What problem does it solve? When a conductor has already chosen an exact OCX model route, dispatching it incorrectly is easy: generated agent types ignore placeholder model arguments, namespaced provider IDs can silently fall through to the default provider and get billed to the wrong upstream, and a healthy response is not proof the requested model actually ran. This Skill enforces a verify-then-admit discipline so no unverified dispatch enters the production handoff. ## Core Features & Use Cases - Exact route injection: Selects the correct injection surface for generated ocx-* agent types, Workflow agent() calls, and namespaced provider IDs, refusing placeholder or prompt-prose model claims. - Pre-dispatch catalog verification: Checks a namespaced ID's provider prefix against the live gateway catalog (ccodex models) before dispatch, since the gateway fails open onto the default provider rather than failing closed. - Post-run identity verification: Correlates requests with gateway attribution (ocx observe logs --jsonl) via requestId, treating routeKind: "default-provider" as an alarm and refusing to accept unverified receipts. - Use Case: A conductor resolves muse/muse-spark-1.2-contributor for a worker. Before spawning, the Skill confirms the muse prefix is live in the gateway catalog, dispatches with the exact ID and effort, then reads attribution to confirm the observed provider and model match the assignment. ## Quick Start Load the model-tier-rightsizing skill first to resolve an exact route, then ask the agent to dispatch that resolved assignment to a worker and verify the run against gateway attribution before accepting the result.
