Well, that would be a complete reworking of the system. Right now the values involved are plugged into the equations used by the AI for controlling its actions. What you are describing is more like a flag system.
However, that does give me an idea for how it could be done if we really wanted to...
We could have the new $maneuvering string to determine how effectively the AI executes the new maneuvers available to it, and also have optional strings added on after it saying things like "+barrellroll: no" if we wanted a specific AI not to execute a specific sort of move. Then perhaps we could code in an extra line onto new maneuvers telling it to make sure there isn't a "+barrellroll: no" string present for that AI class before going on to perform the action.
This would be more managable than making an entire new string for each maneuver, in that we only need to deal with one new set of numbers, and only those maneuvers which are specifically negated would not be used.
All that having been said, I don't understand
why one wouldn't want the Shivans to do barrel rolls.