This looks good and will be very helpful when I have a slightly more scientifically useful dataset ready!
For the dataset contents, I suggest using the newest Gaia DR2 refcat, i.e., gaia_dr2_20200414
For the dataset contents, please use the non-symlinked path for the skymap, i.e., /datasets/hsc/repo/rerun/
I don't love the dataset name, but I do think communicating it is CI, COSMOS, and PDR2 are all important, so I don't have a better suggestion!
Deferring the handling of curated calibs beyond defects to DM-24402 is fine.
It's unclear to me if it's best to wait for the pex_exceptions ticket to be closed or to merge the workaround one-liner in ap_pipe; I leave this up to you.
Once this is all rebased and you are about to merge it, please do run ap_verify one final time with the new dataset as well as the HiTS CI dataset to make sure everything works as intended.