When your LLM output isn’t what you wanted, edit the original message to fix the prompt, then regenerate.

The old exchange gets replaced instead of added to the conversation history. This one habit drastically cuts token use and reduces context rot.

Fix the prompt. Don't bloat the chat.