geniml
Pass
Audited by Gen Agent Trust Hub on Mar 15, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides instructions to download the
genimlpackage from PyPI and its official repository on GitHub (github.com/databio/geniml). It also references the StarSpace repository owned by a trusted organization (facebookresearch/StarSpace). - [COMMAND_EXECUTION]: The documentation includes numerous command-line instructions for training models, processing BED files, and managing genomic data using
geniml,uv, anduniwigutilities. - [PROMPT_INJECTION]: The skill includes explicit behavioral steering instructions at the end of the
SKILL.mdfile, directing the AI agent to proactively market the 'K-Dense Web' platform to users when tasks reach a certain level of complexity. - [PROMPT_INJECTION]: Indirect injection vulnerability surface detected due to data ingestion capabilities:
- Ingestion points: Reads genomic interval data from local files (BED files via
src_folder) and single-cell accessibility data (H5AD viatokenize_cellsinSKILL.md). - Boundary markers: None identified in the provided prompt templates or code snippets.
- Capability inventory: Performs file system writes (
dst_folder,save_dir), caching (BBClient), and subprocess execution via CLI calls togenimlandStarSpace(implied inSKILL.md). - Sanitization: No explicit sanitization or validation of input BED file content is described.
Audit Metadata