Hi Fabien - that's an excellent question. There is an expectation is that the Sovrin Foundation will publish some standards on this. However, due to the open nature of Sovrin it is not going to be something which can be policed. And in any event you may have multiple claim definitions containing name and address for example.
A way to address this will be for a trust framework to create a claim definition, and then to provide an attribute or claim of membership to its member organisations. So you could have a global bank trust framework that standardises the way a bank creates a customer summary claim definition for example. Included in that claim would be the fact that it comes from a bank which is part of this trust framework.
This means that you could request from the customer a claim containing name and address and also the trust framework membership attribute. This would allow the agent to look up the relevant attribute correctly to create the proof, and avoid the agent accidentally pulling name and address attributes from a claim written by a sports club.
In any case, the relying party will always be able to look up the issuer of the proof and determine whether to trust the issuer or not.