zhongerxin/cowart — reverse-engineered prompt
Reverse engineered prompt
Build me a local Codex plugin called Cowart that opens an infinite visual canvas in the browser for the current project. I want it to feel like a simple whiteboard for thinking, marking things up, and working with images, but everything should run locally and save into the project’s own canvas folder instead of inside the plugin repo.
The main flow I need is this. From Codex I should be able to say open the Cowart canvas for this project, and it should start a local web page and load the canvas. On the canvas I should be able to create a special AI image holder, select it, then ask Codex to generate an image into that selected holder at the right size. I also want to upload or paste a screenshot of a Cowart image with annotations and arrows, then ask Codex to create a clean revised version beside the original without moving or deleting the original.
Please wire up whatever MCP tools or skills are needed so Codex can read selection state, insert images, and save assets locally. Look up current docs online if you need to.
Want more depth? Deep Reverse