The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses gcloud and kubectl to interact with Google Cloud infrastructure. It adheres to a strict execution policy where discovery is limited to read-only operations (after obtaining user permission), and mutating operations require explicit user approval.
[EXTERNAL_DOWNLOADS]: Fetches architectural guidance and documentation from official Google Cloud domains (docs.cloud.google.com) and GitHub repositories (GoogleCloudPlatform). These are well-known, trusted sources.
[CREDENTIALS_SAFE]: The skill implements specific safety protocols for handling sensitive Hugging Face tokens (HF_TOKEN). It prevents token exposure by instructing users to create Kubernetes Secrets manually via CLI and using secure secretKeyRef environment bindings in manifests instead of hardcoding values.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from the user's infrastructure (e.g., Cloud Run configurations) to inform its design phase. It mitigates injection risks by using structured --format filters on CLI commands to limit data input and by maintaining an explicit state file (migration-state.md) to track progress.
[DYNAMIC_EXECUTION]: The skill generates a Kubernetes Job manifest that contains a Python script for downloading model weights. This script is restricted to its specific operational task (using huggingface_hub), executes within the user's isolated cluster environment, and uses a standard, stable base image.

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