document-processing

SKILL.md

Document Processing Guide

Work with office documents: PDF, Excel, Word, and PowerPoint.

Format Overview

Format
Extension
Structure
Best For

PDF
.pdf
Binary/text
Reports, forms, archives

Excel
.xlsx
XML in ZIP
Data, calculations, models

Word
.docx
XML in ZIP
Text documents, contracts

PowerPoint
.pptx
XML in ZIP
Presentations, slides

Key concept: XLSX, DOCX, and PPTX are all ZIP archives containing XML files. You can unzip them to access raw content.

PDF Processing

PDF Tools

Task
Best Tool

Basic read/write
pypdf

Text extraction
pdfplumber

Table extraction
pdfplumber

Create PDFs
reportlab

OCR scanned PDFs
pytesseract + pdf2image

Command line
qpdf, pdftotext

Common Operations

Operation
Approach

Merge
Loop through files, add pages to writer

Split
Create new writer per page

Extract tables
Use pdfplumber, convert to DataFrame

Rotate
Call .rotate(degrees) on page

Encrypt
Use writer's .encrypt() method

OCR
Convert to images, run pytesseract

Excel Processing

Excel Tools

Task
Best Tool

Data analysis
pandas

Formulas & formatting
openpyxl

Simple CSV
pandas

Financial models
openpyxl

Critical Rule: Use Formulas

Approach
Result

Wrong: Calculate in Python, write value
Static number, breaks when data changes

Right: Write Excel formula
Dynamic, recalculates automatically

Financial Model Standards

Convention
Meaning

Blue text
Hardcoded inputs

Black text
Formulas

Green text
Links to other sheets

Yellow fill
Needs attention

Common Formula Errors

Error
Cause

#REF!
Invalid cell reference

#DIV/0!
Division by zero

#VALUE!
Wrong data type

#NAME?
Unknown function name

Word Processing

Word Tools

Task
Best Tool

Text extraction
pandoc

Create new
python-docx or docx-js

Simple edits
python-docx

Tracked changes
Direct XML editing

Document Structure

File
Contains

word/document.xml
Main content

word/comments.xml
Comments

word/media/
Images

Tracked Changes (Redlining)

Element
XML Tag

Deletion
<w:del><w:delText>...</w:delText></w:del>

Insertion
<w:ins><w:t>...</w:t></w:ins>

Key concept: For professional/legal documents, use tracked changes XML rather than replacing text directly.

PowerPoint Processing

PowerPoint Tools

Task
Best Tool

Text extraction
markitdown

Create new
pptxgenjs (JS) or python-pptx

Edit existing
Direct XML or python-pptx

Slide Structure

Path
Contains

ppt/slides/slide{N}.xml
Slide content

ppt/notesSlides/
Speaker notes

ppt/slideMasters/
Master templates

ppt/media/
Images

Design Principles

Principle
Guideline

Fonts
Use web-safe: Arial, Helvetica, Georgia

Layout
Two-column preferred, avoid vertical stacking

Hierarchy
Size, weight, color for emphasis

Consistency
Repeat patterns across slides

Converting Between Formats

Conversion
Tool

Any → PDF
LibreOffice headless

PDF → Images
pdftoppm

DOCX → Markdown
pandoc

Any → Text
Appropriate extractor

Best Practices

Practice
Why

Use formulas in Excel
Dynamic calculations

Preserve formatting on edit
Don't lose styles

Test output opens correctly
Catch corruption early

Use tracked changes for contracts
Audit trail

Extract to markdown for analysis
Easier to process

Common Packages

Language
Packages

Python
pypdf, pdfplumber, openpyxl, python-docx, python-pptx

JavaScript
docx, pptxgenjs

CLI
pandoc, qpdf, pdftotext, libreoffice