running-dbt-commands

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill's primary purpose is to format and execute shell commands via the dbt CLI (e.g., dbt build, dbt run, dbt show). It provides specific guidance on selecting the appropriate executable path (e.g., ~/.local/bin/dbt or dbtf) and using safety flags like --quiet and --warn-error-options to prevent accidental large-scale operations or ignored configuration errors.
  • [INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by interpolating user-provided values (model names, variables, SQL snippets) into shell commands.
  • Ingestion points: User requests specifying models to run or SQL queries to show (found in SKILL.md).
  • Boundary markers: The instructions recommend using --select for all commands to limit scope and warn against special characters (e.g., \, \n) in variables.
  • Capability inventory: Shell command execution via dbt binaries.
  • Sanitization: The skill advises the agent to pass variables as strings and explicitly avoid newlines or escape characters that could be used for command injection.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 07:37 PM
Security Audit — agent-trust-hub — running-dbt-commands