Your best bet for this kind of situation is a Finate State Machine. You would proably want to use fuzzy logic variables in the machine, as well. You can have varying degrees of "truth" to several variables, and then switch states (pass, shoot, etc, etc, etc) depending on the relative state of the variables. I''m not a big hockey person, so I''m not sure what variables would be appropriate to quantify the situation.
You might also look into the A* algorithm (or it''s cousin the SMA*) for best-path determination.
Hope this helps.