tao-data-io
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The
selective_download.pyscript identifies a vulnerability surface by processing untrusted data from annotation files (Parquet, JSON, CSV). - Ingestion points: The script reads file paths from columns in external annotation files provided via the
--annotationargument. - Boundary markers: The script lacks validation to ensure that paths extracted from annotations remain within the designated destination directory.
- Capability inventory: The script has the capability to write files to the local filesystem using
boto3andpathlibbased on extracted data. - Sanitization: The path sanitization logic in
download_selectiveis insufficient. Using.lstrip("/")on user-provided relative paths does not filter out directory traversal sequences like.., potentially allowing an attacker-controlled annotation file to cause writes to arbitrary host locations.
Audit Metadata