Not saying that it’s not a bad idea, but I’m curious if there is another problem occurring that you feel this is missing

the agent natively is not running any sort of webserver, and has no listening connections as it is outbound only

however, you can add a liveness probe that tests if the prefect agent is still responsive, with something like this:

livenessProbe:
            exec:
              command:
              - prefect
              - agent
              - '--help'
            initialDelaySeconds: 5
            periodSeconds: 5

Do pending jobs ever recover if the agent couldnt kick the job off

do they eventually get set to failed

or Pending prefect flow runs?

if it’s a kubernetes job that was started, taht wouldnt’ be a prefect issue at that point, it was handed off to execute to k8s and I would be looking at the logs there for why it’s stuck in pending (insufficient cpu/memory potentially, not enough nodes, etc.)

and it will stay in that state given this scenario

do you have logs for the agent?