For UCSC data just follow the suggestions here:
Although to be honest it seems more common to just stick to one assembly for the duration of the work (yes that is anecdotal experience) but if you always reference the build you use throughout the data release and publication process I don't see this being a problem (of course YMMV if what you are working on varies greatly between builds).
Use LiftOver http://genome.ucsc.edu/cgi-bin/hgLiftOver
I also remember that I used the mapping information of dbsNP or UniSTS for both build to find how a genomic segment 'moved'.
Moreover, if you store this information in a DB, or in any file, always include a column for the build or else your data will be a mess when you later have a look at it.