omicverse-single-cell-annotation

Pass

Audited by Gen Agent Trust Hub on Jun 29, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill utilizes library functions like download_reference_pkl and download_scsa_db to retrieve external reference models and databases (e.g., CellTypist .pkl files). These downloads are performed through the OmicVerse library's intended mechanisms and target expected vendor-specific or community resources required for its primary functionality.
  • [PROMPT_INJECTION]: The skill facilitates a potential indirect prompt injection surface as it is designed to extract biological data (marker gene lists) from user-provided AnnData objects and process them through LLM backends (OpenAI, Qwen, Kimi) for cell-type annotation. Ingestion points: Clustered marker gene lists are extracted from the AnnData object and passed to gptcelltype or the gpt4celltype branch of Annotation.annotate. Boundary markers: No explicit delimiters or specific instructions for the LLM to ignore embedded data instructions were identified in the workflow documentation. Capability inventory: The skill uses environment variable configuration for API key management and performs network operations to communicate with LLM providers. Sanitization: There is no evidence of explicit sanitization or filtering of gene markers before their interpolation into LLM prompts. This surface is inherent to the skill's primary purpose of providing LLM-based data annotation.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 29, 2026, 12:31 PM
Security Audit — agent-trust-hub — omicverse-single-cell-annotation