The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill enables the agent to interact with external databases using MCP tools such as execute_sql. This establishes an attack surface where malicious data stored in a database could influence the agent's behavior if processed without adequate safeguards.
Ingestion points: Results from execute_sql, list_instances, and list_users tools defined in the MCP configuration.
Capability inventory: The skill provides extensive capabilities including arbitrary SQL execution, instance management, and user credential modification.
Boundary markers: The documentation does not specify the use of delimiters or 'ignore' instructions for the agent when processing data returned from the database.
Sanitization: No explicit validation or sanitization procedures are described for the data retrieved from external sources.
[COMMAND_EXECUTION]: Contains instructions for executing standard administrative shell commands using gcloud, psql, and the cloud-sql-proxy binary. These are standard tools for managing cloud database infrastructure.
[EXTERNAL_DOWNLOADS]: References official Google Cloud client libraries for Python, Node.js, Java, and Go. It also points to reputable third-party repositories for MCP extensions, such as those hosted by gemini-cli-extensions and mcp-toolbox.dev.

cloud-sql-basics