Expanding on @VCG's comment, which pretty much answered the question:
You can model this by adding a simultanous first move (to sign or not to sign) for both players. If they both choose to sign the contract, a predesignated outcome $A$ is realized and they get the corresponding payoffs. If they do not both sign, the original game is played.
Whether they will sign the contract or not will depend on their beliefs about what payoff they could expect in equilibrium. If a player's expected payoff from the original game is larger than what they would get in $A$ they will not sign.
(There are similar considerations in Aumann's correlated equilibrium, though there is no enforcable contract there.)
An example for when you would not sign the enforcable contract:
Consider a game of chicken with the payoffs
Suppose players could sign a contract forcing them to chose the "Chicken" strategy. A player will not do so if for some reason she expects that the other player will play chicken anyway, allowing her to collect a payoff of 7. In fact, using forward induction not signing may serve as a signal: The only reason he would not sign the contract is because she expects to win a larger than 6 payoff. The only way she can do this is by playing "Dare". Thus if the other player was willing to sign the contract, she has effectively signaled that she intends to play "Dare" by not signing the contract. Unfortunately according to the logic of forward induction the other player will also not sign.