I audited unofficial Openmetadata CLI from romamo/openmetadata-cli against 22 Critical CLI Agent Spec failure modes. Even though it was built during a hackathon, it has the best score among all the other tested tools. Result: 1.5/3 average. Readiness: 12/15 [B]. This CLI already has a serious agent-first foundation: JSON envelopes on normal commands, schema introspection, dry-run mutation previews, MCP mode, bundled agent skills, non-TTY SSO protection, and prompt-injection tagging for external data. The remaining gaps are mostly contract gaps, not conceptual gaps. The two failing Critical checks were: §43 output size: no max-output flag, truncation metadata, or schema-declared output cap. §74 credential scopes: schema does not declare required_scopes and there is no check-permissions preflight. The partial failures matter too: timeout maps to GENERAL_ERROR, credential expiry maps to AUTH_REQUIRED, invocation errors can bypass JSON, and dry-run output lacks affected scope/effect fields. Bottom line: omd is close to being highly agent-ready. The next step is making every failure and credential boundary machine-readable. Full report: cli-agent-spec.github.io/eva… @open_metadata