omicverse-single-cell-cellvote-consensus
Pass
Audited by Gen Agent Trust Hub on Aug 6, 2026
Risk Level: SAFEPROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
- [SAFE]: No malicious patterns detected. The skill implements biological data analysis workflows using the OmicVerse library.
- [PROMPT_INJECTION]: The skill demonstrates an indirect prompt injection surface by interpolating cluster markers and cell labels into LLM prompts. This is a functional requirement for its arbitration logic.
- Ingestion points: Data enters the prompt context from
adata.obsandcluster_markers. - Boundary markers: No explicit boundary markers or instructions to ignore embedded content were observed in the provided snippets.
- Capability inventory: Includes network requests to OpenAI API for consensus voting and local file writing for PopV results (
save_path='tmp'). - Sanitization: No specific sanitization of input biological labels was identified.
- [REMOTE_CODE_EXECUTION]: The skill uses a monkey-patching pattern to dynamically assign the
get_cluster_celltypefunction. This allows users to provide a local implementation instead of the default remote LLM call, serving as a runtime configuration mechanism rather than a vulnerability.
Audit Metadata