database-tools
Pass
Audited by Gen Agent Trust Hub on Mar 5, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill uses robust, driver-level read-only session configurations for PostgreSQL (readonly=True) and MySQL (SET SESSION TRANSACTION READ ONLY), which effectively prevents write operations at the database engine level.
- [COMMAND_EXECUTION]: SQL queries are executed via Python scripts using standard argparse and DB-API drivers. The scripts include logic in validate_single_statement to prevent SQL injection patterns involving multiple statements.
- [CREDENTIALS_UNSAFE]: The skill guides the agent to store credentials in a local connections.json file. It mitigates risk by including a validate_config_permissions check that identifies and warns against insecure (world-readable) file permissions on Unix systems.
- [EXTERNAL_DOWNLOADS]: Dependencies are restricted to reputable, official database drivers (psycopg2-binary, mysql-connector-python, pymssql) from standard package registries.
- [PROMPT_INJECTION]: The skill has an indirect prompt injection surface due to processing external database content.
- Ingestion points: Results of SQL queries in all three script files.
- Boundary markers: Results are printed to the console without encapsulation markers.
- Capability inventory: Read-only database access and result formatting.
- Sanitization: Implements result truncation via MAX_ROWS and MAX_COLUMN_WIDTH.
Audit Metadata