DOCX Creation, Editing, and Analysis

Work with Word documents: create from scratch, edit existing content, add tracked changes for professional review, extract text, and convert to PDF or images.

Overview

A .docx file is a ZIP archive of XML files and resources. Different tasks require different tools:

When to Use

• •User asks to create a new Word document from content or data
• •User asks to edit, modify, or update an existing .docx file
• •User asks to review a document with tracked changes (redlining)
• •User asks to add comments to a Word document
• •User asks to extract or read text from a .docx file
• •User asks to convert a .docx to PDF or images for visual inspection

Workflow Decision Tree

User request
 |- "Create a new document"
 |   -> Creating workflow (docx-js)
 |- "Edit my own document" + simple changes
 |   -> Basic OOXML editing workflow
 |- "Review someone else's document"
 |   -> Redlining workflow (recommended default)
 |- "Legal, academic, business, or government docs"
 |   -> Redlining workflow (required)
 |- "Read/analyze document content"
 |   -> Text extraction with pandoc
 |- "Show me what the document looks like"
     -> Convert to images workflow

Creating a New Document (docx-js)

Use the docx-js library to create Word documents in JavaScript/TypeScript.

Steps

1. •Read the docx-js reference: Load references/docx-js-patterns.md for syntax, formatting rules, and common pitfalls
2. •Write a JS/TS file using Document, Paragraph, TextRun, Table, and other components
3. •Export with Packer.toBuffer() and write to disk

Key Rules

• •Never use \n for line breaks; use separate Paragraph elements
• •Always use LevelFormat.BULLET constant for bullet lists (not unicode symbols or string "bullet")
• •Always specify type parameter for ImageRun (png, jpg, gif, bmp, svg)
• •Set columnWidths array AND individual cell widths on tables
• •Use ShadingType.CLEAR for table cell shading (never SOLID)
• •PageBreak must always be inside a Paragraph; standalone creates invalid XML
• •Override built-in heading styles using exact IDs: "Heading1", "Heading2", "Heading3"
• •Include outlineLevel on heading styles for Table of Contents compatibility
• •Set a default font via styles.default.document.run.font (Arial recommended)

const { Document, Packer, Paragraph, TextRun } = require('docx');
const fs = require('fs');

const doc = new Document({
  styles: {
    default: { document: { run: { font: "Arial", size: 24 } } }
  },
  sections: [{
    properties: {
      page: { margin: { top: 1440, right: 1440, bottom: 1440, left: 1440 } }
    },
    children: [
      new Paragraph({
        heading: HeadingLevel.HEADING_1,
        children: [new TextRun("Section Title")]
      }),
      new Paragraph({
        children: [new TextRun("Body text in Arial 12pt.")]
      })
    ]
  }]
});

Packer.toBuffer(doc).then(buf => fs.writeFileSync("output.docx", buf));

Editing an Existing Document (OOXML)

Use the Document library (Python) for editing existing Word documents via XML manipulation.

Steps

1. •Read the OOXML reference: Load references/ooxml-patterns.md for the Document library API and XML patterns
2. •Unpack: python ooxml/scripts/unpack.py <file.docx> <output_dir>
3. •Write a Python script using the Document class for manipulation
4. •Repack: python ooxml/scripts/pack.py <dir> <output.docx>

Document Library Basics

from scripts.document import Document

doc = Document('unpacked')
# Find nodes by text, line number, or attributes
node = doc["word/document.xml"].get_node(tag="w:r", contains="target text")
# Replace content
doc["word/document.xml"].replace_node(node, "<w:r><w:t>new text</w:t></w:r>")
doc.save()

Set PYTHONPATH to the skill root directory before running:

PYTHONPATH=/path/to/docx-skill python your_script.py

Redlining Workflow (Tracked Changes)

For professional document review with tracked changes. This is the recommended default for editing another person's document and required for legal, academic, business, or government documents.

Principle: Minimal, Precise Edits

Only mark text that actually changes. Keep unchanged text outside <w:del>/<w:ins> tags. Preserve the original run's RSID for unchanged text.

# BAD - replaces entire sentence
'<w:del>..entire sentence..</w:del><w:ins>..entire sentence..</w:ins>'

# GOOD - only marks what changed: "30 days" -> "60 days"
'<w:r w:rsidR="00AB12CD"><w:t>within </w:t></w:r>'
'<w:del><w:r><w:delText>30</w:delText></w:r></w:del>'
'<w:ins><w:r><w:t>60</w:t></w:r></w:ins>'
'<w:r w:rsidR="00AB12CD"><w:t> days</w:t></w:r>'

Steps

1. •Convert to markdown: pandoc --track-changes=all file.docx -o current.md
2. •Identify and group changes into batches of 3-10 related changes (by section, type, or proximity)
3. •Read OOXML reference and unpack: python ooxml/scripts/unpack.py file.docx unpacked
4. •Implement each batch: Use get_node() to find nodes, apply tracked changes with replace_node(), suggest_deletion(), or insert_after()
5. •Repack: python ooxml/scripts/pack.py unpacked reviewed.docx
6. •Verify: Convert back to markdown and grep for expected changes

Method Selection

Reading and Analyzing Content

Text Extraction

# Convert to markdown preserving tracked changes
pandoc --track-changes=all document.docx -o output.md
# Options: --track-changes=accept/reject/all

Raw XML Access

For comments, complex formatting, metadata, or embedded media:

python ooxml/scripts/unpack.py document.docx unpacked/

Key files inside the unpacked archive:

Converting to Images

Two-step process for visual inspection:

# Step 1: DOCX to PDF
soffice --headless --convert-to pdf document.docx

# Step 2: PDF to JPEG images
pdftoppm -jpeg -r 150 document.pdf page
# Creates page-1.jpg, page-2.jpg, etc.

# Specific page range:
pdftoppm -jpeg -r 150 -f 2 -l 5 document.pdf page

Best Practices

Do

• •Read the appropriate reference file before starting any document task
• •Use separate Paragraph elements for each line (never \n)
• •Set a default font and establish visual hierarchy with styles
• •Group tracked changes into batches of 3-10 for manageable debugging
• •Grep word/document.xml before each script to get current line numbers
• •Preserve original run formatting (<w:rPr>) when making tracked changes
• •Use defusedxml for secure XML parsing

Do Not

• •Use unicode bullets ("bullet" string or SymbolRun for lists); use LevelFormat.BULLET
• •Mix up <w:ins>/<w:del> closing tags
• •Use markdown line numbers to locate content in XML (they do not map)
• •Modify text inside another author's <w:ins> or <w:del> directly; use nested deletions
• •Skip validation after repacking (doc.save() validates by default)

Dependencies

Examples

Example 1: Create a Simple Report

User: Create a Word document with a title "Q4 Report", two sections with
      headings, and a bullet list of key metrics.

Action: Read references/docx-js-patterns.md, then write a JS file using
        Document with Heading1 style override, numbered sections, and a
        bullet list via numbering config with LevelFormat.BULLET. Export
        with Packer.toBuffer().

Example 2: Redline a Contract

User: Review this NDA and change the confidentiality period from 2 years
      to 3 years, and update the governing law from California to Delaware.

Action: Use the redlining workflow. Convert to markdown first to understand
        the document, then unpack, read references/ooxml-patterns.md, find
        the target text with get_node(), apply minimal tracked changes
        (only mark "2 years" -> "3 years" and "California" -> "Delaware"),
        repack, and verify with pandoc.

Example 3: Extract Text from a Document

User: What does this contract say about termination clauses?

Action: Run pandoc --track-changes=all contract.docx -o contract.md,
        then read the markdown file and search for termination-related
        sections.

Output Checklist

• • Correct workflow selected based on decision tree
• • Reference file read before starting implementation
• • Document opens without errors in Word or LibreOffice
• • Tracked changes display correctly (for redlining tasks)
• • No unintended formatting changes introduced
• • All changes verified via markdown conversion