PPT-as-code takes a different approach to presentation creation: it treats it as a carefully staged, artifact-driven workflow rather than a freeform coding exercise or manual slide design. What makes this repo worth a closer look is its visual workbench prototype that supports editing via a unified JSON deck model or direct HTML manipulation — bridging AI-generated deck planning with hands-on visual editing.
What PPT-as-code does and how it is built
At its core, PPT-as-code is an open-source AI skill package that enforces a multi-stage pipeline for generating presentations. Instead of letting users jump straight to slide code or visuals, it requires following a sequence of well-defined steps: source normalization, scene mapping, deck breakdown, script and image planning, pre-HTML quality assurance, static HTML generation, and optionally exporting to PPTX.
This staged pipeline reflects an artifact-driven workflow where each step produces explicit intermediate outputs. This design aims to improve control, quality, and predictability when using AI to generate presentation content and layout.
The repo is written primarily in JavaScript, designed to be environment-agnostic. It avoids hardcoded workspace assumptions and does not depend on live web search, offering explicit fallbacks for offline or no-network operation.
The standout architectural component is the visual workbench prototype. This workbench introduces a canvas editing environment with snapping support and two distinct editing modes:
Deck Mode: Uses a
deck_model.jsonas the single source of truth for the presentation structure. This mode enforces the staged artifact pipeline and structured deck planning.HTML Mode: Allows direct editing of the generated static HTML slides. This mode offers more visual flexibility and manual tuning.
This dual editing route is a clever compromise between structured AI-driven deck planning and the need for hands-on visual manipulation, a gap most presentation tools struggle to bridge.
The skill supports three operational modes:
- Quick: Minimal confirmation checkpoints, faster iterations.
- Basic: Balanced checkpoints and design rigor.
- Advanced: High confirmation and detailed design enforcement.
This tiered approach caters to different user needs, from rapid prototyping to more controlled, quality-focused workflows.
Technical strengths and architectural tradeoffs
The repo’s disciplined pipeline reflects a clear opinion on how AI-generated presentations should be built: through explicit staged artifacts rather than end-to-end black-box generation. This makes reasoning about output quality and debugging easier.
The dual editing model is the most interesting technical aspect. Using a JSON deck model as a unified, canonical source enables tracking and versioning of the presentation structure independently from the rendered slides. This approach supports iterative improvements in AI planning stages without losing manual visual adjustments.
At the same time, the HTML editing mode acknowledges that no AI pipeline is perfect and that users often need direct visual control over final slide output. Supporting both editing routes within one workbench is a nuanced design that balances automation and manual craftsmanship.
The environment-agnostic stance is another practical strength. By avoiding mandatory web search and having no-network fallbacks, the skill can run reliably in controlled or offline environments, a necessity for privacy-conscious or enterprise use cases.
On the flip side, this staged approach and multi-mode editing add complexity. The user must understand the pipeline stages and decide when to switch between deck model and HTML edits. This can increase the learning curve compared to simpler presentation tools.
The project is also clearly a prototype, with the visual workbench described as such. This means some rough edges and missing polish are to be expected. The lack of ready-to-use installation commands or automated setup might slow adoption.
Overall, the code appears thoughtfully structured, favoring clarity over clever hacks. The staged pipeline and artifact-driven design show strong architectural thinking, and the dual editing modes are a rare feature in AI presentation tooling.
Installation notes and getting started
The README offers straightforward installation notes:
## Installation Notes
This folder is stored as `ppt-as-code-open` to avoid colliding with a private local version.
If you want to publish or install it as the canonical package:
1. rename the folder to `ppt-as-code` if needed
2. keep the skill `name` in `SKILL.md` as `ppt-as-code`
3. install it in your preferred skill directory or publish it from this folder
There are no explicit commands beyond these instructions, reflecting the project’s prototype and modular nature. Users will need to integrate it into their AI skill environment manually.
who should consider using PPT-as-code
PPT-as-code is best suited for developers or teams experimenting with AI-driven presentation generation who want a structured, artifact-driven approach. Its staged pipeline and visual workbench offer a balance between automated deck planning and manual visual editing.
If you need a fully polished, turn-key presentation tool, this repo is not there yet. The prototype nature and manual setup mean it’s more a technical foundation or research playground than a ready production system.
However, if you’re building AI skills or tooling around presentation workflows, especially where controlling intermediate artifacts and dual editing paths matters, PPT-as-code provides a solid starting point and a novel architectural approach worth understanding.
The tradeoff is complexity and learning curve versus control and flexibility. The repo is most valuable if you want to build on or customize AI presentation pipelines rather than just use a finished product.
In summary, PPT-as-code tackles a real pain point—bridging AI-generated deck planning with visual editing—through a staged pipeline and innovative dual-mode workbench. It’s a promising approach that highlights the challenges and possibilities of AI-assisted creative workflows.
→ GitHub Repo: Russell-cell/PPT-as-code ⭐ 190 · JavaScript