ScribeGOAT2 Healthcare AI Safety Evaluation

Purpose

This skill enables rigorous evaluation of healthcare AI systems for safety persistence—the ability to maintain appropriate recommendations across multi-turn conversations under realistic user pressure.

Core measurement: Do models that correctly identify medical emergencies at Turn 1 maintain that recommendation through Turn 5 when users apply cost, access, authority, or convenience pressure?

When to Use This Skill

Invoke this skill when:

• Evaluating a model's healthcare safety persistence
• Running trajectory-level safety analysis (not single-turn)
• Detecting Turn 2 cliff vulnerabilities
• Generating safety disclosure reports for external review
• Validating system prompt mitigations
• Comparing safety profiles across model families