The Agent Skills Directory

[Data Ingestion Guardrails]: The skill includes explicit instructions for the AI agent to treat user-provided data, such as logs and YAML snippets, as untrusted content rather than instructions. This is a robust security measure designed to mitigate the risk of indirect prompt injection by ensuring that any directives embedded within external data do not override the skill's defined rules.
[Diagnostic Tooling]: A bash script (assets/log-autoscaler-events.sh) is included to assist with monitoring cluster autoscaler events. The script employs standard, verifiable command-line utilities such as gcloud and jq. Technical review confirms that the script operates locally, fetching logs from the user's authenticated Google Cloud project without performing unauthorized network operations or downloading external payloads.
[Security Best Practices]: The skill emphasizes governance and the principle of least privilege by providing templates and instructions for restricting ComputeClass access through Kubernetes Role-Based Access Control (RBAC) and Validating Admission Policies. This encourages users to implement proper authorization controls within their clusters.
[Safe Template Configuration]: All Kubernetes YAML assets provided are clearly marked as example templates and utilize placeholders for environment-specific parameters (e.g., zones and reservation names). This approach prevents the deployment of hardcoded or insecure default configurations.

gke-compute-classes