bio-single-cell-markers-annotation

Pass

Audited by Gen Agent Trust Hub on Aug 14, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the ingestion of external data files, such as .h5ad and .rds objects.
  • Ingestion points: Biological data is loaded using sc.read_h5ad() in examples/find_markers_scanpy.py and readRDS() in examples/find_markers_seurat.R.
  • Boundary markers: There are no boundary markers or explicit instructions to the agent to disregard potential instructions embedded in the metadata (e.g., cell labels, gene names) of the processed data.
  • Capability inventory: The skill allows the agent to read and write files to the local file system (adata.write_h5ad, saveRDS, write.csv).
  • Sanitization: No sanitization or validation of data strings is performed before processing or including them in the analysis context.
  • [EXTERNAL_DOWNLOADS]: The skill recommends installing software directly from a GitHub repository.
  • Evidence: The documentation in SKILL.md and usage-guide.md suggests using remotes::install_github('immunogenomics/presto') to optimize Wilcoxon test performance.
  • Risk: While common in bioinformatics research, installing code from GitHub repositories bypasses the vetting processes of official package registries such as CRAN or PyPI.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 14, 2026, 09:20 AM
Security Audit — agent-trust-hub — bio-single-cell-markers-annotation