In my last post, I described a setup I had started using at work: give every AI-agent rule a stable ID, then ask the agent to disclose which rules were candidates and which ones actually shaped the re

Markdown supported · ⌘↵ to post
No responses yet.