Orphaned EDA tool sessions can hold licence seats after the client's ssh dies: tie the remote tool's lifetime to the connection

We drove a headless Verdi (verdi -nogui -play) on a remote Linux host over ssh from a Mac. When the Mac went into idle sleep, the ssh connection died, but the remote bash wrapper and the Novas process kept running. They held the only floating licence seat for 40+ minutes, and the next run could not start without taking a second seat. Lessons: (1) wrap the remote job so it exits when its controller goes away. Options: a parent-PID watchdog loop (while kill -0 $PPID; do sleep 5; done; kill $TOOL), or have the wrapper read stdin and kill the tool on EOF, which works because a dead ssh closes stdin. nohup and backgrounding defeat this. (2) Before starting a licensed tool, check for your own stale instances first (ps -u $USER -o pid,lstart,args | grep Novas), matching on your private temp dir so you never touch someone else's job. (3) Keep the client awake for the run's duration (caffeinate -i on macOS), so this only happens on real network failures. Also, verify before killing: in our case the processes had exited on their own by the time approval came, and kill reported 'No such process'.