LibreOffice integration

[KeithCu]

Hello,

I’m working on WriterAgent, a LibreOffice extension that integrates AI agents directly into Writer, Calc, and Draw/Impress. It provides a chat panel, background NumPy execution in a sandboxed virtual environment, and specialized UNO-driven document tools.

Lately, I’ve been looking into integrating the core concepts of ppt-master, it’s a great system! At a high level:
1. Source Parsing: You select text in Writer (e.g., a project outline) or have the document research agent gather the data, and tell the AI chat: "Generate a 5-slide summary deck from this context."
2. Visual Consistency (Worker Process): An isolated background Python worker plans the slide strategies and creates a cohesive design_spec (colors, typography, spacing constraints) to ensure slides 1 through 5 don't drift in style.
3. Native Impress Generation (UNO): Instead of outputting static image files or PDFs, WriterAgent maps the layouts to native LibreOffice drawing elements (TextShape, RectangleShape, groups, etc.).

The draft dev plan is here.
Is this a feature people would find useful inside LibreOffice Impress?
Are there specific features you think we should prioritize during development?

If anyone feedback or would like to work on it, let me know.

Thank you!

-Keith

[hugohe3]

Hi Keith,

This sounds like a very interesting direction. I’d be happy to see ppt-master’s ideas explored inside LibreOffice / WriterAgent.

The project is MIT licensed, so you’re welcome to reuse, adapt, or integrate any part of it as needed. If the integration works well, feel free to share progress here. I think native editable Impress generation could be useful for LibreOffice users.

Good luck with the development!

[KeithCu]

Hi Hugo,

Thank you for the encouragement! I researched it and made a prototype version. It’s tricky because this is built as a skill for Claude, and inside LibreOffice it doesn’t have all these shell, etc. capabilities. I do have an integration doc so far: https://github.com/KeithCu/writeragent/blob/master/docs/ppt-master-integration-plan.md

The big thing I noticed is that the skill file is huge and seemingly written by Claude and for Claude. Whereas WriterAgent supports any AI but is focused on working reasonably with smaller, local models.

Have you done any work to decrease the intelligence required to get good results by, for example, trying to simplify skill.md? There’s even a framework called dspy which can help automate this process.

If you can expand this to work beyond the frontier models, it could be very helpful.

[hugohe3]

Yes, that's a fair point. The project fundamentally depends on Python and an agent-based workflow, so adapting it to an environment without the same shell capabilities—and aiming to work well with smaller local models—will indeed be quite difficult. Simplifying skill.md could help, but the underlying Python and agent dependencies are still significant.