bio-expression-matrix-metadata-joins
Pass
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: LOWDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION] (LOW): The skill utilizes the 'mygene' library which transmits gene identifiers to an external web service (mygene.info). While this is a standard community resource in bioinformatics, it constitutes data egress to a non-whitelisted domain.
- [PROMPT_INJECTION] (LOW): The skill ingests untrusted data from multiple CSV and TSV files, creating a surface for indirect prompt injection. 1. Ingestion points: 'sample_info.csv', 'counts.tsv', 'metadata.csv', 'clinical.csv', 'sequencing.csv', and 'qc.csv' via pandas.read_csv and R's read.delim. 2. Boundary markers: Absent. 3. Capability inventory: Local file system read/write (e.g., writing .h5ad files) and network access via the mygene library. 4. Sanitization: Absent; the skill relies on standard CSV/TSV parsing without data validation or instruction filtering.
Audit Metadata