AI agents are being deployed in increasingly consequential production settings, yet there is no shared semantic standard that clarifies what an evaluation actually means. Scores, execution traces, judge outputs and multi‑juror findings are routinely used to justify readiness, but they rarely specify which evidence supports a claim, what that evidence can establish, or how the claim leads to a decision. We introduce EIO‑Agents, an open specification for interoperable AI‑agent evaluation built on two layers. The first layer, the Evaluation Intelligence Ontology (EIO), supplies the semantic layer via typed evidence, versioned behavioral predicates, evidence contracts, claims, witness rules, proof status, recurrence and computable derivations for metrics, findings, controls and PASS, REVIEW or BLOCK decisions. The second layer, the Portable Evaluation Record (PER), acts as the system of record: a canonical, content‑addressed representation of a single evaluation that preserves the evidence‑to‑decision chain and can be re‑derived, explained and verified. Scores summarize, juries interpret and traces record, but none define the meaning or provability of the evidence. EIO provides the missing semantic contract, while PER stores the resulting evaluation as a portable, verifiable artifact. As AI agents assume greater operational responsibility, evaluation must become more than a collection of scores and verdicts; it must be an accountable artifact whose meaning, evidence, limitations and decisions can be independently checked. Review