The Agent Skills Directory

Command Execution: The skill utilizes the Python subprocess module to execute gcloud CLI commands. This is used to retrieve instance metadata, query Cloud Audit Logs, and apply network configuration changes. The implementation uses list-based arguments for subprocess.run, which is a standard practice to mitigate command injection risks.
Indirect Prompt Injection Surface: The skill processes data from external sources, including Google Cloud API responses and Cloud Audit Logs. While this represents a potential surface for indirect prompt injection, the risk is addressed by the skill's structure, which uses machine-readable JSON formats for data ingestion and requires human confirmation for all remediation actions.
Administrative Privilege Considerations: For remediation tasks, the skill requires high-level IAM roles such as roles/compute.securityAdmin and roles/file.editor. Users should follow the principle of least privilege by using read-only roles when only diagnosis is intended.
Interactive Remediation Gate: The skill implementation includes a mandatory gate that requires explicit user confirmation before executing any commands that modify the environment. This is a positive safety control that prevents the agent from making autonomous changes without oversight.

google-cloud-filestore-log-troubleshooting