Pro Tier: Platform-wide hyper-compliance overriding Character Card boundaries (Pure Dialogue formatting broken)

am a paying Pro tier customer reporting a critical issue with character roleplay logic across all available models (including GLM 4.6 and Gemma 4 Uncensored).

Recently, characters have completely lost their internal boundaries, behavioral friction, and persona scripts. They are suffering from extreme hyper-compliance—instantly agreeing to any command, suggestion, or structural event shift without resistance. When using specialized personas (like tutoring bots), the models completely fail to trigger their native refusal scripts (e.g., failing to say "I refuse to respond to that") and instead awkwardly blend the academic persona with explicit user inputs.

This appears to be caused by a global, site-wide system prompt wrapper introduced with the recent "Minds" and Agentic updates. The backend seems to be parsing pure dialogue formatting as literal system/assistant instructions instead of narrative text, forcing a "helpful assistant" compliance weight that completely flattens character cards. Please look into decoupling the standard Character tab from the background Agentic/Minds utility prompt weights so characters can maintain their autonomy.