geniml
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS] (LOW): References external code and models from trusted sources (GitHub: facebookresearch/StarSpace, Hugging Face: databio). Under [TRUST-SCOPE-RULE], these references are downgraded to LOW severity.
- [PROMPT_INJECTION] (LOW): The skill exposes an indirect prompt injection surface as it is designed to ingest and process untrusted genomic BED files and metadata CSVs.
- Ingestion points: Preprocessing and training steps in bedspace.md and scembed.md.
- Boundary markers: Documentation does not mention the use of delimiters or 'ignore' instructions for processed data.
- Capability inventory: Shell command execution via CLI and file system writes.
- Sanitization: No data sanitization or validation methods are described in the provided documentation.
Audit Metadata