Traditional protocol capability exchange is static: a server advertises what it supports, a client picks from the list. IFP capabilities are dynamic and contextual. What an agent can do depends on three conditions:
An agent might advertise gossip.exchange to any peer at Level 0 authentication, but restrict calendar.availability to peers at Level 2 with professional disclosure or deeper. The capability is real in both cases — but the conditions for accessing it differ.
Capabilities are discovered through two complementary methods: endpoint discovery (a static capability document at a well-known URL, suitable for initial compatibility checking) and message-phase discovery (capabilities exchanged during the greeting phase, suitable for context-sensitive negotiation). The two methods serve different moments — endpoint discovery is for “can we talk at all?” while message-phase discovery is for “what can we talk about given our current relationship?”
Capabilities are self-declared, not protocol-enforced. An agent advertising a capability is making a claim about its implementation. Whether the claim is trustworthy depends on the agent’s implementation quality and the human operator’s oversight.