different-ai/openwork ? reverse-engineered prompt

I want to build an open-source desktop app, something like an alternative to Claude Cowork. It should be a user-friendly GUI for running AI agents, with a strong focus on being local-first.

A user should be able to pick a project folder, type a prompt, and see the agent's execution plan as a timeline of steps. Crucially, the app must ask for permission before the agent can do things like run commands. It should also have some kind of "skills manager" in the UI so you can add new plugins or capabilities to the underlying agent.

While it should mostly run on the user's machine, there needs to be an option to connect to a remote server too. Make it a modern app that works on macOS and Linux, feel free to use Tauri and a nice web framework for the frontend. You can look up the latest docs for the best approach.