Comment by nine_k

I wonder what prevents a GH action from connecting to your VPN (Wireguard is fine) and post tons of diagnostics right onto your screen, and then, when something goes badly wrong, or when a certain point is reached, to just keep polling an HTTP endpoint for shell commands to execute.

I mean. I understand, it would time out eventually. But it may be enough time to interactively check a few things right inside the running task's process.

Of course this should only happen if the PR contains a file that says where to connect and when to stop for interactive input. You would only push such a file when an action is misbehaving, and you want to debug it.

I understand that it's a band-aid, but a band-aid is better than the nothing which is available right now.