lecture-slides for Claude Code: a workflow skill for editable lecture pptx decks

lecture-slides is a workflow skill that takes a lecture deck from a single content JSON file to an editable pptx. Unlike the lecture-slide-make generator from yesterday’s post, it ships no scripts. It is a written procedure that tells the agent what to read, what to decide, and what to check, in what order. The book’s author no longer uses it and now works with a different in-house skill, so treat this one as a design to borrow rather than a tool to install.

What problem it solves

Asking an agent for lecture slides usually produces a deck in one jump, with no agreed outline, no stated venue, and no limit on how much text each slide carries. You find out the deck is too dense only after it is built. Because instructors end up editing the deck by hand, a PDF or HTML version alone does not help either.

This skill puts decisions before generation. The agent reads the writing style, typography, and brand reference documents first, then confirms the outline for the cover, part openers, and content slides. Next it declares a _meta block with the genre, the presentation venue, and the maximum word count, which the generator later uses as validation criteria. Only then does it fill the content JSON, show nine sample slides across three design approaches, and build the full pptx from the one you choose. Each slide holds one concept, and the body takes one of eight forms such as a diagram, table, contrast, cards, or bars, with bullets as the fallback.

Who it fits

The chapter rates four situations. These three cover most decisions.

Situation	Fit	Reason
Decks for VOD lecture sessions	Good fit	Provides a 10- to 14-slide outline
Seminar room presentation decks	Good fit	Body text sizes vary by venue
Proposals centered on a brand	Poor fit	Workflow is dedicated to lecture and course material decks

Setup and first run

There is no public repository. The chapter gives this folder layout as a reference for building it yourself.

# Example folder layout
~/.claude/skills/
  lecture-slides/
    SKILL.md
<lecture project>/
  slides/ch01/
    content.json
    assets/shots/
    ch01.pptx
    ch01.pdf

You do not need to call it by name. It activates when you ask for lecture material or a VOD deck.

Create lecture material on [lecture topic]

From there the agent walks the sequence above: reference documents, outline, _meta, content JSON, nine samples, your choice, then generation and final checks.

What trips people up

The generator is not in the skill folder. SKILL.md contains only the workflow, and it assumes a build_vod_pptx.py script for pptx generation and a compare_deck.py script for comparison exist in a separate kit. If you copy the folder layout and stop there, the agent will follow every step up to generation and then have nothing to run. Prepare a python-pptx-based generator of your own first, or pair the workflow with a generator skill such as lecture-slide-make.

A related catch sits at the end of the process. Missing screenshots leave empty spaces in the deck, so the workflow requires the --strict-assets option just before finalization to confirm none are missing. Your generator needs to support that check, too.

Where this comes from

This post adapts part of chapter 04-07 of The Claude Code & Codex Skills Guidebook. The full chapter covers how a script compares the generated pptx with your hand-edited version to improve the generator, and which external programs, fonts, and review subagent the final checks expect, and the book has 52 more skills like this one. Read the full chapter in The Claude Code & Codex Skills Guidebook on Gumroad.