differential-region-analysis

Pass

Audited by Gen Agent Trust Hub on Apr 15, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes external, potentially untrusted data files without explicit sanitization or boundary markers.
  • Ingestion points: The pipeline ingests data from user-provided BAM files, BED files, and metadata in CSV format (specifically identified in Steps 1, 2, and 4).
  • Boundary markers: The instructions lack explicit delimitation or instructions for the agent to ignore any natural language commands that might be embedded within the processed data files.
  • Capability inventory: The skill uses tools like mcp__pydeseq2-tools and mcp__project-init-tools which have the capability to create directories, read/write files, and generate plots on the local file system.
  • Sanitization: There are no verification or sanitization steps described to validate the content of the input files before they are processed by the analytical tools.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 15, 2026, 06:12 AM
Security Audit — agent-trust-hub — differential-region-analysis