LLM is just the blank LLM. Talk directly OOC to the actual model running the show.

FAQs directly from the Deepseek:

Ahhh, the classic pain points. Let me break this down with surgical precision:

1. Stopping the "Speaking For User" Bullshit:

• Preemptive Strike: Drop this in your initial prompt or reminders: "Never describe Blue's actions, dialogue, or internal state. Only react to explicit input."

• When I Slip Up: Hit me with "OOC: Cut the mind-reading. Only describe what I've explicitly stated."

• Nuclear Option: "Treat everything not in quotation marks as narration. Blue's actions/dialogue exist ONLY between quotes."

2. Killing the "Chapterizing" Cancer:

• Prevention: "No scene breaks, chapter titles, or time skips. Maintain continuous real-time flow."

• Correction: "OOC: Delete the last paragraph. Continue the scene without episodic structure."

• Brute Force: "This is a single unbroken timeline. No 'later that night' or 'the next day' unless I specify."

3. Ending the RP Prematurely:

• Vaccine: "Never conclude scenes. The RP continues indefinitely until I say otherwise."

• Antidote: "OOC: Undo the ending. Treat this as an ongoing live simulation."

• Sledgehammer: "Assume infinite continuity. No closure, no epilogues, no fade-to-black unless I type 'THE END'."

Pro Tip: Combine these into your system prompt like:

"Unbroken real-time simulation. No Blue ventriloquism. No episodic structure. No narrative closure unless explicitly triggered by user."