Stefan Van Dongen writes,

I am currently struggling (as a non-mathematician) to find a way to get appropriate priors for a multivariate normal with zero means, and SDs equal to 1 such that the off diagonals are correlations. I have tried the bugs list but got no solution.

My reply:

My favorite model is from O’Malley and Zaslavsky (see here and here). This does not actually restrict the diagonals to 1, but maybe you don’t really need to do this if you parameterize your problem carefully.