The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for utilizing Model Context Protocol (MCP) tools and a command-line fallback using the psql utility. Specifically, it guides the agent to generate temporary IAM authentication tokens via the aws CLI and execute read-only queries for plan analysis. These operations are standard for database administration and follow least-privilege principles by preferring read-only tools.
[INDIRECT_PROMPT_INJECTION]: The skill acknowledges an attack surface where it processes externally-sourced SQL data (e.g., migration scripts, user-pasted DDL). It proactively mitigates this risk by mandating the use of the dsql_lint tool for safety validation and the safe_query.build() utility for consistent parameterization and identifier quoting to prevent injection attacks.
[DATA_EXPOSURE_AND_EXFILTRATION]: Security guidelines are enforced throughout, including multi-tenant isolation requirements (mandatory tenant_id filters) and IAM-based role mapping (dsql:DbConnect). These practices ensure that sensitive data remains scoped to authorized tenants and roles.
[REMOTE_CODE_EXECUTION]: The skill references official AWS documentation and a known sample repository (github.com/aws-samples/aurora-dsql-samples) for guidance. No execution of untrusted or unverified remote scripts was found.

distributed postgres