for processing uploaded files.

• "Multiple output formats?" — If the user doesn't specify, choose the most natural format: tables → spreadsheet, reports → Word/PDF, data → CSV/JSON. Ask if ambiguous.