Guizang PPT Skill
An AI-agent skill that generates polished single-file HTML slide decks with editorial magazine and Swiss International layouts, plus image prompts and presenter tools.
Evidence shows: no permission declarations, user confirmation processes, data flow descriptions, sensitive data handling, dependency security, external effects, rollback mechanisms, or source attribution are provided. All trust criteria lack evidence, hence score 0.
Evidence shows: README and SKILL.md descriptions are consistent, workflow files exist, but no dependency list or failure handling details are provided. Self-consistency scores 2, dependency availability scores 1 (no dependencies listed), failure messages scores 1 (no error handling documentation).
Evidence shows: README clearly lists fits and doesn't fit scenarios, capability boundaries are clear, trigger phrases are explicit, platform support list is detailed. Hence each criterion scores 2.
Evidence shows: directory structure is clear, installation instructions are detailed, naming is stable, examples and FAQ exist, known limitations are explicit, license is AGPL-3.0. However, versioning changelog is missing, maintenance responsibility is not clear. Hence information architecture, install notes, naming stability, examples and FAQ, known limitations, license score 2, versioning changelog and maintenance responsibility score 1.
Evidence shows: output is single-file HTML, directly usable, marginal value is high (multiple layouts and presenter mode), cost-benefit is reasonable (no build required). Hence each criterion scores 2.
Evidence shows: claims in README are not backed by verifiable test results or external validation, facts and inferences are not clearly separated. Hence each criterion scores 1.
- No permission declarations or data flow descriptions provided; review code before use.
- Dependencies not listed; potential supply chain risk.
- Versioning changelog missing; updates hard to track.
What does this agent do, and when should you use it?
Guizang PPT Skill is a structured skill for Claude Code / Codex agents to generate single-file HTML horizontal-slide decks, supporting two visual systems: Style A (editorial magazine, narrative-focused) and Style B (Swiss International, grid-driven for facts/products/analysis). It includes reusable templates, 22 locked Swiss layouts, 10 magazine layouts, preset themes (5 e-ink, 4 Swiss anchor colors), and a full presenter mode with dual-window audience view, speaker notes, rehearsal timing, laser pointer, and annotation. The skill guides agents through a 7-question clarification, content filling, optional Codex-based image generation, automatic quality checks via Node scripts, and final preview.
Takes user input (article, Markdown), runs a clarification workflow (style, audience, duration, etc.), copies template.html (Style A) or template-swiss.html (Style B), fills content based on layout skeletons, and optionally generates images via GPT-Image 2.0 / GPT-M 2.0 in Codex, inserting them into predefined image slots. Generates presenter notes (purpose, points, transitions) and validates against checklist.md and scripts (validate-swiss-deck.mjs, validate-presenter-mode.mjs). Output is a standalone HTML file with presenter mode (press 'P') featuring audience sync, timer, rehearsal log, automatic page turning, laser pointer, and drawing tools.
- A speaker turns a long article into a 6-10 page Swiss-style deck for an internal talk or demo day.
- A product analyst uses Style B to create a fact-driven presentation with 21:9 hero images and locked layouts.
- An individual creates an editorial-style deck for a personal sharing session, emphasizing narrative and personal voice.
- A content creator generates platform covers (WeChat 21:9, 1:1 share card, Xiaohongshu 3:4) from the same source material.
- A presenter wants to rehearse with speaker mode, using generated notes and timing to practice.
- A designer uses Codex to generate custom images (photos, infographics, UI mockups) for existing PPTs, matching template proportions.
What are this agent's strengths and limitations?
- Dual design systems with strict Swiss layout lock ensure high-quality, consistent output.
- Single-file HTML output requires no build step or server, simplifying sharing and presentation.
- Built-in presenter mode with audience sync, rehearsal timing, and laser pointer enhances live delivery.
- Automated validation scripts catch layout issues, making it reliable for large decks.
- Integrates with Codex image generation and cover generation, covering the full visual content pipeline.
- Tightly coupled to agent environments (Claude Code / Codex); unreliable on plain chatbots without filesystem/browser.
- Delivers HTML only; export to PPTX requires manual conversion, not a core workflow.
- Theme colors are preset and non-customizable, limiting brand-specific or bespoke design.
- Requires shell and file system access, which may be restricted in some managed AI platforms.
How do you install or deploy this agent?
Recommended one-liner:
npx skills add https://github.com/op7418/guizang-ppt-skill --skill guizang-ppt-skill
Or manually clone to ~/.claude/skills/guizang-ppt-skill:git clone https://github.com/op7418/guizang-ppt-skill.git ~/.claude/skills/guizang-ppt-skillVerify SKILL.md, assets/, and references/ exist.
How do you use this agent?
After installation, ask your agent:
"Create a Swiss-style PPT from this article, about 7 pages, with 2-3 images."
Or try:
"Turn this Markdown into a magazine-style presentation."
"Based on this PPT's core ideas, generate a WeChat 21:9 header image."
"Add speaker notes and per-slide timing, then rehearse with presenter mode."
Open the generated HTML and press 'P' to enter presenter mode.