alterlab-geniml

Pass

Audited by Gen Agent Trust Hub on Sep 25, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the geniml package and its dependencies (e.g., scanpy, hnswlib) via uv pip. It also references downloading pre-trained models from Hugging Face (databio organization) and the StarSpace tool from Facebook Research's GitHub repository. These sources are established platforms for machine learning and research software.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process genomic data formats such as BED, AnnData (H5AD), and CSV metadata. These ingestion points represent a surface for indirect prompt injection if an attacker were to embed malicious instructions within genomic metadata. However, the skill treats these files as structured data for mathematical operations and does not interpret their content as instructions. The presence of boundary markers is not explicitly mentioned, but the highly specialized nature of the data reduces the risk of accidental obedience to embedded text.
  • Ingestion points: BED files, AnnData files, and CSV metadata files (SKILL.md, references/bedspace.md).
  • Boundary markers: None explicitly defined for natural language parsing within data files.
  • Capability inventory: File read/write/edit and Bash command execution via geniml and uv (SKILL.md).
  • Sanitization: Standard handling of genomic formats via libraries like pandas, scanpy, and geniml.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 25, 2026, 02:36 PM