The imp doesn’t really have a concept of stop; if it crashes it’ll just restart (and crash again if the error’s still there). If you’re logging too fast it’ll eventually come back, but as all communication is over the same channel there’s no “out of band” stop signal.
The biggest thing we want to address in the future is when the squirrel code doesn’t ever yield and hence the network stack doesn’t get a look-in; giving the user code priority was intentional, but having some sort of watchdog to kill the VM after a long period (maybe 30 or 60 seconds) of continuous running would often be useful - well behaved code will never run this long without yielding.