ml-stack-deployment
Prepare an ML artifact for serving when a user asks to deploy, publish, expose an endpoint, launch a Space or local service, configure auth, scale, smoke-test, or roll back; separate release approval
SKILL.mdskills/ml-stack-deployment/SKILL.md
GitHubArtifact promotion and serving
Require an independent audit, passing evaluation thresholds, immutable model/data/environment/resource manifests, model card and provenance, target, auth policy, secret references, health probes, rollback version, and scaling/cost policy.
Workflow
- Reconcile artifact hashes and manifests with
ml-stack-auditand verify evaluation go/no-go. - Select local service, endpoint, or Space; define request schema, health and smoke probes, capacity, timeout, observability, and rollback.
- Review security, authentication, authorization, secret handling, PII, license, and public exposure.
- Obtain explicit release approval; perform no upload, publication, endpoint mutation, or repository change through this skill unless an equivalent approved runtime operation exists.
- Hand exact target and release manifest to
ml-stack-hubfor approved publication andml-stack-trackingfor post-release metrics.
Boundary and outputs
Never silently publish to Hub, Spaces, endpoints, or repositories. Abstain on missing hashes, audit, thresholds, model card, auth, rollback, or cost policy. Return a release checklist, target plan, probes, rollback command/intent, measured versus unknown values, and approval record. Next owner: ml-stack-hub for artifact publication or ml-stack-audit for a failed gate.
See artifact promotion, serving and Spaces, health and rollback, and publication policy.