pptx-render

Announce: "I'm using pptx-render to extract PPTX slide content."

PPTX Slide Inspector

What this skill carries — grep references/ for any subject the names below miss: !d=${CLAUDE_SKILL_DIR}; command -v skill-toc >/dev/null 2>&1 && exec skill-toc "$d"; s=$HOME/.claude/skills/plugin-utils/bin/skill-toc; [ -x "$s" ] && exec "$s" "$d"; echo "(skill-toc unavailable: references and scripts are NOT listed here — install the plugin-utils plugin, or start a new session so its bin/ reaches PATH)"

Extracts content from PPTX slides using python-pptx. Primary use case: understanding what a PPTX slide contains (shapes, text, positions, images) for comparison against Typst slides, especially diagrams and visual items (VIS-* in content inventories).

Prerequisites

Tool	Source
python-pptx	pixi project dependency

Step 1: Identify the PPTX File and Slide Number

If the user references a content inventory item (e.g., VIS-3, DQ-7), look up its PPTX slide number:

grep "VIS-3\|the-item-id" inventory/content-inventory-XX.md

Step 2: Extract Slide Shapes

from pptx import Presentation
import json

prs = Presentation('path/to/slides.pptx')
slide = prs.slides[SLIDE_NUM - 1]  # 0-indexed

for shape in slide.shapes:
    info = {
        'name': shape.name,
        'left_in': round(shape.left / 914400, 2),
        'top_in': round(shape.top / 914400, 2),
        'width_in': round(shape.width / 914400, 2),
        'height_in': round(shape.height / 914400, 2),
    }
    if shape.has_text_frame:
        info['text'] = shape.text_frame.text
    if shape.shape_type == 13:  # MSO_SHAPE_TYPE.PICTURE
        info['is_image'] = True
    if shape.has_table:
        info['is_table'] = True
        info['rows'] = len(shape.table.rows)
        info['cols'] = len(shape.table.columns)
    print(json.dumps(info))

Step 3: Extract Images (if needed)

To save embedded images from a slide:

from pptx import Presentation
from pptx.enum.shapes import MSO_SHAPE_TYPE

prs = Presentation('path/to/slides.pptx')
slide = prs.slides[SLIDE_NUM - 1]

for i, shape in enumerate(slide.shapes):
    if shape.shape_type == MSO_SHAPE_TYPE.PICTURE:
        image = shape.image
        ext = image.content_type.split('/')[-1]
        with open(f'/tmp/pptx-slide-{SLIDE_NUM}-img-{i}.{ext}', 'wb') as f:
            f.write(image.blob)
        print(f'Saved image {i}: {image.content_type} ({shape.width/914400:.1f}x{shape.height/914400:.1f} in)')

Step 4: Interpret the Layout

Shape positions use inches from top-left corner:

• left_in / top_in: position of shape's top-left corner
• Standard slide is 10" × 7.5" (widescreen) or 10" × 5.63" (16:9)
• Shapes with is_image: true and generic names ("Picture 5") are usually clipart
• Group shapes may contain sub-shapes (connectors, arrows) — inspect .shapes on groups

Classifying Slide Content

Shape Pattern	Likely Content
Multiple text boxes + arrows/lines at specific positions	Substantive diagram — reproduce in Typst
Single large Picture shape filling the slide	Clipart/stock photo — skip or replace
Table shape	Data table — reproduce as Typst #table
Text boxes only, no connectors	Text slide — no diagram needed
Group shapes with AutoShapes inside	Flow diagram — extract sub-shapes

Quick Reference

# One-liner to dump all shapes from slide N
uv run python3 -c "
from pptx import Presentation; import json
prs = Presentation('PPTX_PATH')
for s in prs.slides[N-1].shapes:
    d = {'name': s.name, 'text': s.text_frame.text if s.has_text_frame else None,
         'pos': f'{s.left/914400:.1f},{s.top/914400:.1f}',
         'size': f'{s.width/914400:.1f}x{s.height/914400:.1f}'}
    print(json.dumps(d))
"

Rendering slides to PDF/PNG

For actual rasterization (not content extraction), use the shared x2t wrapper — ONLYOFFICE x2t is stateless and parallel-safe, unlike soffice:

# pptx -> PDF (all slides, then split with pdftoppm if per-slide PNGs needed)
python3 ${CLAUDE_SKILL_DIR}/../../scripts/doc_render.py deck.pptx deck.pdf
# pptx -> PNG (first slide only)
python3 ${CLAUDE_SKILL_DIR}/../../scripts/doc_render.py deck.pptx slide1.png

Do NOT call soffice --headless directly — it silently fails on macOS (returns 0, no output) due to profile lock issues. The wrapper prefers x2t and only falls back to soffice where x2t is absent.