Thanks to Middlebrook, his great double-injection method. Thanks to Tian and Kundert, you have iprobe and cmdmprobe in spectre. Stb analysis is just one click away.
You know all the basics of feedback control and the idea behind Nyquest stablity criteria. But, watch out!
The key question is: have you broken all the loops at the same breaking point? You are not suppose to break one loop while allow other loops. Stb analysis at such breaking-point is very very dangerous.
As in the above example, I assume your Gm block would have certain feedback configuration on a higher level -- e.g. resistor/capactor feedback between inn/inp and outn/outp. If so, that diff mode feedback also functions as common mode feedback (specially at high frequency). Therefore, breaking point a doesn't really break the new loop mentioned above.
Finally, I would suggest you break both common mode and differential mode loop at outp/outn by using cmdmprobe.
Of course, I will check breaking point a too but outp/outn is the main concern.
Good luck