geniml

Warn

Audited by Gen Agent Trust Hub on Mar 31, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits a metadata mismatch where the YAML frontmatter author ('K-Dense Inc.') contradicts the system-provided author context ('LeonChaoX'), which could lead to confusion about the skill's origin.\n- [EXTERNAL_DOWNLOADS]: The skill facilitates the installation of dependencies from PyPI and suggests installing software directly from a third-party GitHub repository ('databio/geniml'). It also references tools hosted by Facebook Research.\n- [REMOTE_CODE_EXECUTION]: Installation instructions that utilize 'pip' or 'uv' with git URLs involve the execution of build and setup scripts from the remote repository during the installation process.\n- [COMMAND_EXECUTION]: The skill's primary functions and installation steps are driven by shell commands, including the 'geniml' CLI and package management tools.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted external genomic data and metadata.\n
  • Ingestion points: Genomic interval data (BED files) and experiment metadata (CSV files) are processed from local directories.\n
  • Boundary markers: The instructions do not define clear delimiters or provide 'ignore instructions' warnings for the data being processed.\n
  • Capability inventory: The skill allows the agent to execute arbitrary shell commands and perform file system write operations.\n
  • Sanitization: There is no evidence of input validation or content filtering for the genomic data files before they are processed.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 31, 2026, 08:21 AM
Security Audit — agent-trust-hub — geniml