It works this way => everytime an agent stops or crashes the object defined in EXECUTE_ON_LOST, EXECUTE_ON_END etc. will be executed.
You can check - if you stop one agent - the notification should be started => check via statistics of this object or message window