An agent is "reflectively stable" if, asked to construct a successor agent or modify its own code, it would construct a successor that thinks much the same way it does. For instance, we say that causal decision theory is not "reflectively stable" because causal decision theorists don't construct other causal decision theorists. Similarly, having a utility function that only weighs paperclips is "reflectively stable" because paperclip maximizers try to build other paperclip maximizers. If, thinking the way you currently do (in some regard), it seems optimal or acceptable to think that way (in that regard), then you are reflectively stable (in that regard).