bio-differential-splicing
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external genomic data (BAM, GTF, TPM, and junction files) using various bioinformatics command-line tools such as rMATS, MAJIQ, and leafcutter. This creates an inherent attack surface common to data-processing tools, where malformed input files could theoretically attempt to influence agent command generation or interpretation. However, this functionality is the primary purpose of the skill and follows standard scientific workflows. * Ingestion points: BAM files, GTF annotation files, Salmon/Kallisto TPM files, and junction files defined in SKILL.md and usage-guide.md. * Boundary markers: None explicitly mentioned; the skill relies on standard biological file format specifications. * Capability inventory: The skill utilizes shell command execution for scientific software (rmats.py, regtools, majiq, suppa.py, snakemake) and R/Python data manipulation as documented in SKILL.md and the examples/ directory. * Sanitization: Processing is delegated to established bioinformatics tool suites which implement their own parsers for genomic data formats.
- [EXTERNAL_DOWNLOADS]: The skill references and provides instructions for installing specialized bioinformatics software from established community registries (Bioconda, PyPI) and official academic repositories (e.g., GitHub repositories of researchers like David Knowles, and official tool websites like majiq.biociphers.org). These are standard and trusted sources within the bioinformatics research community.
- [COMMAND_EXECUTION]: The skill provides examples of executing shell commands to run genomic analysis pipelines, including starting a local visualization server via voila and running snakemake workflows with Singularity containerization. These commands are documented for their intended scientific use cases.
Audit Metadata