mlops

What problem does it solve? Machine learning operations span many disconnected tools and subdomains, making it hard to route a request to the right capability while enforcing accuracy, evidence traceability, and reproducibility. This Skill acts as a governed router and contract layer for MLOps tasks, rejecting out-of-scope requests with contextual error messages instead of fabricating results. ## Core Features & Use Cases - Subdomain Routing: Directs requests to the correct subdomain such as codex-llm-routing, ellipse-3d-anatomy-constrained, evaluation, inference, models, research, or training, with routing decisions traceable to the request description. - Contract Enforcement: Applies boundary definitions, IO contracts, and strategy genes (MLOP-001 to MLOP-007) covering input validation, output consistency, error context, and a no-arbitrary-code safety rule. - Golden Set Verification: Ships golden input/output/error test cases (routing success and rejection paths) as the single source of truth, requiring a weighted score of at least 0.80 with all critical checks passing. - Use Case: A user asks to route Codex CLI v0.139+ to a local LLM via a Responses API proxy; the Skill routes to codex-llm-routing with traceable evidence, while an unrelated request like writing a poem is rejected with context and recovery suggestions. ## Quick Start Ask the assistant to route your machine learning operations request, such as deploying a model or running an evaluation, to the appropriate mlops subdomain.