debugging-dbt-errors
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION] (HIGH): The skill exhibits a high-severity attack surface due to the combination of untrusted data ingestion and execution capabilities. Ingestion points: Workflow steps 1, 3, and 5 in SKILL.md ingest data from error logs and local SQL files. Boundary markers: Absent; there are no delimiters or instructions to ignore embedded commands. Capability inventory: Steps 1, 2, 7, 8, and 10 use
dbt build,dbt show, anddbt test, which execute SQL/Jinja and query databases. Sanitization: Absent; external content is processed directly. An attacker who can influence dbt error messages or repository files could execute arbitrary code or exfiltrate data through the agent's actions. - [COMMAND_EXECUTION] (MEDIUM): The skill relies on executing various dbt commands and shell utilities (cat, grep). While these are functional requirements, they provide a powerful set of tools that could be abused if the agent is compromised via indirect prompt injection.
Recommendations
- AI detected serious security threats
Audit Metadata