The pptx skill is the PowerPoint guide from Anthropic’s public skills repository, one of the document skills that power Claude’s file creation features. It triggers whenever a .pptx or .potx file is involved, or the user mentions a deck, slides or a presentation, and picks the approach by task: write a pptxgenjs script to create a new deck, unpack the archive and edit ppt/slides/slideN.xml to change an existing deck or fill a template, and run slide_text.py or thumbnail.py to read one.
Much of the guide is a list of pptxgenjs footguns, from hex colors with a # that corrupt the file to shared option objects, negative shadow offsets and stacked-chart label positions. It asks for structured decks built on a theme, named layouts, placeholders and sections, with apply_theme.js writing the theme’s colors into the file afterwards. It then covers deck structure, design ideas with palettes, safe fonts and a list of things to avoid, and a required QA pass: slide text checks, validate.py, and a render to images with LibreOffice and pdftoppm to look at every slide.
What You Get
- A SKILL.md guide with a task table for creating, editing and reading .pptx and .potx files
- A list of pptxgenjs gotchas covering layout size, hex colors, shadows, lists, charts, combo axes, icons, slide layouts and unescaped XML
- A structured-deck method: a theme object, scheme colors, sections, one layout per slide frame and placeholders filled by name
- slide_text.py, which prints every slide’s title, text, tables, chart data, alt text and speaker notes as markdown
- thumbnail.py, which renders a labeled grid of slides for picking template layouts
- add_slide.py and clean.py, which duplicate slides or layouts with all the package bookkeeping and remove orphaned slides, media and relationships
- apply_theme.js, which writes the theme’s colors and name into a pptxgenjs deck
- validate.py, which checks schemas, relationships, content types, charts and slides and names a fix for each failure
- Design guidance: palettes, typography with QA-safe fonts, spacing, and a list of common mistakes to avoid
Common Use Cases
- Building a new pitch deck or report deck with native charts, icons and speaker notes
- Filling a company template with new content while keeping its layouts and styling
- Extracting the text, tables and notes from a deck to use in a summary or email
- Adding, duplicating, reordering or deleting slides in an existing presentation
- Converting a legacy .ppt to .pptx, or rendering a deck to images to check it
What Makes It Unique
Like the docx and pdf skills from the same repository, it is Anthropic’s own reference for a complex skill used in production, and like docx it splits the work by whether the file exists yet: pptxgenjs for new decks, direct XML edits for existing ones and templates. Where those skills stop at producing a valid file, this one also has opinions about the deck itself: one message per slide, a theme-and-layout structure that can be restyled in one place, palettes, fonts chosen so the LibreOffice preview measures text the way PowerPoint will, and a list of patterns that read as AI-generated slides. Its validator targets faults PowerPoint refuses but other tools accept, such as chart XML pptxgenjs writes.
Under The Hood
The guide has the agent write and run its own Node script requiring pptxgenjs, which it expects to be preinstalled and installs from npm only if the require fails, and the same for react-icons, react, react-dom and sharp when it renders icons. It shells out to Python, zip, pdftoppm and LibreOffice. slide_text.py reads a deck with python-pptx and prints it. thumbnail.py renders the deck to PDF and JPEG pages in a temporary folder and writes the grid images to the working directory. add_slide.py writes new slide parts and edits the content types, relationships, presentation.xml and app.xml; given a .pptx without -o it rewrites that file in place. clean.py deletes slides, media, notes, themes and relationships nothing references in an unpacked folder, and refuses when it cannot resolve the relationships. apply_theme.js loads the deck with jszip, rewrites its theme part, mends one chart fault, drops ZIP folder entries and writes the deck back. validate.py unpacks with a safe extractor that refuses symlinks and paths outside the target folder, and with —auto-repair writes fixes back into the file. soffice.py runs LibreOffice headless; before rendering it copies the deck with links to outside files cut, and where the sandbox blocks Unix sockets it writes a small C shim to the temp folder, compiles it with gcc and loads it with LD_PRELOAD. The shared office folder also carries the docx and redlining validators, which this skill does not use.