Document Creation

Core Principle

Documents are deliverables. Match the format to the audience: Word for collaborative editing, PDF for final distribution, PowerPoint for presentations, Excel for data and models. Every document should look like it was crafted by a professional, not generated by a script.

Format Selection

Word Documents (.docx)

Reading Content

# Text extraction with pandoc
pandoc document.docx -o output.md

# With tracked changes preserved
pandoc --track-changes=all document.docx -o output.md

Creating New Documents

Use the docx npm package (JavaScript) for creating Word documents programmatically.

npm install -g docx

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

const doc = new Document({
  sections: [{
    properties: {
      page: {
        size: { width: 12240, height: 15840 },  // US Letter in DXA
        margin: { top: 1440, right: 1440, bottom: 1440, left: 1440 }
      }
    },
    children: [
      new Paragraph({
        heading: HeadingLevel.HEADING_1,
        children: [new TextRun("Document Title")]
      }),
      new Paragraph({
        children: [new TextRun("Body text goes here.")]
      }),
    ]
  }]
});

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

Critical Rules for Word Documents

- Set page size explicitly (defaults to A4, not US Letter)
- Never use \n for line breaks — use separate Paragraph elements
- Never use unicode bullets — use LevelFormat.BULLET with numbering config
- Tables need dual widths: columnWidths on table AND width on each cell
- Always use WidthType.DXA, never PERCENTAGE (breaks in Google Docs)
- ImageRun requires the 'type' parameter (png, jpg, etc.)
- Override built-in heading styles with exact IDs: "Heading1", "Heading2"

PDF Documents

Reading PDFs

# Text extraction
import pdfplumber

with pdfplumber.open("document.pdf") as pdf:
    for page in pdf.pages:
        print(page.extract_text())

# Table extraction
with pdfplumber.open("document.pdf") as pdf:
    for page in pdf.pages:
        tables = page.extract_tables()
        for table in tables:
            for row in table:
                print(row)

Creating PDFs

from reportlab.lib.pagesizes import letter
from reportlab.platypus import SimpleDocTemplate, Paragraph, Spacer
from reportlab.lib.styles import getSampleStyleSheet

doc = SimpleDocTemplate("report.pdf", pagesize=letter)
styles = getSampleStyleSheet()
story = []

story.append(Paragraph("Report Title", styles['Title']))
story.append(Spacer(1, 12))
story.append(Paragraph("Body text content.", styles['Normal']))

doc.build(story)

PDF Operations

from pypdf import PdfReader, PdfWriter

# Merge PDFs
writer = PdfWriter()
for pdf_file in ["doc1.pdf", "doc2.pdf"]:
    reader = PdfReader(pdf_file)
    for page in reader.pages:
        writer.add_page(page)

with open("merged.pdf", "wb") as output:
    writer.write(output)

Critical Rules for PDFs

- Never use Unicode subscript/superscript characters in ReportLab
  (renders as black boxes) — use <sub> and <super> tags instead
- Use pdfplumber for reading, reportlab for creating, pypdf for manipulation
- For scanned PDFs, use pytesseract OCR (convert to images first)

PowerPoint Presentations (.pptx)

Reading Content

# Text extraction
python -m markitdown presentation.pptx

Design Principles

BEFORE STARTING:
  → Pick a bold, content-informed color palette
  → One color dominates (60-70%), 1-2 supporting, one accent
  → Dark backgrounds for title/conclusion, light for content
  → Commit to ONE visual motif and repeat it across slides

EVERY SLIDE NEEDS:
  → A visual element — image, chart, icon, or shape
  → Text-only slides are forgettable
  → Vary layouts: two-column, icon rows, grids, half-bleed images

TYPOGRAPHY:
  → Choose a header font with personality + clean body font
  → Titles: 36-44pt bold | Body: 14-16pt | Captions: 10-12pt
  → Never default to Arial — use Georgia, Cambria, Trebuchet MS

AVOID:
  → Same layout on every slide
  → Centered body text (left-align paragraphs and lists)
  → Accent lines under titles (hallmark of AI-generated slides)
  → Text-only slides with plain bullets

Creating from Scratch

Use pptxgenjs for programmatic creation:

npm install -g pptxgenjs

Excel Spreadsheets (.xlsx)

Reading Data

import pandas as pd

df = pd.read_excel('file.xlsx')            # First sheet
all_sheets = pd.read_excel('file.xlsx', sheet_name=None)  # All sheets

Creating Spreadsheets

from openpyxl import Workbook
from openpyxl.styles import Font, PatternFill, Alignment

wb = Workbook()
sheet = wb.active

# Add data
sheet['A1'] = 'Revenue'
sheet['B1'] = 'Q1'
sheet.append(['Product A', 50000])

# ALWAYS use formulas, not hardcoded calculations
sheet['B10'] = '=SUM(B2:B9)'

# Formatting
sheet['A1'].font = Font(bold=True)
sheet.column_dimensions['A'].width = 20

wb.save('output.xlsx')

Critical Rules for Excel

- ALWAYS use Excel formulas, never hardcode calculated values
  BAD:  sheet['B10'] = 5000  (calculated in Python)
  GOOD: sheet['B10'] = '=SUM(B2:B9)'

- Financial models: blue text for inputs, black for formulas,
  green for cross-sheet links, red for external links

- Number formatting:
  → Currency: $#,##0 with units in headers
  → Percentages: 0.0% (one decimal)
  → Negatives: parentheses (123) not -123
  → Years: text format ("2024" not "2,024")

- Place all assumptions in separate cells, never inline in formulas
- Document data sources for hardcoded values
- Recalculate after creating (formulas stored as strings, not values)

Anti-Patterns

Power Move

"Create a [document type] for [purpose]. Use professional formatting, proper typography, and make it look like it was designed by someone who cares about the details. Include [specific content requirements]. Output as [format]."

The agent becomes your document production team — creating polished, professional deliverables in any format.