Rather than adding an agents.md file written to stop LLM agents, I’m thinking about putting random comments into my code telling the agent to stop and warn the user, also some comments that are just word salads/non-sequiturs.