deployment-automation
Pass
Audited by Gen Agent Trust Hub on Mar 8, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Indirect prompt injection vulnerability surface detected in automated evaluation and promotion workflows.\n
- Ingestion points: Evaluation data is ingested from Unity Catalog tables (e.g.,
gold.evaluation.agent_eval_dataset) inreferences/dataset-lineage.mdandreferences/deployment-job-patterns.md.\n - Boundary markers: No explicit boundary markers or "ignore instructions" wrappers are used when passing dataset content to the evaluator LLM via
mlflow.genai.evaluate.\n - Capability inventory: The skill has the capability to automatically promote models to production aliases (
mlflow.set_registered_model_alias) and deploy them to Model Serving endpoints (databricks.agents.deploy) inreferences/deployment-job-patterns.md.\n - Sanitization: No sanitization, filtering, or schema validation of the input dataset content is performed before processing.
Audit Metadata