About 15 minutes ago, I believe I have successfully managed to get two Agent Device's to exchange information about the other, and in so doing, acquire information about each other that facilitates further communication.
I have called this the engage process. There is an opposite - the disengage process - where two Agent Devices formally break this relationship and mutually forget about their association with the other. The disengage process hasn't yet been implemented, but I'm pretty sure it is much easier to implement than the engage process.
While engaging and disengaging are a device-specific thing (specifically Agent Devices), this process uses the basic Clique Space data model, with perhaps some liberties. It's a fairly straightforward process; two Agent Devices create information about each other on their "local" Agent Device Clique Spaces, and then mutually exchange information each has created about the other along with information about themselves by creating "foreign" Agent Device Clique Space copies of the other's local Agent Device Clique Space. The process requires that one of the two Agent Devices be the "initiator" and the other be the "responder", but once the process is complete, it is completely symmetrical; one cannot determine which device was the initiator and which the responder simply by inspecting the relationship artefacts on each Agent Device.
So there you go. Now that the Agent Device engagement process has been (one hopes) stabilised into a fully functional realisation, I will begin concentrating on the Agent Collaboration.
