hypogenic-hypothesis-generation

Pass

Audited by Gen Agent Trust Hub on Apr 30, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted external data (tabular JSON and research PDFs) by interpolating them into LLM prompts (e.g., in observations and inference templates). This creates a surface for indirect prompt injection where malicious instructions embedded in the data could influence agent behavior.
  • Ingestion points: JSON datasets (train/val/test) and literature PDFs (via GROBID).
  • Boundary markers: Absent in prompt templates; data is directly interpolated into system and user prompts.
  • Capability inventory: Calls to task.generate_hypotheses, task.inference, and shell script execution for tool setup (setup_grobid.sh).
  • Sanitization: No explicit sanitization or validation of the content within the ingested data is documented.
  • [EXTERNAL_DOWNLOADS]: Fetches example datasets and literature resources from the Chicago Human+AI Lab's (ChicagoHAI) public GitHub repositories.
  • [COMMAND_EXECUTION]: Provides instructions for setting up and running the GROBID service and dataset preprocessing scripts to facilitate literature-integrated hypothesis generation.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 30, 2026, 04:05 AM
Security Audit — agent-trust-hub — hypogenic-hypothesis-generation