JayTheCoder77/Zox ? reverse-engineered prompt

Build me a local coding agent app that runs on my machine and talks to an LLM with my own API key.

I want a terminal experience where I can ask it to work on a repo, let it read files, edit code, search, and run shell commands, but only after asking for permission when needed. It should keep each session saved, show what it did, and make it easy to resume or export a session later. The server should be the main part that actually runs the tools, and the terminal UI should just connect to it and show the conversation and tool activity.

Please include a safe default setup that uses a git worktree so it does not edit my current checkout directly, plus an option to use the host workspace if I want that. It should support a few model providers, work in a real terminal or a simpler line mode, and have basic hooks or commands for customizing behavior. If you need to check current docs online while building it, go ahead.