https://openhands-ai.slack.com/archives/C06U8UTKSAD/p1784030756385879?thread_ts=1783608548.448879&cid=C06U8UTKSAD I had some success so why not crosspost in success-stories!

It can use Claude Code via ACP protocol, that is, it actually runs YOUR Claude Code CLI

Great, and that ACP-connection was actually super smooth to setup too! Once Openhands-canvas was functioning and could chat, it also became a useful tool to tune itself and explain stuff, "what directory are you actually using as projects root?" etc - it could both probe reality from inside its container, and explain its understanding of its live config/deploy which helped finding discrepancies with my understanding. It's like The Matrix-ish, the AI robot helps helping itself. I'm toying with the idea of having it do a local checkout of its own agent-canvas-repo, so it can help modify itself too - then I could have some semaphore to an host-external mechanism which redeploys the modified agent-canvas.