docx

name	docx
description	Extract text and tables from Word (.docx) files. Use when a user uploads a Word document that needs to be processed.

Word Document Processing for Text Extraction

Purpose

Extract readable text and structured tables from DOCX files. This skill feeds extracted content into downstream processing (e.g., document summarization for PPT generation).

When to Use

• User uploads a .docx or .doc file
• Need to extract document content for summarization, analysis, or conversion
• Document contains structured tables that need to be captured
• Legacy .doc files must be converted to .docx first

Approach

Step 1: Try python-docx First

from docx import Document

doc = Document("document.docx")

# Extract paragraphs (skip empty)
paragraphs = [p.text.strip() for p in doc.paragraphs if p.text.strip()]
raw_text = "\n\n".join(paragraphs)

# Extract tables
tables = []
for table in doc.tables:
    rows_data = []
    for row in table.rows:
        cells = [cell.text.strip() for cell in row.cells]
        rows_data.append(cells)
    if rows_data:
        tables.append(rows_data)

Why python-docx:

• Native DOCX support, no external dependencies
• Preserves paragraph boundaries and table structure
• Handles both text and tables in one pass

Step 2: Fallback to pandoc (Command Line)

# Convert to plain text
pandoc document.docx -o output.txt --to plain

# Preserve more formatting
pandoc document.docx -o output.md --to markdown

# Track changes (if present)
pandoc document.docx -o output.txt --track-changes=all

import subprocess

result = subprocess.run(
    ["pandoc", "document.docx", "-o", "-", "--to", "plain"],
    capture_output=True, text=True, timeout=60
)
if result.returncode == 0:
    raw_text = result.stdout

Use pandoc when:

• python-docx is not installed
• Need markdown output for richer formatting info
• Document has complex formatting that python-docx misses

Step 3: Legacy .doc Files

Convert to .docx first using LibreOffice:

soffice --headless --convert-to docx document.doc

Then process as .docx (Step 1 or 2).

Handling Common Issues

Problem	Solution
.doc not .docx	Convert with LibreOffice first
Empty paragraphs	Filter if p.text.strip()
Table cells with nested content	Use .text on each cell (simplest)
Track changes in document	Use pandoc with --track-changes=all
python-docx not installed	Use pandoc fallback
pandoc not installed	Use LibreOffice to convert to text

Document Structure Preservation

• Headings: Check paragraph.style.name for "Heading 1", "Heading 2", etc.
• Lists: Detect bullet/number patterns in text
• Bold/Italic: Inspect run.bold and run.italic for emphasis
• Tables: Preserve as 2D arrays, include header row separately

# Preserve heading structure
for para in doc.paragraphs:
    style = para.style.name
    if "Heading" in style:
        level = style  # "Heading 1", "Heading 2", etc.
        text = para.text.strip()
        # ... heading with level

Integration Notes

For pptagent pipeline:

1. Extract text paragraphs → pass to document-summarizer skill
2. Extract tables → pass alongside text
3. For DOCX with clear headings: preserve heading hierarchy, summarizer will map to sections
4. Estimate page count: max(1, len(paragraphs) // 30) (rough estimate)

Output Format

Return (raw_text: str, tables: list[list[list[str]]], estimated_pages: int):

• raw_text: paragraphs joined with double newlines
• tables: list of tables, each table is list[list[str]] (rows → cells)
• estimated_pages: rough page estimate based on paragraph count