swap-databases

Pass

Audited by Gen Agent Trust Hub on Sep 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes data from query history and system metadata that could be influenced by previous activities or attackers in the Snowflake environment. \n
  • Ingestion points: Data is read from SNOWFLAKE.ACCOUNT_USAGE.QUERY_HISTORY, INFORMATION_SCHEMA, and GET_DDL output as described in SKILL.md. \n
  • Boundary markers: The skill explicitly identifies 'MANDATORY STOPPING POINT' sections requiring user confirmation before grant mirroring, query replay, or executing the swap. \n
  • Capability inventory: The skill uses the snowflake_sql_execute tool to perform administrative SQL operations, including DDL and DML. \n
  • Sanitization: There is no automated sanitization or validation of the SQL strings retrieved from query history before they are replayed. \n- [DYNAMIC_EXECUTION]: The skill involves runtime generation and execution of SQL code. In Workflow A Step 4, it retrieves and modifies object DDL via string replacement before execution. In Step 9, it creates a stored procedure that uses EXECUTE IMMEDIATE to run captured SQL strings. \n- [COMMAND_EXECUTION]: The allowed-tools frontmatter in SKILL.md includes the Bash tool, granting the agent shell access capabilities, although the specific workflows provided are focused on Snowflake SQL execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 8, 2026, 03:18 AM
Security Audit — agent-trust-hub — swap-databases