tao-data-io
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external annotation files (Parquet, JSON, CSV, JSONL) to determine which objects to download from S3 and where to store them locally.
- Ingestion points: The
references/selective_download.pyscript reads data from external files specified by the--annotationargument. - Boundary markers: There are no explicit delimiters or instructions to the agent to ignore potentially malicious content within the annotation data.
- Capability inventory: The skill performs network operations (S3 downloads via the
boto3library) and file system writes based on the contents of the annotation file. - Sanitization: The script uses
pandasfor structured parsing and converts inputs to strings, but it lacks robust validation to prevent path traversal or ensure that the referenced file paths are strictly confined to the intended staging directory.
Audit Metadata