Guizang PPT Skill: An Agent-Driven HTML Deck Generator with Two Strict Visual Systems
AI-agent Skill for generating polished HTML slide decks: editorial magazine and Swiss layouts, image prompts, social covers, and a WebGL/low-power presentation runtime.
At a glance
- What is it?
- Guizang PPT Skill turns Claude Code or Codex into a slide factory that outputs single-file HTML decks, with a magazine style for narrative and a Swiss grid style for facts. It pairs those layouts with a local presenter runtime, but it is not a replacement for dense data tables or collaborative editing.
- Who is it for?
- Adopt Guizang PPT Skill if you are a solo speaker, a startup founder, or an internal team lead who wants a fast, agent-generated HTML deck with a distinct visual identity and no build step. Skip it if your content is mostly large tables, if you need multi-user editing, or if your audience expects a standard 16:9 PPTX workflow.
- Can I use it commercially?
- Yes, with strict conditions. AGPL-3.0 is a network copyleft licence: if people use a modified version over a network, for example as a hosted service, you must offer them its source code under the same licence.
- Is it still maintained?
- Yes. The repository last received commits 40 days ago.
- What is it written in?
- Mainly HTML, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What Problem It Solves and Who It Is For
The project addresses a specific pain: AI agents generate Markdown well, but Markdown slides look plain and are hard to control. Guizang PPT Skill gives Claude Code or Codex a structured workflow to produce single-file HTML decks with editorial polish. The README states it is suited for offline talks, internal industry speeches, private sharing sessions, AI product launches, demo days, and presentations with a strong personal style. It is explicitly not suited for content with large data tables, training courses that need high information density, or multi-person collaborative editing. The target user is someone who already works with an agent that has shell access and a browser, not a general chatbot user.
Two Visual Systems: Magazine and Swiss Grid
The skill ships two distinct visual systems, not one reskinned template. Style A, called electronic magazine or e-ink, is described as Monocle with code, aimed at narrative, opinion, and personal expression. It offers 10 layouts including cover, chapter, data poster, image grid, pipeline, and comparison. Style B is Swiss International Style: a 16-column grid, right-angle blocks, 1px hairline rules, no shadows, no gradients, no rounded corners, and a single high-saturation anchor color. Style B has 22 named layouts from S01 to S22, and the documentation says the body pages must only pick from those, not invent new structures. The Swiss system also locks image slots to template-defined data-image-slot attributes and requires common hero images in 21:9 or 16:10 ratios. This is a stricter system than a CSS swap, and the README explicitly says so.
Installation and Triggering: Real Commands
Installation is straightforward. The recommended command is npx skills add https://github.com/op7418/guizang-ppt-skill --skill guizang-ppt-skill. Alternatively, you can paste a paragraph into an agent that asks it to git clone the repository into ~/.claude/skills/guizang-ppt-skill and verify that SKILL.md, assets/, and references/ exist. Manual clone works too. Once installed, Claude Code auto-discovers the skill when you say phrases like make me a magazine-style PPT or generate a horizontal swipe deck. The workflow is a structured sequence: choose a style, answer a 7-question clarification list, copy the template from assets/template.html or assets/template-swiss.html, fill content, optionally generate images in Codex with GPT-Image 2.0 or GPT-M 2.0, write speaker notes, run self-checks against references/checklist.md, preview in a browser, and rehearse via presenter mode.
The Presenter Runtime: Local, Not Cloud
Both templates embed the same presenter runtime. Pressing P in the lower right opens a presenter view and a separate audience window. The runtime shows current and next slide in a 16:9 preview, a grid for page selection, structured notes with purpose, talking points, and transitions, plus progress and connection status. It tracks elapsed, per-page, and remaining or overtime time. Rehearsal data is stored locally in the browser, and the README stresses there is no AI scoring. Auto-advance is off by default and only turns on if the outline gives explicit seconds or the user enables a global interval. The runtime includes a laser pointer, circle tool, black and white screen, freeze, and recovery when the audience window closes or loses sync. A pre-talk check covers popups, fullscreen, fonts, images, video, and 16:9 preview, and it reminds the user to manually confirm HDMI, adapters, and projector. This is a genuinely local tool, which means no cloud relay or phone remote, but it also means the presenter must manage the audience window manually.
Validation Scripts and the Swiss Quality Gate
A key differentiator is the validation layer. The Swiss style is enforced by scripts that block centered titles, experimental layouts, text inside SVG, and images outside their slots. When Playwright is available, the validation goes further and measures real rendered overflow, bottom whitespace, and title gap. The README gives a specific command for presenter mode validation: node scripts/validate-presenter-mode.mjs path/to/index.html, with an optional --target-minutes 30 flag. The checklist.md file collects lessons from real talks, and the skill requires all P0 issues to pass before delivery. This is an unusual level of quality control for a skill, and it suggests the author cares about output consistency. However, the validation scripts are only as good as the Playwright environment; without it, the checks are static and may miss rendering issues.
Limitations and Wrong-Use Cases
The README is honest about where this fails. Large data tables are a poor fit because the layouts prioritize visual impact over information density. Training courses that need dense content are also not suitable. The output is a static HTML file, so multi-author collaboration is not possible in the same way as a shared PPTX. Another limitation is the platform dependency: the skill requires an agent with file system and shell access, and a browser for preview. Ordinary chatbots without those capabilities cannot reliably produce a full deck. The presenter mode depends on the browser allowing popups and local windows; if the browser blocks the audience window, the speaker must manually reopen it. Also, the skill is designed for horizontal page-flipping decks, which may feel unusual for audiences used to vertical scrolling or standard slide software.
Alternative Approaches and a Real Comparison
The obvious alternative is to use a Markdown-to-PPT tool like Marp or Slidev, which also generate HTML slides but from a plain-text source. The difference is in the workflow and the output control. Marp and Slidev give the user direct control over the source file, and they work without an agent because the user edits the Markdown directly. Guizang PPT Skill instead delegates the entire design decision to the agent, which then enforces a strict layout system. That is a trade-off: you gain a consistent editorial look without learning CSS, but you lose the ability to fine-tune every pixel by hand. Another difference is the presenter runtime; Marp and Slidev have their own speaker notes and presentation modes, but Guizang's is bundled into the single HTML file, so you do not need a separate tool. If you want full manual control and no agent dependency, Marp is a lighter choice. If you want an agent to produce a branded deck quickly and you accept the layout constraints, this skill is the stronger option.
Maintenance, License, and Upgrade Cost
The repository is under AGPL-3.0, which has implications if you distribute modified versions or offer the output as a service. The README does not discuss license details beyond the LICENSE file, so you should check the exact terms before using it in a commercial product. The project has a recent release v1.1.0 dated 2026-05-15, labeled Swiss style and community-ready, which suggests active maintenance. Updating the skill is a simple git pull inside the skill directory, as the README shows with the update prompt. The cost of maintenance is low for the user: you just re-run the pull command and verify the latest commit. However, if you have generated decks with older templates, a new version may change layout defaults, so you may need to regenerate or manually adjust existing decks. The validation scripts are part of the skill, so upgrading also brings new checks, which could force you to fix previously acceptable decks.
Editorial conclusion
Adopt Guizang PPT Skill if you are a solo speaker, a startup founder, or an internal team lead who wants a fast, agent-generated HTML deck with a distinct visual identity and no build step. Skip it if your content is mostly large tables, if you need multi-user editing, or if your audience expects a standard 16:9 PPTX workflow. Before adopting, verify that your agent environment can run shell commands and open local HTML files, and confirm that the AGPL-3.0 license is acceptable for your distribution model. Then test the presenter mode with your actual projector and browser popup settings, because the failure recovery depends on those local conditions.
Community notes