What this is

This is an open-source Skill called dashi-ppt-skill (a "Skill" is a plugin-style package that teaches an AI Agent how to perform a specific task), which has earned 4.3k stars on GitHub. Its approach is straightforward: feed your document to an AI Agent such as Claude Code, Codex, or Doubao—agents capable of autonomously executing multi-step tasks—and the Agent uses this Skill to produce a web-based PPT. But the key isn't the "generation" itself; it's that "the generated output is itself a visual editor."

You can drag sliders, click dropdowns to switch layouts, edit text inline, drag-and-drop images, and one-click swap color schemes—all with full WYSIWYG preview. The final exported PPTX isn't a "picture-based PPT" (where text is rasterized into images and becomes uneditable in PowerPoint); it's a real, node-by-node reconstructed editable file. Everything runs locally on your machine, with no document content uploaded to any server.

The project also bundles 12 visual themes, 1,020 layouts, and 8,576 adjustable components, with built-in chart types including radar, waterfall, and Sankey diagrams, plus analytical model layouts such as SWOT and Porter's Five Forces. Installation is a single command: npx dashi-ppt-skill@latest.

Industry view

Supporters argue this approach hits a real pain point. The AI PPT space has been crowded in the past two years—Gamma, AiPPT.cn, ChatPPT, and many others—with most players competing on "generation quality": more templates, flashier charts, smoother animations. But what actually consumes users' time has never been "generation"—it's "editing." When your boss asks you to tweak a number, add a competitor-comparison slide, or change the cover color, these "small edits" are more painful in most AI PPT tools than regenerating the whole deck. dashi-ppt-skill's logic is "the output isn't the endpoint, it's the starting point." This idea could reshape how the AI PPT space plays out—tools that "can only generate but not edit" will increasingly struggle to sell.

But there are sober voices too. First, the open-source Skill format means it depends on Agent runtimes like Claude Code, Codex, and Doubao, which creates a barrier for ordinary users who haven't used AI Agents yet, and adoption may not go as smoothly as SaaS (subscription-based online software) products. Second, 1,020 layouts sounds impressive, but "layout richness" is something mature players like Gamma have already done deeply—whether an open-source tool can keep up with maintenance and updates is an open question. Third, the AGPL license (which requires derivative works to also be open-sourced) raises compliance concerns for some enterprise users, and the path to commercialization may not be smooth.

Impact on regular people

For enterprise IT: worth a small-scale pilot. If your company is already using Claude Code or similar tools, this Skill can be deployed at near-zero cost, and documents stay on-premises, making it relatively compliance-friendly. However, it's not recommended for procurement as an "enterprise-wide PPT solution"—it's more of a developer's tool than a turnkey product.

For individual professionals: for high-frequency PPT scenarios like industry research, fundraising reviews, and project reports, you can adopt a workflow of "AI-generated first draft + your own polishing" to skip the from-scratch formatting time. But if you don't have an AI Agent set up yet, just configuring the environment may be enough to turn you off—first assess whether you're willing to tinker.

For the consumer market: the bigger signal here is that the next battleground for AI office tools is shifting from "can it generate?" to "can it continue to edit?" In the next six months, you'll very likely see a wave of AI PPT tools announce "we support editing too"—and that is what this open-source project truly leaves behind for the industry.