pptx

name	pptx
description	Extract text and tables from PowerPoint (.pptx) files. Use when a user uploads a PPTX presentation that needs to be processed.

PowerPoint Processing for Text Extraction

Purpose

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

When to Use

• User uploads a .pptx file
• Need to extract presentation content for summarization, analysis, or conversion
• Presentation contains text in shapes, tables, or speaker notes
• Extract content to generate a new presentation based on existing one

Approach

Step 1: Try python-pptx First

python-pptx is the standard library for reading PPTX content.

from pptx import Presentation

prs = Presentation("presentation.pptx")
text_parts = []
tables = []
slide_count = len(prs.slides)

for slide in prs.slides:
    for shape in slide.shapes:
        # Extract text from shapes
        if shape.has_text_frame:
            for paragraph in shape.text_frame.paragraphs:
                line = paragraph.text.strip()
                if line:
                    text_parts.append(line)

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

Why python-pptx first:

• Native PPTX support, no external dependencies beyond the library
• Iterates all shapes (text boxes, placeholders, tables, etc.)
• Can access slide metadata (slide count, layout info)

Step 2: Fallback to markitdown (CLI Tool)

markitdown converts PPTX to plain text/markdown for easy extraction.

python -m markitdown presentation.pptx

import subprocess

result = subprocess.run(
    ["python", "-m", "markitdown", "presentation.pptx"],
    capture_output=True, text=True, timeout=60
)
if result.returncode == 0:
    raw_text = result.stdout.strip()

Use markitdown when:

• python-pptx is not installed
• Need richer text extraction (includes speaker notes, metadata)
• Output in markdown format is preferred

Step 3: Alternative — markitdown as Python Library

# pip install "markitdown[pptx]"
import markitdown

md_converter = markitdown.MarkItDown()
result = md_converter.convert("presentation.pptx")
raw_text = result.text_content

Step 4: Raw XML Access (Advanced)

PPTX files are ZIP archives containing XML. Extract raw text if needed:

import zipfile

with zipfile.ZipFile("presentation.pptx", "r") as z:
    slide_files = sorted([f for f in z.namelist() if f.startswith("ppt/slides/slide") and f.endswith(".xml")])
    for slide_file in slide_files:
        content = z.read(slide_file).decode("utf-8")
        # Extract text between <a:t> tags
        import re
        texts = re.findall(r"<a:t>([^<]+)</a:t>", content)

Use XML extraction when:

• Both python-pptx and markitdown are unavailable
• Need raw slide XML for custom processing

Handling Common Issues

Problem	Solution
Empty text (encrypted PPTX)	File may be password-protected — prompt user to unlock first
Text in grouped shapes	python-pptx automatically flattens groups; XML fallback may be needed
Slide master text bleeding through	Filter out text from slide masters/layouts (use shape.is_placeholder check)
Tables not extracting	Verify shape.has_table is True; some tables are images, not shapes
Very large presentations	Process slide-by-slide, limit total text (e.g., 100,000 characters)
.ppt (legacy format, not .pptx)	Convert with LibreOffice: soffice --headless --convert-to pptx file.ppt
python-pptx not installed	Use markitdown fallback

Content Structure Preservation

PPTX content follows a slide hierarchy:

# Preserve slide-by-slide structure
for slide_idx, slide in enumerate(prs.slides):
    slide_title = None
    slide_content = []

    for shape in slide.shapes:
        if shape.has_text_frame:
            text = shape.text_frame.text.strip()
            if not text:
                continue
            # First text shape with large font is likely the title
            if shape.shape_type == 1:  # TITLE placeholder
                slide_title = text
            else:
                slide_content.append(text)

    slide_text = f"Slide {slide_idx + 1}"
    if slide_title:
        slide_text += f": {slide_title}"
    if slide_content:
        slide_text += "\n" + "\n".join(slide_content)
    all_slides.append(slide_text)

Speaker Notes Extraction

Speaker notes can contain additional context:

if slide.has_notes_slide:
    notes_frame = slide.notes_slide.notes_text_frame
    notes_text = notes_frame.text.strip()

Integration Notes

For pptagent pipeline:

1. Extract text from each slide → join with double newlines: "\n\n".join(text_parts)
2. Extract tables → pass alongside text to summarizer
3. Slide count serves as page count estimate: max(1, slide_count)
4. Consider speaker notes as supplementary content if available

Output Format

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

• raw_text: all text parts joined with double newlines
• tables: list of tables, each table is list[list[str]] (rows → cells)
• slide_count: total number of slides (used as page count estimate)