genomics-workflow-acceleration
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-provided genomics workflow files (Nextflow, Snakemake, WDL, and Python), which constitute an ingestion surface for untrusted data. The agent has capabilities to edit these files and execute validation commands (e.g.,
nvidia-smi,pbrun). This risk is effectively mitigated by the skill's mandatory 'default-off' toggle architecture, which prevents the execution of new code paths without explicit user configuration, and instructions requiring human-verified A/B comparisons between original and accelerated outputs. - [DYNAMIC_EXECUTION]: The skill performs dynamic code generation by modifying existing workflow files to insert GPU-accelerated steps. This behavior is the primary intended function of the skill and is constrained by strict mapping templates (e.g., mapping BWA-MEM to
fq2bam). The instructions follow development best practices by recommending the use of feature branches and preserving original logic alongside the new additions. - [EXTERNAL_DOWNLOADS]: The skill refers to official container images from NVIDIA's registry (
nvcr.io) and community modules from thenf-corerepository. These are well-known and reputable sources in the bioinformatics domain. While the internal evaluation report notes a potential supply chain risk if users do not pin container versions, the skill instructions explicitly advise users to pin image tags during implementation.
Audit Metadata