| name | mlops-paas-deployment |
| description | Implement or review K3s GitOps resources, Kustomize overlays, Argo CD, Argo Events/Workflows, Kubeflow jobs, platform services, autoscaling, CI/CD promotion, rollback, and operational validation. Use for changes under k8s/ or cluster deployment workflows. |
MLOps PaaS deployment
Workflow
- Inspect the root Kustomization, relevant base/overlay, Argo application, Ansible add-on ownership, and CI/CD workflow.
- Read gitops-and-kustomize.md, then the affected workload/platform/execution reference.
- Identify whether GitOps, Ansible, Terraform, or a runtime workflow owns the resource.
- Render manifests before editing live resources; avoid duplicate operator ownership.
- Preserve namespace, service-account, security, and callback contracts.
- Record existing topology gaps rather than assuming every manifest is reconciled.
- Preserve user changes and avoid unrelated manifest formatting/refactors.
- Update this skill when deployment ownership or topology changes.
Load $mlops-paas-security for exposure, credentials, workload isolation, IAM, or webhooks; load $mlops-paas-architecture for cluster/bootstrap changes.
For cross-service deployment changes also load $mlops-paas-overview. Treat rendered manifests, workflows, tests, and live ownership configuration as more authoritative than README or architecture prose; reconcile conflicts explicitly. Never embed passwords, tokens, private keys, real environment values, or SSH credentials.
Validation
Render the affected Kustomization, run schema/dry-run checks where available, and inspect the final namespace, image, secret/config references, probes, and service selectors.