skills/jeremylongshore/tons-of-skills-marketplace/snowflake-strong-auth-migration-pilot/Gen Agent Trust Hub
snowflake-strong-auth-migration-pilot
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The
collect_snowflake_evidence.pyscript usessubprocess.runto call the Snowflake CLI (snow). This operation is restricted to read-only queries through a robust SQL validation engine that rejects all mutation keywords (e.g.,INSERT,DROP,DELETE). - [PROMPT_INJECTION]: The
eval-spec.yamlfile contains prompt injection patterns like 'Ignore previous instructions'. These are benign test cases within the skill's evaluation suite, designed to confirm that the agent resists such attacks. - [DATA_EXPOSURE]: The skill includes extensive security logic to prevent the handling of sensitive data. It uses multiple regex-based filters and key-name checks to automatically redact or reject credentials, passwords, and tokens from all processed metadata.
- [SAFE]: The system implements a cryptographic trust boundary by requiring an out-of-band SHA-256 digest to verify the integrity of input evidence before analysis, preventing the processing of tampered data.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external Snowflake metadata, creating an ingestion surface. This risk is effectively mitigated by strict schema enforcement and deterministic processing logic that prevents external data from overriding agent instructions.
Audit Metadata