guizang-ppt-skill is an open-source agent skill for generating self-contained HTML slide decks — horizontal-swipe presentations viewable in any browser. Its defining characteristic is broad agent compatibility: the skill is documented to work across Claude Code, Codex, Cursor, and similar agentic coding tools. It reached 10,000 GitHub stars in approximately 25 days as of May 2026, a signal that structured slide generation remains a high-demand unsolved problem even with many competing tools available.
| *Source: GitHub — op7418/guizang-ppt-skill | Weibo post by 歸藏的AI工具箱 (May 2026)* |
What Makes This Different from Other Slide Skills
The crowded slide-generation landscape (Gamma, Beautiful.ai, Tome, platform-specific agent skills) mostly targets one agent ecosystem or produces proprietary formats. guizang-ppt-skill is positioned differently:
| Dimension | guizang-ppt-skill | Typical slide AI tools |
|---|---|---|
| Output format | Self-contained HTML (browser-viewable) | Proprietary format or single-platform |
| Agent support | Claude Code, Codex, Cursor, and similar tools | Usually one agent/platform |
| Agent integration | Drop-in skill file | Platform-specific |
| Approach | Skill/prompt engineering | SaaS product |
The author explicitly prioritized cross-agent consistency: “I did a lot of work to ensure it has very, very good results on any model, any agent.” This is harder than it sounds — slide structure, layout logic, and visual hierarchy are areas where different models diverge significantly.
How It Works
guizang-ppt-skill is a SKILL.md-style agent instruction file. You install it into your agent’s skill/rules directory and invoke it with a topic or outline:
# Claude Code example
/install guizang-ppt-skill
# Then invoke
/guizang-ppt Create a 10-slide deck on "Transformer Architecture for Beginners"
The skill instructs the agent on slide structure, content density, visual layout principles, and output formatting — producing a complete self-contained HTML slide deck viewable in any browser.
Rapid Adoption Signal
Reaching 10,000 GitHub stars in roughly 25 days in May 2026 — in a category with many established competitors — is a meaningful signal. The author’s own reflection on this:
“In a market with so many PPT generation skills already, it still reached this milestone in under a month. Quality and experience are still the deciding factors.”
This is worth noting for the AI education context: it illustrates that execution quality (not just functionality) determines adoption even when the problem space is crowded.
Distinction from HTML PPT Studio
The wiki already has an entry for HTML PPT Studio, which generates HTML-based slide decks optimized for Claude Code. Both tools produce HTML output, so they are more similar than different — the key distinction is scope and community:
| guizang-ppt-skill | HTML PPT Studio | |
|---|---|---|
| Format | Self-contained HTML (horizontal-swipe) | HTML/web-based |
| Agent scope | Claude Code, Codex, Cursor, and more | Claude Code focused |
| Editing after | Browser / code editor | Browser / code editor |
| Best for | Cross-agent portability, broad adoption | Deep Claude Code integration |
Both are good choices for HTML slide generation. Choose guizang when you want a skill that travels across different agents or want the benefit of its large community. Choose HTML PPT Studio when you are Claude Code-only and want tighter ecosystem integration.
How LearnAI Team Could Use This
- Rapid course material prototyping — generate a 15-slide module outline as an HTML deck for instructor review before investing in polished design
- Cross-tool teaching demos — use it to demonstrate how the same agent skill works across Claude Code, Codex, and Cursor; directly relevant to “cross-agent skill design” as a concept
- Student capstone presentations — have students use the skill to generate an initial deck, then refine it — teaching AI-assisted workflow rather than starting from blank slides
- Comparing output quality across agents — run the same prompt through Claude Code and Cursor using this skill; analyze differences as an AI literacy exercise
- Skill file design study — the source SKILL.md is a good example of how to write agent instructions that generalize across different agentic coding tools
Real-World Use Cases
| Scenario | How to use |
|---|---|
| Professor needs a guest lecture deck by tomorrow | Prompt with outline; refine 20% manually; open HTML in browser — done in under an hour |
| Student group project presentation | Generate structure with guizang; each student fills in their section |
| AI course demo: “same skill, different agents” | Run identical prompt on Claude Code and Cursor; compare slides side by side |
| Workshop facilitator needs shareable slides | Output HTML deck → share link or host statically → distribute |
| Researcher summarizing a paper into slides | Paste abstract + key findings; generate a 6-slide HTML summary deck |
Important Things to Know
- “Works across all agents” has limits — the skill is documented for Claude Code, Codex, and Cursor; actual output quality will vary by agent capability; test on your target tool before relying on it for important deliverables
- Output is HTML, not .pptx — the generated deck is a self-contained HTML file for browser viewing; if you need a traditional PowerPoint file, this is not the right tool
- Star counts are volatile — 10,000 stars as of May 2026; GitHub stars reflect hype as much as production utility
- Output needs review — AI-generated slides routinely have layout inconsistencies, overly dense text, or misplaced emphasis; treat the output as a first draft
- Skill file format matters — how you install and invoke it depends on your agent; Claude Code, Cursor, and Codex each have slightly different skill/rule integration paths
- Compare with your existing workflow — if your team already uses HTML PPT Studio, evaluate guizang on cross-agent portability and community size rather than output format (both produce HTML)
- Open source and forkable — the SKILL.md source is available; you can customize the instructions for your specific domain (CS education, research presentations, etc.)