bio-single-cell-cnv-inference

Warn

Audited by Gen Agent Trust Hub on Aug 14, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill instructs the user to install R packages from non-standard GitHub repositories, which constitutes unverifiable third-party code.\n
  • Evidence: usage-guide.md contains remotes::install_github('navinlabcode/copykat') and remotes::install_github('AntonioDeFalco/SCEVAN').\n- [COMMAND_EXECUTION]: The skill utilizes shell commands to execute R scripts that are not provided within the skill package, creating a reliance on external, unverified code.\n
  • Evidence: SKILL.md contains an example execution using Rscript pileup_and_phase.R with various file-based arguments.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes numerous external data files and includes capabilities for file system access and shell execution without providing sanitization or boundary markers.\n
  • Ingestion points: The skill reads counts.matrix, cell_annotations.txt, gene_ordering.txt, barcodes.tsv, genetic_map_hg38_withX.txt.gz, and genome1K.phase3.SNP.vcf.\n
  • Boundary markers: Absent; there are no instructions or delimiters to ensure the agent ignores potential malicious instructions embedded within these data files.\n
  • Capability inventory: The skill performs shell command execution (Rscript), file reading (read.table), and directory/file writing (out_dir parameters).\n
  • Sanitization: Absent; there is no evidence of validation or sanitization for the content of the ingested files before they are processed by analysis tools or shell commands.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 14, 2026, 09:19 AM
Security Audit — agent-trust-hub — bio-single-cell-cnv-inference