The Agent Skills Directory

Infrastructure Management via Command Execution: The skill provides a detailed reference for using gcloud commands to provision and manage Cloud SQL instances, databases, and users. These patterns are consistent with the skill's primary purpose of database administration.
Official Library Dependencies: The documentation references the installation of official Google Cloud connectors and client libraries using standard package managers (pip, npm, Maven, and Go). These resources are well-known, vendor-provided tools designed to facilitate secure database connections.
Security and Identity Management Guidance: A significant portion of the skill is dedicated to security best practices. It explicitly documents the risks associated with broad IAM permissions, such as the setIamPolicy capability, and recommends using IAM database authentication to avoid the use of static credentials.
Indirect Prompt Injection Surface: The Model Context Protocol (MCP) reference in references/mcp-usage.md describes tools like execute_sql that allow an agent to interact with a database.
Ingestion points: Untrusted data could theoretically enter the agent context through SQL query results or be passed into the execute_sql tool parameters.
Boundary markers: The documentation suggests using a restricted readonly toolset endpoint to limit the agent's capabilities.
Capability inventory: Tools include instance management, user creation, and SQL execution.
Sanitization: The skill recommends following the principle of least privilege and using IAM roles to restrict what the agent can perform at the resource level.

cloud-sql-basics