Dear Christophe,
Thanks for reading and for your question.
In an extrinsic view (but see last paragraph below), a self-referential object X would be equipped with a series of maps to itself, the endomaps {a}, and its evolution to another self-referential state, Y, with endomaps {b}, would be carried out by some extrinsic map, say, f: X -> Y, wherein the composition f o a = b o f holds, for all a and b.
But for an intrinsic view, the {a} have a new meaning within the system: a map a_i operates with itself and this must result in a "change of state" within X. That is, the composition a_i o a_j = a_k (not necessarily i=j) results in a new state Y, with self-referential operators {b}, mixed from the original ones {a}, which I suppose to be countably infinite. This requires an "autonomy" for the evolution, which is fed by a combinatorial, potentially non-exhaustive, set of endomaps. This self-characterization acquires an internal meaning as "active", and externally as "autonomous".
On the other hand, the external map f has no meaning inside the self-referential object. So it is difficult (or impossible) to characterize it-- a different mathematics, perhaps?
Best,
Christine