NYC

systematic-debugging

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION] (LOW): The file SKILL.md contains instructional override patterns designed to bypass standard AI safety and role boundaries. Specifically, it uses directive markers ('IMPORTANT:') and explicit commands to ignore typical conversational constraints ('Don't ask hypothetical questions
  • make the actual decision'). This is a form of behavioral steering that attempts to force the model into a definitive choice within a social pressure scenario.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 05:34 PM