The Agent Skills Directory

[SAFE]: The skill implements a detailed 'Agent Safety Model' that classifies commands into READ, WRITE, DESTRUCTIVE, EXPENSIVE, and SECURITY categories. It explicitly instructs the agent on when to proceed autonomously versus when to require user confirmation.
[SAFE]: It contains a 'FORBIDDEN' category that restricts the agent from performing high-risk actions such as deleting projects, creating service account keys, or including passwords in command-line arguments.
[SAFE]: The documentation promotes security best practices, such as using Service Account Impersonation and Workload Identity Federation instead of static, leak-prone JSON key files.
[SAFE]: It includes explicit cost warnings for expensive cloud resources (e.g., GKE clusters, Cloud SQL instances, and Load Balancers) to prevent accidental billable resource creation.
[SAFE]: The skill recommends using machine-parseable output (--format=json) and server-side filtering (--filter) which reduces the risk of parsing errors and minimizes data transfer.
[SAFE]: All external references are to official Google Cloud documentation, well-known package managers (Scoop), or standard CI/CD tools (GitHub Actions).

gcloud