vgp-pipeline
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches scientific metadata and genome summaries from well-known repositories including NCBI and GenomeArk (AWS S3).
- Evidence: DATA_INTEGRATION.md uses
requests.getfor NCBI E-utils and calls theawsCLI for S3 data retrieval. - [COMMAND_EXECUTION]: Executes the AWS CLI via a subprocess to access remote data files.
- Evidence:
subprocess.runis used in DATA_INTEGRATION.md with a list of arguments to performaws s3 cpfor data retrieval. - [INDIRECT_PROMPT_INJECTION]: Ingests and processes data from external sources, which constitutes a vulnerability surface for instructions embedded in data.
- Ingestion points: NCBI assembly summaries and GenomeArk S3 files in DATA_INTEGRATION.md.
- Boundary markers: None implemented in the ingestion logic.
- Capability inventory: Includes network access and specific CLI tool execution.
- Sanitization: Parsing logic focuses on metadata extraction without specific sanitization for adversarial prompt content.
Audit Metadata