[survirtual]

This is potentially exactly what I was looking for, thank you for putting it together.

When running a bunch of parallel agents locally, they can step on each other's shoes a bit. The ideal setup is to give them isolated workspaces, have them pull code in, do work, then push code back upstream.

When they do work, they sometimes go off the rails. They'll delete files they don't understand or think are irrelevant, explore other parts of the FS and get confused once their context is contaminated. By giving them a sterile workspace, it allows near risk-free multi-agent operations.

Containers offer most of this, but I was concerned about the security boundaries if it really goes haywire. For example, if I have an agent working at a very low level, it might start messing with the OS in a way that can damage things in a difficult way to reverse. They get confused easily.

Anyway, bookmarked. I will check it out in more detail when I get to that portion of my workflow. Thanks again.

[binsquare]

Heck yea, I'm trying to do something similar to manage running bunch of parallel agents locally.

Let me know if you run into any issues