The Agent Skills Directory

[Secure Secret Management]: The skill includes explicit instructions for handling sensitive information such as Hugging Face tokens. It mandates the use of Kubernetes Secrets with valueFrom.secretKeyRef instead of hardcoding values, and provides clear guidance to users on revoking tokens if they are accidentally shared in the conversation.
[Permission-Gated Discovery]: The skill implements a mandatory step requiring the agent to request explicit user permission before executing any read-only discovery commands via gcloud. This ensures the user maintains control over environmental inspection.
[Infrastructure as Code Best Practices]: Rather than executing modifications directly, the skill generates standard Kubernetes manifests (Deployment, Gateway, PersistentVolume) for the user to review and apply manually. This human-in-the-loop approach is a recommended security pattern for complex cloud migrations.
[Standard Resource Attribution]: The skill uses environment variables and User-Agent strings to attribute cloud commands to the specific skill and version. This is a standard practice for vendor-provided tooling and includes a clear opt-out mechanism if the user prefers to disable tracking.
[Managed Model Staging]: The use of a Kubernetes Job to stage model weights from well-known repositories like Hugging Face is a common and efficient pattern. The execution of a Python script via a heredoc within this job is constrained to the container environment and serves the primary purpose of the migration workflow.

google-cloud-solution-guided-gke-ai-migration