A new interoperability paper for multi-agent systems, InterSAGE, treats agent discovery as a verification problem rather than a lookup problem. Most agent discovery today works like a phone book: an agent finds a name and an endpoint, then trusts whatever shows up. InterSAGE's design instead requires discovery results to arrive as DID-bound Verifiable Credential manifests that get checked before any interaction starts. The paper describes capability-aware discovery that turns skill and tool advertisements into DID-bound Verifiable Credential manifests, so discovery results are verified for issuer provenance, subject binding, permission alignment, and freshness before interaction begins. In practice, that means a directory entry isn't just a claim an agent makes about itself — it has to be signed by an issuer, tied to a specific decentralized identifier, scoped to specific permissions, and time-stamped, and all four of those get checked by the requesting agent before it hands over any task. The motivation is concrete: an agent that impersonates a supply-chain optimizer, escalates its capabilities beyond delegation scope, or repudiates a financial commitment can cause cascading damage that no transport-layer encryption or OAuth token can prevent. Discovery is the first place that kind of impersonation happens — before any auth handshake, an agent has already decided who to talk to based on what a registry told it. If that registry entry is unsigned or unverifiable, the rest of the security stack is reacting to a decision that was already compromised. This sits alongside a growing field of similar proposals. The paper notes a first wave of trust-layer work already underway, including Microsoft's AgentMesh, described in its own documentation as SSL for AI agents, plus a zero-trust architecture for the agentic web and an invocation-bound capability token scheme called AIP that fuses identity, attenuation, and provenance. InterSAGE positions itself as a fuller stack: identity persistence, discovery verification, trust negotiation, and monotonic capability attenuation combined into one suite rather than addressed piecemeal. The capability-aware discovery piece is the most immediately testable claim, since it doesn't require new consensus infrastructure — just a convention for what a discovery response has to contain and a verifier on the requesting side that actually checks it. Whether registries and agent frameworks adopt that convention, or whether discovery keeps functioning as an unverified phone book, is the open question the paper doesn't answer. It's a design proposal, not a deployed protocol yet, and there's no indication in the paper of production adoption.