hurricanewrist/agent-zero — reverse-engineered prompt

Reverse engineered prompt

GitHub

Build me a self hosted AI agent workspace like Agent Zero. I want to run it locally, preferably in Docker, open a web UI, connect my AI provider, and give the agent real tasks.

The agent should be able to use a Linux environment, run terminal commands, write and edit code, manage files, browse the web in a visible browser, create tools when needed, and remember project context. I also want a shared canvas where I can watch what it is doing, open documents, review browser sessions, and add comments or corrections while it works.

Please include projects, skills, plugins, prompts, knowledge, logs, and a way to keep user files separate. Add support for office style documents, spreadsheets, and presentations if possible, plus a simple connector so the agent can work on my real local machine when I explicitly allow read and write access.

Make it usable with clear setup steps for Mac, Linux, Windows, and Docker. Look up current docs online if you need to.

Want more depth? Deep Reverse