dataset-evaluation
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes a local script
scripts/format_detector.pyto validate dataset file formatting. This execution is a functional component of the skill intended to ensure training and evaluation data meets SageMaker requirements.- [INDIRECT_PROMPT_INJECTION]: The skill processes external data from datasets, which constitutes an indirect prompt injection surface. The risk is mitigated as the tool performs structural and schema validation (e.g., checking for required keys like 'messages' or 'query') rather than interpreting the natural language content as instructions.\n - Ingestion points:
scripts/format_detector.pyreads user-specified local files and S3 objects resolved during the workflow.\n - Boundary markers: Not explicitly implemented in the agent's summary instructions.\n
- Capability inventory: The skill can execute the
format_detector.pyscript and read from S3 using theboto3library.\n - Sanitization: The script uses the standard
jsonlibrary for parsing and compares data against strictly defined schemas for Nova and Open Weights model fine-tuning.
Audit Metadata