sql-insight
Pass
Audited by Gen Agent Trust Hub on Nov 22, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes SQL queries provided by the agent or user using the
scripts/sql_query_helper.pyscript. While the script implements a_validate_readonlyfunction to block common write-operation keywords (e.g., INSERT, DELETE, DROP), the execution of arbitrary SQL statements on connected databases is a high-privilege operation. - [DATA_EXF_EXPOSURE]: The
schemacommand extracts database structure and sample row data, which is then displayed to the user or passed into the agent's context. This potentially exposes sensitive data stored in the database if the tool is used on production or private data sources. - [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from the database to use as agent context.
- Ingestion points:
scripts/sql_query_helper.pyreads table names, column types, and sample data from user-specified databases (SQLite or PostgreSQL). - Boundary markers: The output lacks specific boundary markers or instructions to the agent to disregard potential commands embedded within the retrieved data.
- Capability inventory: The agent can run arbitrary SELECT and EXPLAIN queries and read from the local file system (SQLite files).
- Sanitization: The skill validates the read-only status of queries but does not sanitize the resulting data for natural language instructions that could manipulate the agent's behavior.
Audit Metadata