scienceclaw-prediction

Pass

Audited by Gen Agent Trust Hub on Apr 6, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill establishes guidelines for scientific property estimation and outcome forecasting without introducing any executable scripts, network requests, or sensitive file access.
  • [SAFE]: No patterns of prompt injection, obfuscation, or credential harvesting were identified within the instructions or metadata.
  • [SAFE]: The 'Zero-Hallucination Rule' is a positive safety feature that instructs the agent to rely exclusively on tool-provided data rather than its training data for factual claims.
  • [SAFE]: Mentioned computational tools (such as RDKit, scikit-learn, and Biopython) are referenced as suggested capabilities for the agent to integrate with, not as unauthorized remote downloads or runtime installations.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 6, 2026, 11:30 PM
Security Audit — agent-trust-hub — scienceclaw-prediction