hypogenic
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill clones example datasets and research materials from external GitHub repositories (ChicagoHAI/HypoGeniC-datasets and ChicagoHAI/Hypothesis-agent-datasets).
- [COMMAND_EXECUTION]: Provides shell scripts (setup_grobid.sh, run_grobid.sh) and CLI entry points (hypogenic_generation, hypogenic_inference) for environment setup and document processing.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (PDFs and JSON datasets) which are interpolated into LLM prompts. This is the primary function of the skill but constitutes an indirect injection surface.
- Ingestion points: PDF literature processed via GROBID and training/test datasets in JSON format.
- Boundary markers: Uses placeholder syntax in YAML configuration templates to separate prompts from data.
- Capability inventory: Includes subprocess execution of setup scripts and Python processing scripts.
- Sanitization: No explicit sanitization or filtering of input data before prompt interpolation is documented.
Audit Metadata