I am familiar with this capability. I don't want it to run automatically in response to the alert; this would be inherently problematic. Because the system isn't intelligent you can end up with false-positives and you don't want to go executing scripts against a system that isn't really down; instead it's better to have our NOC techs do a sanity check to validate that the problem really does exist before executing the action.
The feature that I am requesting is twofold...
The core of it is having a set of pre-defined executable actions that can manually be ran on any node(s) at any time with the appropriate permissions.
The additional piece is linking those actions to alerts so that in response to the alert a NOC tech can manually run the action if they deem necessary right from within the alert; this was functionality we had with OpenView. This part would just be a cherry on top, the first piece is what I would really like to see.