On Tue, 14 Apr 2026 15:54:45 -0400 Aaron Conole <[email protected]> wrote:
> I am not a prompt optimization engineer, but I've been told by one that > we should try to avoid 'do not' type orders in favor of 'do' orders. > That's maybe a good second pass optimization, because we have a mix. On > the plus side, there are also lots of examples of good behavior in here, > which I was also told is a good thing to include. A lot of the 'do not' orders in current prompt are to suppress false positives. Mostly AI edits after telling it to shut up about something.
