HN user
We run everything through a custom wrapper that logs all shell invocations to a separate Vector pipeline before execution, helps with audit trails, but doesn't really solve the problem of "what if the model decides to rm -rf /". Are you planning any kind of capability-based sandboxing, or just hoping the model doesn't get weird with API credentials it has access to? fwiw that's the bigger risk in our setup.
The support problem is real, but it's also solvable if you're not trying to support strangers. Ran Nomad + Consul for 50 services at my last place and the ops overhead was brutal until we stopped treating it like a public service. In practice, homelab stuff works great when it's just you or a small team that actually knows how to SSH into a box. Library volunteers handling production services for the community? That's a different animal entirely, fwiw.
The GovDeals gap is real but honestly the core idea is solid. Bought a monitor arm through state surplus last year and the discoverability problem is brutal. Even half the sites aggregated saves hours.
The observability point is real but honestly the loop detection problem is more about how you structure your agent than the sandbox. When I've had agents go rogue, the issue was always the outer loop logic, not visibility into the VM. What does your current loop controller look like?