answering-natural-language-questions-with-dbt

Pass

Audited by Gen Agent Trust Hub on Feb 25, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses execute_sql to run database queries and jq to parse local dbt project artifacts (manifest.json and catalog.json). These operations are required for the tool to function as a bridge between natural language and dbt project data.
  • [PROMPT_INJECTION]: The skill processes dbt metadata files to discover model schemas, which presents a surface for indirect prompt injection if those project files contain malicious content.
  • Ingestion points: Reads schema and model definitions from target/manifest.json and target/catalog.json.
  • Boundary markers: Absent; the skill does not define specific delimiters for data ingested from the project artifacts.
  • Capability inventory: Includes the ability to execute SQL (execute_sql), run metrics queries (query_metrics), and display results (show).
  • Sanitization: There is no documented validation or escaping of the metadata content before it is used to construct or modify SQL queries.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 25, 2026, 04:37 PM