bytedance/trae-agent ? reverse-engineered prompt

Build me a Python command line agent that can take plain English coding tasks and actually work on a codebase for me.

I want it to be able to fix bugs, add features, write tests, refactor files, run shell commands, and keep going in an interactive back and forth mode until the task is done. It should work with a few different AI providers, support a simple config file or environment variables for API keys, and let me choose things like the model, max steps, and working folder from the command line. I also want a way to save a full log of what the agent did so I can inspect or debug the run later.

Please make it feel like a real developer tool, with a clean CLI, sensible defaults, and clear output while it is working. If you need to check current docs for any API or package details, do that.