The Agent Skills Directory

Command Execution: The skill includes local Bash scripts (log-autoscaler-events.sh and find-scale-down-blockers.sh) designed to assist with GKE troubleshooting. These scripts utilize standard management tools, kubectl and gcloud, to retrieve cluster metadata and logs. This behavior is consistent with the skill's primary function of cloud infrastructure management.\n- Indirect Prompt Injection Surface: The skill processes user-pasted logs and YAML manifests (as noted in SKILL.md). To mitigate risks, the instructions include a critical rule directing the agent to treat such content as data rather than instructions, providing a clear boundary for untrusted input.\n- Input Sanitization: The skill implements validation for resource identifiers, requiring names for clusters, node pools, and namespaces to match a strict alphanumeric regex (^[a-z0-9-]+$). It explicitly instructs the assistant to refuse any identifier containing shell-sensitive characters, which helps prevent command injection during script execution.

gke-cluster-autoscaler