bio-read-qc-umi-processing

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • Indirect Prompt Injection (HIGH):
  • Ingestion points: FASTQ and BAM genomic sequence files (processed by umi_tools, samtools, etc.).
  • Boundary markers: None. Data is processed directly from external sources without delimiters or instructions to ignore embedded content.
  • Capability inventory: Command-line execution with filesystem write permissions and subprocess management via umi_tools, STAR, and samtools.
  • Sanitization: None. The skill assumes the structural integrity of the biological data files.
  • Analysis: Per the capability-based risk model, any skill that ingests untrusted data and possesses write/execute capabilities is Tier: HIGH. While genomic data is specialized, the potential for malicious inputs to exploit the toolchain or influence agent behavior via tool outputs cannot be discounted.
  • Unverifiable Dependencies (MEDIUM): The skill relies on external software including umi-tools, samtools, STAR, fastp, and pandas. None of these dependencies are associated with the defined trusted repositories or organizations, and no version constraints are provided to ensure supply chain integrity.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 07:29 AM