Recently, I have been working extensively with LLM-powered coding assistants - Claude Code, Cursor, Windsurf - and I noticed a frustrating pattern. Every time these tools run a command for me, they're processing massive amounts of irrelevant log output. You know how it goes - the LLM runs npm run...