geniml

Pass

Audited by Gen Agent Trust Hub on Sep 25, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the geniml package directly from the databio repository on GitHub and references the StarSpace tool from the facebookresearch organization as a necessary external dependency.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple sources, creating a potential vector for indirect prompt injection.
  • Ingestion points: Processes genomic BED files, metadata CSV files, and AnnData (.h5ad) files as seen in SKILL.md, references/bedspace.md, and references/scembed.md.
  • Boundary markers: Instructions do not include delimiters or specific guardrails to prevent the agent from following instructions that might be embedded in the genomic data or metadata labels.
  • Capability inventory: The skill performs file system operations (writing BED/Parquet files), shell command execution (uniwig, cat, geniml CLI), and network requests (fetching models from Hugging Face and BED files via BBClient).
  • Sanitization: There is no evidence of validation or sanitization for the content of the ingested genomic or metadata files.
  • [COMMAND_EXECUTION]: The skill documentation provides and expects the execution of several shell-based workflows, including package management with uv, file concatenation with cat, and the use of specialized bioinformatics CLI tools like uniwig and the geniml suite.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 25, 2026, 02:34 PM