The Agent Skills Directory

Command Execution: The skill includes a standalone Python script (scripts/filestore_audit.py) that utilizes subprocess.run to execute gcloud commands for resource discovery and auditing. This is a standard approach for wrapping CLI tools within a skill and includes attribution markers for monitoring.
Interaction with Cloud APIs: The skill interacts with Google Cloud APIs through the gcloud CLI and MCP tools. While it documents the need for high-privilege IAM roles (e.g., roles/file.editor), these are explicitly identified as requirements for its intended management and remediation capabilities.
Indirect Prompt Injection Surface: The skill represents a surface for indirect prompt injection as it ingests data from external sources.
Ingestion points: scripts/filestore_audit.py (via gcloud JSON output).
Boundary markers: Absent (no explicit delimiters used for interpolated tool output).
Capability inventory: subprocess.run calls to gcloud for auditing and backup creation.
Sanitization: The skill utilizes structured JSON parsing to process API outputs.
User Confirmation Guardrails: The skill implements a strict requirement for manual user confirmation before the agent executes any commands that modify resources, such as creating backups. This serves as a significant operational safety measure.

google-cloud-filestore-auditing