/pptx

name: pptx
description: Read and inspect existing .pptx PowerPoint files (text, speaker notes, slide images). Use only for reading uploaded PowerPoint files — to AUTHOR a deck use the slides skill instead.

pptx — PowerPoint presentations

Reading and inspection of `.pptx` files. Editing and creation are not yet supported.

> **Making a presentation, slide deck, or "slides"?** This is NOT the > skill — use the **`slides`** skill (`/app/org/public/core/slides/SKILL.md`), > which authors HTML decks with a live, editable preview. Reach for `pptx` > ONLY when the user explicitly works with a PowerPoint `.pptx` file — > reading one they uploaded, or when they specifically ask for `.pptx` > output (not yet supported here).

Quick reference

| Task | Command | | ----------------------------- | --------------------------------------------- | | Plain-text dump | `pptx-extract <file.pptx>` | | Plain-text + speaker notes | `pptx-extract --notes <file.pptx>` | | Visual overview (grid image) | `pptx-thumbnail <file.pptx>` | | One slide at full resolution | `pptx-rasterize --pages 3 <file.pptx>` | | Several slides | `pptx-rasterize --pages 1,3-5 <file.pptx>` | | Raw XML inspection | `pptx-unpack <file.pptx>` |

Every command writes its output next to the input file and prints the output path(s) on stdout.

Reading text

`pptx-extract` prints each slide as a `## Slide N` section followed by the visible text. Speaker notes are excluded by default; pass `--notes` to include them under a `### Notes` subsection per slide.

Visual overview

`pptx-thumbnail` produces two artifacts:

• A composite grid image at `<input-stem>.thumbnail.jpg` — 4-column grid

with a small slide-number overlay on each cell. Suitable for a single multimodal read to get the shape of the deck.

• A directory `<input-stem>.slides/` containing per-slide JPGs at the

same resolution. When a slide in the grid looks interesting, read `slide-NNN.jpg` directly instead of re-running `pptx-thumbnail`.

To actually see the rendered image, call the `read` tool on the path `pptx-thumbnail` printed:

read /home/sandbox/deck.thumbnail.jpg

The image is injected into the next turn as a vision input.

Per-slide rasterization

`pptx-rasterize` renders specific slides at higher DPI (default 150) into `<input-stem>.rasterized/slide-NNN.jpg`. Use `--pages 3` for a single slide or `--pages 1,3-5` for a selection. Override DPI with `--dpi`.

Pass the resulting path to the `read` tool to actually see the slide.

Raw XML

`pptx-unpack` unzips the `.pptx` into `<input-stem>.unpacked/`. Useful locations inside:

• `ppt/slides/slideN.xml` — slide content
• `ppt/notesSlides/notesSlideN.xml` — speaker notes
• `ppt/media/` — embedded images
• `ppt/theme/` — color palette and font definitions

Direct python-pptx usage

For programmatic access beyond what these scripts cover (table cell data, shape positions, layout metadata, etc.), import `pptx` directly. The library is preinstalled.

from pptx import Presentation
prs = Presentation("/path/to/file.pptx")
for slide in prs.slides:
    for shape in slide.shapes:
        ...