dbt-engineering
Pass
Audited by Gen Agent Trust Hub on Jun 21, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: The skill is configured to automatically load and process local project files such as .sql models, schema.yml, and dbt_project.yml, which presents a surface for indirect prompt injection if these files are malicious.
- Ingestion points: Files in the models/ directory, schema.yml, and dbt_project.yml as specified in the skill description.
- Boundary markers: Absent; there are no specified delimiters to isolate these files from the agent's core instructions.
- Capability inventory: Command execution via uv run dbt and cloud storage operations via gcloud storage.
- Sanitization: Absent; the instructions do not specify validation or filtering of ingested SQL or YAML content.
- [COMMAND_EXECUTION]: The skill documentation includes examples of shell commands for local development and CI/CD automation using project-specific tooling.
- Evidence: The Slim CI section explicitly provides uv run dbt build and gcloud storage cp commands as part of the operational workflow.
- [EXTERNAL_DOWNLOADS]: The project configuration references standard community packages from well-known sources in the dbt ecosystem.
- Evidence: The packages.yml section includes dbt-labs/dbt_utils and calogica/dbt_expectations, which are official or widely adopted packages in the community.
- [DATA_EXFILTRATION]: The skill demonstrates how to synchronize dbt state files with external cloud storage infrastructure.
- Evidence: The Slim CI section uses gcloud storage cp to download and upload manifest.json files to a Google Cloud Storage bucket (gs://BUCKET).
Audit Metadata