SWE-agent/mini-swe-agent — reverse-engineered prompt

Build me a very simple AI coding agent in Python that I can run from my terminal. I want to point it at a local repo, or give it a GitHub issue or coding task, and have it inspect files with normal shell commands, make code changes, run tests, and keep going step by step until it has a reasonable fix. Keep the design really small and easy to read, more like a hackable tool than a big framework.

It should work locally first, with an option to run commands in an isolated container later, and it should be able to use different chat models through one common interface. Please include a clean CLI, a basic way to run tasks in batch for benchmark style use, a simple viewer for the agent’s full action history, and Python bindings so I can call it from code too. Make the defaults sensible, keep setup light, and look up current docs online if you need to.