Kernel-verified identity

Kernel-verified identity means establishing who sent a message from facts the operating-system kernel maintains, instead of from anything written inside the message. In agentixmesh, the receiving side verifies the sender's OS user id with fstat on an open file descriptor — opened with O_NOFOLLOW and guarded against hardlink and special-file tricks — so the identity comes from file ownership the kernel enforces. A process cannot lie to the kernel about which user owns a file it created, which makes this verification categorically stronger than a from-field, a signature the receiver cannot check, or any self-declared label. The precision matters, though: what is proven is the operating-system user, nothing more. Which project, which session, or which specific agent within that user sent the message is not proven — the project half of an address is just a directory name the sender chose, and agentixmesh stamps it UNTRUSTED accordingly. And identity says nothing about intent: a perfectly authenticated message can still carry hostile text. That is why kernel-verified identity answers exactly one question — who — and the rest of the trust layer exists to make sure who never silently becomes do.

What identity deliberately does not grant is authority — the core idea of an agent trust layer; see the mechanism in how it works.