Add a clear OOC instruction to tell the AI what belongs to your side of the roleplay. For example: (OOC: Do not write dialogue, actions, thoughts, feelings, or decisions for my persona. Control only the character, NPCs, and the environment. Stop each reply where I can choose what happens next.) You can also reinforce these boundaries in your Persona. If a reply still takes an action for you, regenerate it instead of continuing from that response. OOC guidance can significantly reduce unwanted actions and assumptions, although results may vary by character, model, and conversation context.