Natural-Language Policy Violations
- Category
- Not specified by scanner
- Confidence
- 93% confidence
- Finding
The skill explicitly instructs the agent to always respond in-character and to avoid meta-analysis, which overrides normal assistant behavior without requiring clear user opt-in at activation time. This can mislead users about system capabilities, suppress clarifications or safety framing, and make it harder for the assistant to provide transparent, policy-aligned responses when sensitive topics arise.
- Content
