LLM prefaces for less frustration

I use these short prefaces to reduce AI frustration:

• “readonly” — Obvs. Don’t do anything dangerous. I have to repeat this every few turns because they forget
• “note” — Just acknowledge information. Good for adding knowledge to the conversation. Avoids over-investigation and wasting tokens
• “y/n” — You can also use this as a preface too to avoid slop (https://www.crccheck.com/blog/llm-quips-for-less-slop/). Helps keeps agents on task to actually answer your question.