The Agent Skills Directory

Secret Redaction and Data Protection: The skill includes instructions to redact sensitive information—such as API keys, service account JSONs, and .env files—before local data is shared with the operator service. This significantly reduces the potential for credential exposure.
Command Execution and User Oversight: The agent is instructed to use gcloud, kubectl, and Terraform to manage resources. To ensure security, the skill mandates that every command or deployment action must be reviewed and approved by the user before execution.
File System Safety: When generating or modifying local files, the skill specifically restricts file operations to relative paths within the workspace. This prevents the agent from attempting to write to sensitive system directories or using path traversal to access unauthorized areas.
Management of External Context: The skill processes local artifacts like logs and runbooks which could potentially contain untrusted instructions. It manages this indirect prompt injection surface by using structured headings for data ingestion and maintaining strict user oversight for all resulting tool invocations.

cloud-operator