Hi everyone,
I wanted to share a small original project I built while experimenting with Codex skills:
codex-excalidraw-skill
The idea was to make Codex generate diagrams that are not just static images, but real editable Excalidraw scenes. The skill creates structured .excalidraw JSON, SVG/HTML previews, ready-to-open Excalidraw Web links, and local session state so follow-up requests can modify the previous diagram instead of starting from scratch.
I built it because I wanted a workflow closer to an interactive diagramming assistant: ask Codex for a visual, inspect the preview, open it in Excalidraw, and keep iterating. The repository includes the skill instructions, helper scripts, references, examples, and an example tactical diagram rendered in the README.
A few things the skill currently supports:
- Natural-language requests to editable Excalidraw diagrams
- SVG and HTML previews
- Open-in-Excalidraw web links
- Session-state reuse for incremental edits
- Validation checks for unreadable text and broken scene structure
- Specialized templates for tactical football rosters and building elevations
This is my own implementation, created through hands-on testing and iteration in Codex. I did not copy an existing skill; I used the project as a way to learn how far Codex skills can go for visual, editable artifacts.
I am sharing it here in case it helps other builders thinking about Codex skills, diagram generation, or editable AI-created artifacts.
I would love feedback from the community, especially on:
- How to make the skill easier for others to install and adapt
- Better patterns for preserving manual edits across follow-up requests
- Whether this kind of workflow should eventually become a proper MCP/widget integration
Made by Fabricio Artur.