HKUDS/OpenHarness ? reverse-engineered prompt

Build me a personal AI assistant that runs in my terminal. I want to be able to give it high-level tasks, and it should be able to use tools to get things done on my computer. This means it needs to be able to run shell commands, browse the web, and read local files to help me with my work. Safety is a huge priority, so it must always ask for my permission before it executes a command, especially if it could change or delete files.

I also want it to have a persistent memory so it remembers our previous conversations and any specific instructions I've given it. It'd be great if it could handle complex problems by breaking them down and having multiple "sub-agents" work together on the solution. The interface should be a polished, chat-style UI right in the terminal, all controlled by a simple command.