Skip to content

About

No description, website, or topics provided.

Resources

Stars

2 stars

Watchers

0 watching

Forks

Latest commit

 

History

4 Commits

Folders and files

Repository files navigation

JustSlides

Made for presenting, not decorating.

An agent skill for Claude Code and Codex that turns a topic into a finished, click-by-click slide deck with a speaker script, as a single HTML file.

Presenter view: current slide, next-slide preview, clicks left on this slide, and the speaker notes for the current click highlighted

What This Does

Most slide generators stop at how the slides look. JustSlides is about how you give the talk. The style stays plain on purpose: white slides, one serif typeface, one accent color, so the audience follows your argument instead of the decoration. You give it a topic, and it plans the talk, builds every diagram one click at a time, and writes what to say after each click. You get one .html file that opens in any browser. You present it with the arrow keys, and your notes follow along in a second window.

It is built for teaching and research talks: lectures, discussion sections, micro-teaching lessons, group meetings and paper presentations.

Key Features

  • Speaker notes tied to clicks: the script is split by ★, one segment per click, and the sentence you should be saying right now is highlighted.
  • Presenter view: a second window with a timer, the next slide, and how many clicks are left on this slide. It stays in sync with the projector window.
  • Built step by step: boxes appear one at a time, arrows draw themselves, numbers count up, and elements glide between slides like Keynote's Magic Move.
  • Academic structure: hook, one running example, one idea per slide, evidence, discussion questions, takeaways. Real citations in footnotes, no invented numbers.
  • Checks its own work: a checker catches mismatched clicks and notes, and a screenshot script lets the agent look at every slide and fix overflow before handing it over.
  • Zero dependencies: one HTML file with inline CSS and JS. No npm, no build step, works offline.

🎤 Presenting: the only part you do yourself

Everything else is automated. On stage, open the .html file in a browser and use these keys.

Key Action
➡️ → Space Enter next click, then next slide
⬅️ ← Backspace previous click / slide
⏭️ Shift+→ / Shift+← skip a whole slide forward / back
🧑‍🏫 P presenter view in a second window
📝 N notes panel under the slide (same window)
🔲 G overview grid of all slides, click one to jump
✨ A animations on / off
🖥️ F fullscreen
⏮️ Home / End first / last slide
🖨️ Cmd/Ctrl+P print to PDF, one page per slide

🖱️ Clicking the right two-thirds of the slide advances and the left third goes back. 📱 Swipe works on touch screens. 🔗 deck.html#7 opens slide 7.

💡 Recommended setup for a talk: open the deck, press F for fullscreen on the projector, then press P and drag the presenter window to your laptop screen.

Gallery

Generated from one prompt

output/paper-diffusion-transformers.html was made by the skill from the prompt "create slides explaining DiT": 14 slides, 43 clicks, full speaker script.

DiT: Work in Latent Space, Not Pixel Space DiT: adaLN-Zero block diagram DiT: State of the Art on ImageNet table

Examples the skill learns from

Open any of these in a browser and press →.

Paper presentation: Native Sparse Attention shows magic move, fade-outs, count-ups and embedded figures.

NSA: three branches combined by a gate NSA: one query at 64k context, step by step NSA: results on a 27B MoE model

Teaching lesson: Binary Search shows polls, exercises, a playable guessing game, and a tree built one level at a time.

Binary search: what strategy did I use Binary search: why so few guesses, as a tree Binary search: other search methods

Course discussion: Agents and Images shows box-and-arrow diagrams, SVG connectors and discussion questions.

Agents: from text to pixels Agents: how a model reads an image Agents: acting on an image

Installation

Claude Code

Clone the repository and start Claude Code inside it. The skill in .claude/skills/justslides/ is picked up automatically.

git clone https://github.com/YumengHe/JustSlides.git
cd JustSlides
claude

To use it in every project, copy the skill into your personal skills directory:

cp -R .claude/skills/justslides ~/.claude/skills/

Codex and other agents

The skill is plain Markdown. Send the agent this repo link and ask it to use the JustSlides skill, or point it at .claude/skills/justslides/SKILL.md. Everything the skill needs (engine template, layout snippets, checker, screenshot script) is inside that folder.

Usage

/justslides make a 15-minute lecture on public-key cryptography for a second-year CS class

or just ask for slides in plain words; the skill triggers on "slides", "deck", "lecture", "paper presentation", "幻灯片", "课件" and similar. The skill will:

  1. Ask who is presenting, unless it already knows
  2. Outline the talk, about one slide per minute, with what each click reveals
  3. Write the slides into output/<type>-<topic>.html, where the type is lesson, discussion or paper
  4. Write the full speaker script, one ★ segment per click
  5. Run the checker and screenshot every slide to fix layout problems
  6. Tell you the file path, slide count and how to present

Repository layout

.claude/skills/justslides/
├── SKILL.md               instructions the agent follows
├── template.html          engine + two placeholder slides; every new deck starts as a copy
├── reference/patterns.md  copy-paste snippets for 17 slide types
├── scripts/check_deck.py  validates click counts, notes and JS syntax
└── scripts/screenshot.sh  renders every slide to PNG with headless Chrome
examples/                  finished decks that show the style
output/                    decks generated by the skill
assets/                    images used in this README

Writing a deck by hand

Copy template.html, add <section class="slide"> blocks between the SLIDES START/END markers, and add one string per slide to the NOTES array. Three attributes drive the animation:

<section class="slide" data-steps="2">                <!-- number of clicks on this slide -->
  <div class="ttl a fx-down" data-s="0">Title</div>    <!-- data-s: appears after click k -->
  <p class="txt a fx-up" data-s="1" style="--d:.2s">… <!-- --d: delay within the click -->
  <div class="call a fx-up" data-s="2">…</div>
</section>

In the notes, ★ marks a click: the text after the k-th ★ is what to say after click k, so each note contains exactly data-steps stars. Run the checker to confirm:

python3 .claude/skills/justslides/scripts/check_deck.py my-deck.html

See reference/patterns.md for cards, tables, diagrams, magic move, fade-outs, count-ups and interactive widgets.

License

MIT. See LICENSE.

About

No description, website, or topics provided.

Resources

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages