Show HN: Recreated Claude Code's behavior in 280 lines of Python

Now this is what an AI-related Show HN should be. You demonstrate something about how the technology works; you provide something usable; and you aren't trying to sell anything. Kudos.

I don't think the "straightforward architecture" is all that surprising, though. Once you've identified an entire LLM as a single "moving part", and have the idea of allowing its output to be interpreted as commands (and have wrappers available to connect to the LLM's API), the rest pretty much writes itself.