bio-atac-seq-deep-learning-atac
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The script
examples/chrombpnet_pipeline.shperforms agit cloneof an external repository (https://github.com/kundajelab/variant-scorer) to obtain a necessary variant scoring utility. - [REMOTE_CODE_EXECUTION]: The example pipeline in
examples/chrombpnet_pipeline.shexecutes a Python script (variant_scoring.py) that was downloaded from the external repository during runtime. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external genomic data files (BAM, BED, and TSV formats), which serves as an attack surface for indirect prompt injection.
- Ingestion points: Files provided as input parameters to the
chrombpnetCLI and thevariant_scoring.pyscript inexamples/chrombpnet_pipeline.sh. - Boundary markers: No specific delimiters or safety instructions are defined for the agent when processing the contents of these data files.
- Capability inventory: The skill utilizes shell command execution, file system access, and network access (via
git). - Sanitization: The skill relies on standard bioinformatics parsing tools; there is no documented sanitization for natural language instructions that might be embedded in the metadata or data fields of processed files.
Audit Metadata