hypogenic-hypothesis-generation
Pass
Audited by Gen Agent Trust Hub on Mar 16, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill installs the
hypogenicPython package via pip and clones research datasets from the ChicagoHAI GitHub organization. - [COMMAND_EXECUTION]: The skill references the execution of setup and runner shell scripts (
setup_grobid.sh,run_grobid.sh) and utilizes CLI tools (hypogenic_generation,hypogenic_inference) for its core hypothesis processing pipeline. - [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its data ingestion points.
- Ingestion points: Reads dataset JSON files (train/val/test) and PDF literature files.
- Boundary markers: Prompt templates for data observation and inference lack delimiters or instructions to ignore embedded commands.
- Capability inventory: Includes execution of local scripts and filesystem writes.
- Sanitization: No sanitization or filtering is performed on external text features before they are interpolated into LLM prompts.
Audit Metadata