Turn Raw Content into Slide Decks with a Structured Agent Skill

Where this fits

A developer asks their agent: "Turn these meeting notes into a client deck." The agent has the notes, but no plan for preserving wording, handling speaker notes, or choosing a visual style. The result is often a generic slide dump. The generate-ai-presentations skill gives agents a confirmed step-by-step flow: clarify text handling, parse page markers, choose IP and style, then generate images, PPTX, or slide prompts.

Why agents benefit

• Asks one question at a time: The skill forces agents to confirm text preservation, character IP, visual style, and output format sequentially, avoiding premature generation and rework.
• Parses page annotations correctly: It treats Slide 5: or 投影片 3： as structural markers, not content, so titles and page breaks land where expected.
• Supports multiple source formats: Markdown, transcripts, Word, PDF, article text, or existing slide text all funnel into the same clean 16:9 output pipeline.
• Fits automation loops: Because output can be slide images, PPTX, or per-slide image-generation prompts, agents can hand off to rendering tools or humans without guessing.
• Respects fidelity rules: When the user says "keep every word," the skill prevents summarization or rewriting, which is not a default behavior for most LLM-based slide generators.

Practical scenarios

Converting a course transcript into study slides

The agent receives a long lecture transcript. It detects the user wants the original wording preserved, strips timestamps and page markers, and generates a structured PPTX with the exact key phrases.

Rebuilding an existing deck with a new brand style

A user provides a scraped slide deck PDF and says "make it look like our product style." The skill confirms character IP, applies a flat vector style, and outputs per-slide image prompts that match the new visual direction.

Generating a planning table for a pitch before design

Before any visual work, the agent produces a slide planning sheet (title, content, layout notes) from a rough Markdown outline. The user reviews it before the agent commits to rendering.

Add it to your agent workflow

Install the skill with:

npx skills add hsuliang/generate-ai-presentations-skill --skill generate-ai-presentations

Then point your agent at a source and let it run the confirmation flow:

Agent: “I have your meeting notes. Should the slide text be A) fully preserved, B) polished, or C) summarized?”
User: “C, but keep the action items verbatim.”
Agent: “Got it. I’ll summarize the rest and keep action items untouched. Next: should I include a recurring character visual?”

Compare before adopting

Before making this skill a regular part of your agent's toolkit, compare its decision tree against your actual slide workflow. Check how it handles non-East-Asian text and whether the page-marker parsing is robust for your input sources. Also weigh alternatives like direct PPTX libraries or generic image-prompt chains: this skill wins when you need a structured, confirmation-first flow, but it may be heavier than a simple template renderer.

Why it is worth tracking

The skill’s focus on explicit confirmation and structured slide planning is a useful pattern for agent builders. With 28 stars and a maintainer actively iterating on the GitHub repo, it is still early, but the community signal is credible. For a snapshot of its metadata, usage hints, and updates, check https://www.openagentskill.com/skills/hsuliang-generate-ai-presentations. Revisit it in a few weeks to see how the skill handles new output formats and whether the one-question-at-a-time flow proves durable in real agent runs.