PPTX Skill

Source: https://github.com/aipoch/medical-research-skills

When to Use

• You need to generate a new .pptx deck from a short prompt or structured outline (e.g., "5 slides about machine learning").
• You want to update an existing presentation by adding slides or editing text without manually opening PowerPoint.
• You need to extract structured information from a deck (e.g., slide titles) for indexing, review, or QA.
• You want to export slides to images (thumbnails) or PDF for previews, sharing, or downstream processing.
• You need to insert images into slides (local files or downloaded assets) as part of automated reporting.

Key Features

• Presentation creation: Create new .pptx files and populate them with slides.
• Slide authoring: Add slides with titles, body text, and images.
• Text editing: Modify text content on existing slides.
• Image support: Insert and handle images (including basic manipulation via Pillow).
• Template support: Start from existing .pptx templates and extend them.
• Export options: Export slides as images (thumbnails) and optionally export to PDF (via external tooling).
• Information extraction: Read slide metadata such as slide titles.

Dependencies

• Python: >=3.7
• python-pptx: >=0.6.21
• Pillow: >=9.0.0 (image handling)
• requests: >=2.28.0 (downloading remote images)
• Optional (advanced export): LibreOffice >=7.0 (e.g., PPTX → PDF conversion)

Example Usage

# pip install python-pptx Pillow requests

from pptx import Presentation
from pptx.util import Inches
from PIL import Image
import requests
from io import BytesIO

def create_presentation(output_path: str) -> None:
    prs = Presentation()

    # Slide 1: Title slide
    slide = prs.slides.add_slide(prs.slide_layouts[0])
    slide.shapes.title.text = "Machine Learning"
    slide.placeholders[1].text = "A 5-slide overview generated programmatically"

    # Slide 2: Bullets
    slide = prs.slides.add_slide(prs.slide_layouts[1])
    slide.shapes.title.text = "What is Machine Learning?"
    tf = slide.shapes.placeholders[1].text_frame
    tf.clear()
    tf.text = "A field of AI focused on learning patterns from data"
    for bullet in [
        "Supervised learning",
        "Unsupervised learning",
        "Reinforcement learning",
    ]:
        p = tf.add_paragraph()
        p.text = bullet

    # Slide 3: Add an image (downloaded)
    img_url = "https://upload.wikimedia.org/wikipedia/commons/4/44/Neural_network.svg"
    resp = requests.get(img_url, timeout=30)
    resp.raise_for_status()

    # Ensure the image is in a format python-pptx can embed reliably
    img = Image.open(BytesIO(resp.content)).convert("RGBA")
    buf = BytesIO()
    img.save(buf, format="PNG")
    buf.seek(0)

    slide = prs.slides.add_slide(prs.slide_layouts[5])  # Title Only
    slide.shapes.title.text = "Neural Networks (Illustration)"
    slide.shapes.add_picture(buf, Inches(1), Inches(1.6), width=Inches(8))

    # Slide 4: Edit text on an existing slide (example: update slide 2 title)
    prs.slides[1].shapes.title.text = "Machine Learning: Definition & Types"

    # Slide 5: Summary
    slide = prs.slides.add_slide(prs.slide_layouts[1])
    slide.shapes.title.text = "Summary"
    tf = slide.shapes.placeholders[1].text_frame
    tf.clear()
    tf.text = "Key takeaways"
    for bullet in [
        "ML learns from data to make predictions or decisions",
        "Model choice depends on task and constraints",
        "Evaluation and iteration are essential",
    ]:
        p = tf.add_paragraph()
        p.text = bullet

    prs.save(output_path)

def list_slide_titles(pptx_path: str) -> list[str]:
    prs = Presentation(pptx_path)
    titles = []
    for slide in prs.slides:
        title_shape = slide.shapes.title if hasattr(slide.shapes, "title") else None
        if title_shape is not None and getattr(title_shape, "text", "").strip():
            titles.append(title_shape.text.strip())
        else:
            titles.append("(no title)")
    return titles

if __name__ == "__main__":
    out = "machine_learning.pptx"
    create_presentation(out)
    print("Created:", out)
    print("Slide titles:", list_slide_titles(out))

Implementation Details

• Core library: Uses python-pptx to read/write the Open XML .pptx format.
• Slide layouts: Slides are created from built-in layouts (e.g., prs.slide_layouts[0] for title slide, prs.slide_layouts[1] for title+content). Layout availability can vary by template.
• Text editing model: Text is edited via TextFrame and Paragraph objects. Clearing and rebuilding a text frame is a common approach to ensure consistent bullet structure.
• Image insertion:
  • Remote images can be downloaded with requests.
  • Images are normalized with Pillow (e.g., converting to PNG) before embedding to improve compatibility.
  • Placement uses absolute positioning (e.g., Inches(x)) and optional sizing parameters.
• Extraction: Slide titles are typically accessed via slide.shapes.title when present; some slides may not have a title placeholder.
• Export limitations:
  • python-pptx does not natively render slides to images or PDF. Thumbnail/PDF export generally requires external rendering (commonly LibreOffice in headless mode).
• Known constraints:
  • Complex animations and some advanced PowerPoint features may not be editable.
  • Large decks and high-resolution images increase processing time and memory usage.

When Not to Use

• Do not use this skill when the required source data, identifiers, files, or credentials are missing.
• Do not use this skill when the user asks for fabricated results, unsupported claims, or out-of-scope conclusions.
• Do not use this skill when a simpler direct answer is more appropriate than the documented workflow.

Required Inputs

• A clearly specified task goal aligned with the documented scope.
• All required files, identifiers, parameters, or environment variables before execution.
• Any domain constraints, formatting requirements, and expected output destination if applicable.

Recommended Workflow

1. Validate the request against the skill boundary and confirm all required inputs are present.
2. Select the documented execution path and prefer the simplest supported command or procedure.
3. Produce the expected output using the documented file format, schema, or narrative structure.
4. Run a final validation pass for completeness, consistency, and safety before returning the result.