bio-de-results
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructions and usage guide recommend installing and using standard bioinformatics libraries from Bioconductor and CRAN, such as DESeq2, edgeR, biomaRt, and openxlsx. These are well-known, trusted resources within the scientific computing community.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external biological data, including gene annotation files (e.g., CSV) and results from remote databases via the biomaRt package. While this creates a theoretical surface for indirect prompt injection, it is standard for bioinformatics workflows.
- Ingestion points: The skill reads local files using
read.csv()and fetches data from Ensembl viabiomaRt::getBM(). - Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore potentially malicious content within the biological data files.
- Capability inventory: The skill possesses capabilities to write files to the local system (
write.csv,saveWorkbook) and perform network requests to the Ensembl database. - Sanitization: The provided R code does not include explicit sanitization of the data being processed, relying on the structure of data frames and standard R library behavior.
Audit Metadata