This can be systemic of agents that are copies of each other - say if two systems were imaged with the same image, but the certificate stores weren't cleared first. When two identical certificate agents connect, the appliance assumes it was a missed disconnect and disconnects the old one. If it's actually two different systems, they'll happily kick each other offline all day long. Solution is to clear one agent's certificate store so it re-issues a new, clean store (and if you're imaging, to do this on the image before you deploy it).
Now, if that's NOT the case, and it's a single PC kicking itself offline, either somehow there's two agent processes running (having the same effect) or the network connection is unstable, causing it to partially disconnect on the agent side but not on the appliance side, then reconnect on the agent side.