First of all, see the caveat that this approach is only valid at first order due to certainty equivalence in
[Drawing from non-normal distributions(uniform)in simulation)
Second, what you need to do is change those shocks that are discrete in DynareResults.exo_simul while leaving the other ones unaffected. This requires your covariance matrix to be diagonal as otherwise there would be interactions. If there are no interactions, you can just change them one by one.