I realize that I haven't documented my Colemak[eD] solution for Romanian, sorry.

Here's what I did (xkb implementation, abbreviated):

    key <AD11> { [        abreve,        Abreve,         bracketleft,           braceleft ] };// from 'a'
    key <AD12> { [   icircumflex,   Icircumflex,        bracketright,          braceright ] };// from 'i'
    key <LSGT> { [   acircumflex,   Acircumflex,                  oe,                  OE ] };// from 'q'
    key <AC03> { [             s,             S,           0x1000219,           0x1000218 ] };// comma-s from 's'
    key <AC04> { [             t,             T,           0x100021b,           0x100021a ] };// comma-t from 't'
    key <AC05> { [             d,             D,             dstroke,             Dstroke ] };// from 'd'

I have ș and ț on dead keys but the compose method uses comma for the cedilla so I chose to overload the ogonek dead key instead. That works in my Windows PKL files, but it's not a very convenient solution if you use these letters regularly! For someone only working with some Romanian people now and then it should be good enough. So, for Ș I'll type 'AltGr+8 S' for instance.

In the PKL layout.ini file, it's done like this (under ogonek dead_key):

83   =  536    ; S -> Ș
115  =  537    ; s -> ș
84   =  538    ; T -> Ț
116  =  539    ; t -> ț

I've noted that the cedilla letters ş and ţ are sometimes used. As commabelow is in practice used for only two letters it seems wasteful but who am I to argue with the Romanians about that... ;) (And yes, I'm also willing to change my own letter Ø to, say, Ō if it can make typing more logical.)