How to define custom distributions in JAGS?

Either the "zeroes trick" or the "ones trick" will do the job for you; see the WinBUGS documentation online for how to set these up. (WinBUGS is, as you almost certainly know, almost identical to JAGS syntatically, with a few exceptions noted in the JAGS documentation and irrelevant to the two tricks above.)

From the feedback on the newsgroup, it seems that JAGS is better suited for known distributions and composition of known distributions. Its faster / easier writing your own sampler in R for the rest.