docx

name	docx
description	Corporate Word document generation and editing for .docx files. Use when Claude needs to create professional Word documents from structured input, preserve document templates, or produce styled report outputs.
license	Proprietary. LICENSE.txt has complete terms

SKILL.md — /docx Corporate Word Document Generator

What It Does

Generates professional-grade Word documents (.docx) from structured input. Output is generated from the reference template at: .claude/skills/docx/templates/reference.docx

Mode Selection Rules

Choose the workflow based on what the user actually provided:

1. Existing .docx provided and user asked to edit/revise/update it

   • Treat this as an edit-in-place task.
   • Build on the user's document.
   • Preserve their existing structure, wording, comments, tables, and styles unless they explicitly ask for a redesign.
   • Do not switch to .claude/skills/docx/templates/reference.docx just because the file is a Word document.

2. No existing .docx provided and user wants a new polished document

   • Treat this as a new document generation task.
   • Use .claude/skills/docx/templates/reference.docx and the generator script.

3. Existing .docx provided but user explicitly asks to restyle, rebuild, or convert it into the corporate format

   • You may use the reference template.
   • Make that transformation explicit in your reasoning and output.

If there is any ambiguity, ask a clarifying question before proceeding:

• do they want edits to the uploaded document
• or a fresh rebuilt version in the corporate template

How to Invoke

/docx [document title] — [brief description of what the document should contain]

The agent will generate a complete corporate document.

If the task is to modify an existing document, do not use the invocation above as a reason to regenerate from scratch. Edit the provided document instead.

---

Output

• File: output/[title]_YYYY-MM-DD.docx
• Format: A4, Times New Roman, professional corporate styling
• Template: Loads .claude/skills/docx/templates/reference.docx directly and preserves its styles, margins, headers, and footers

---

Review / Validation Utilities

The skill now also includes document review and validation helpers for existing Office files.

Accept tracked changes in a reviewed .docx

Use:

python3 .claude/skills/docx/scripts/accept_changes.py input_reviewed.docx output_clean.docx

What it does:

• uses LibreOffice (soffice) in headless mode
• accepts all tracked changes
• writes a clean output document

Validate unpacked or packed Office files

Run the validator from inside the office script directory so its relative imports resolve cleanly:

cd .claude/skills/docx/scripts/office
python3 validate.py /absolute/path/to/file.docx --original /absolute/path/to/original.docx --author "Claude"

You can also validate an unpacked Office directory:

cd .claude/skills/docx/scripts/office
python3 validate.py /absolute/path/to/unpacked_docx_dir --original /absolute/path/to/original.docx

Useful flags:

• --auto-repair to fix common XML issues automatically
• --verbose for detailed output
• --author to control tracked-change validation assumptions

What these utilities add

• tracked-change cleanup
• XML/schema validation for Office documents
• redlining validation for .docx review flows
• LibreOffice-aware execution helpers for sandboxed environments

---

Document Structure

Every generated document follows this structure:

1. COVER PAGE
   ├── Document Title (centered, large bold)
   ├── Subtitle (if applicable)
   ├── Date
   ├── Classification (if applicable: CONFIDENTIAL / RESTRICTED / INTERNAL)
   └── Prepared By

2. TABLE OF CONTENTS
   └── Generated from the actual heading structure in the document

3. BODY
   ├── Executive Summary
   ├── Section 1 — [Heading 1: 14pt bold]
   │   ├── [Heading 2: 13pt bold]
   │   │   ├── [Heading 3: 12pt bold]
   │   │   └── Body paragraphs (12pt, justified, 1.15 line spacing)
   │   └── Bullet lists (12pt)
   │   └── Tables (Light Shading Accent 1, bold header row, gray header background)
   ├── Section 2...
   └── Conclusion / Recommendations

4. FOOTER
   └── Page numbers: "Page X of Y"

---

Formatting Standards (MUST follow exactly)

Element	Font	Size	Style	Spacing
Cover Title	Times New Roman	18pt	Bold	180pt before
Heading 1	Times New Roman	14pt	Bold, Black	18pt before, 6pt after
Heading 2	Times New Roman	13pt	Bold	12pt before, 6pt after
Heading 3	Times New Roman	12pt	Bold	10pt before, 6pt after
Body	Times New Roman	12pt	Regular	0pt before, 6pt after, justified
Tables	Times New Roman	11pt	Regular	Header: bold, gray (#D9D9D9) fill
Footnote/Caption	Times New Roman	10pt	Italic	—
Header	Times New Roman	9pt	Regular, gray	—
Footer	Times New Roman	9pt	Regular, gray	—

Page: A4 (595pt × 842pt), 1-inch margins all sides

Table Style: Light Shading Accent 1 — alternating row shading (white / #F2F2F2)

Header: Document title (left) | Chapter name (right), gray, 9pt, thin bottom border

Footer: "Page X of Y", centered, gray, 9pt

---

Generation Script

Use the provided generator: .claude/skills/docx/scripts/generate_corporate_docx.py

python3 .claude/skills/docx/scripts/generate_corporate_docx.py \
  --title "Fuel Oil Market Analysis" \
  --subtitle "Q1 2026 Benchmark Assessment" \
  --date "April 2026" \
  --classification "CONFIDENTIAL" \
  --prepared-by "BedRock AI Research" \
  --chapter "Market Analysis" \
  --output output/fuel_oil_report.docx \
  --sections-json '[
    {"type": "heading1", "text": "Executive Summary"},
    {"type": "body", "text": "This report examines..."},
    {"type": "heading2", "text": "Market Overview"},
    {"type": "body", "text": "Global fuel oil demand..."},
    {"type": "bullet", "items": ["Point one", "Point two", "Point three"]},
    {"type": "heading2", "text": "Key Data"},
    {"type": "table", "headers": ["Month", "Volume", "Price", "YoY Change"], "rows": [["Jan", "42M bbl", "$89", "+3.2%"], ["Feb", "38M bbl", "$91", "+1.8%"]], "col_widths": [1.0, 1.5, 1.5, 1.5]},
    {"type": "heading1", "text": "Regulatory Considerations"},
    {"type": "body", "text": "CFTC guidelines require..."}
  ]'

---

What the Agent Must Do

1. Determine document structure from the request:

   • What type of document? (report, memo, analysis, briefing, legal document)
   • Who is the audience? (executive, legal, technical, client)
   • What are the 3-5 key sections?

2. Gather / write content for each section:

   • Use available research files (BedRock project, memory, web search)
   • Write clear, professional prose (not bullet-point lists in body text)
   • Ensure tables contain actual data (do not fabricate numbers)

3. Call the generator script with structured section JSON

4. Verify output:

   • Document opens without errors
   • Cover page has all required fields
   • Table of contents reflects actual sections
   • Tables are properly formatted
   • Page numbers appear in footer

---

Example Prompt

Generate a professional report on the Platts fuel oil benchmark manipulation as a Word document.
Title: "Platts Fuel Oil Benchmark Manipulation: Regulatory Analysis"
Classification: CONFIDENTIAL
Audience: Legal and compliance team
Sections: Executive Summary, Background on Platts MOC Mechanism, Glencore CFTC Case, 
          Regulatory Implications, Detection Methods, Recommendations

---

Notes

• Always use the generator script — do not manually construct docx XML
• If the user supplied an existing .docx for revision, modify that file in place unless they explicitly asked for a template-based rebuild
• Do not fabricate data in tables — use real data or mark cells as "TBD"
• Classification levels: INTERNAL / CONFIDENTIAL / RESTRICTED / PUBLIC
• The reference template (reference.docx) is the style authority — always match it
• Prerequisite: python-docx must be available in the Python environment used to run the script