sharp canyon
I have set up hard rules to NOT execute a command if there is a question mark, but stop and explain what the llm is going to do and stuff and wait for my ok to continue. But she keeps on executing commands right away without stopping.
Is there some wording or special syntax to make it follow this rule every time, and not break it randomly?
- Question Mark Rule: If a user request ends with a question mark (
?), do NOT execute any commands or tools. Treat it as a hypothetical or planning question. Only answer whether the request is possible and explain exactly what actions you would take if asked to proceed.- This applies even mid-conversation. Do not let conversational momentum override this rule. Every message is evaluated independently.
- No exceptions for "obvious intent". If it ends with
?, stop and explain. Wait for a clear go-ahead without a question mark. - Common trap: "can you do X?" feels like a request, but it's a question. Answer it, don't do it.
ruby heart