I also saw that, seems interesting. I also thought about qemu vm, but that probably wont work on simple VPSes.
I am at the moment using a wrapper script that sets up the container around the project folder. Works nice. The only issue is I have to separately connect to the projects, and the connection is not harness agnostic yet. Herdr seems an interesting solution for that.
Have you thought about using a deterministic orchestration framework where the framework makes sure the tests are passing, agents are not cheating, and executes the plan, development, review, fix etc loops deterministically?
