omni-to-snowflake-semantic-view
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to construct and execute shell commands using identifiers provided by the user, such as topic names and view names, which are interpolated into CLI calls without explicit instruction to sanitize these inputs.
- Evidence: The skill uses
<topic_name>and<view_name>directly inomni models yaml-getshell commands in Steps 2b and 2d. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources (Omni YAML model files) and processes their content to generate and execute Snowflake SQL commands. This creates a vulnerability surface where malicious logic in a model file could influence the generated SQL executed in the warehouse.
- Ingestion points: Model and view YAML files fetched via
omni models yaml-get(SKILL.md, Step 2). - Boundary markers: None identified in the prompt templates or interpolation logic.
- Capability inventory: Shell command execution (
omni,snow,snowsql) and direct SQL execution (CALL SYSTEM$CREATE_SEMANTIC_VIEW_FROM_YAML). - Sanitization: The instructions do not specify validation or sanitization of the SQL expressions extracted from the YAML views.
- [DYNAMIC_EXECUTION]: The skill dynamically generates complex Snowflake SQL statements by translating expressions and metadata from Omni YAML files, then executes this code within the Snowflake environment.
- Evidence: Step 7 describes generating a
CALLstatement where the body is a dynamically built YAML string containing logic derived from external files. - [EXTERNAL_DOWNLOADS]: The skill recommends installing the
snowflake-connector-pythonpackage and uses vendor-specific CLI tools. - Evidence:
pip install snowflake-connector-pythonin the Runtime Environment Detection section. The package is a well-known library for a established service.
Audit Metadata