Notes on LLMs

The owloops/claude-powerline status bar plugin for code added a nice new setting today. In your ~/.claude/claude-powerline.json set the .display.lines[0].segments.context.displayStyle to "bar", as so: { "display": { "lines": [ { "segments": { "context": { "displayStyle":"bar", "enabled": true, "showPercentageOnly": true }, } } ] }, } This gives you a context segment that’s simple and easy to read at a glance: Here’s my full updated config that generates that style. (theme is set to light because that works best across light and dark terminal themes, I’ve found.) ...

Tip Turn on code’s verbose output mode. Unfortunately, it’s a not something you can set in ~/.claude/settings.json. But you can toggle it by running: ...

From Claude Code’s CHANGELOG.md for 2.1.3: Merged slash commands and skills, simplifying the mental model with no change in behavior. This is great news. The division between skills and slash commands was arbitrary and confusing. Skills now appear as slash commands. For example, I have a writing-bash-scripts skill that now is available as /writing-bash-scripts. When I run it, it loads the skill into context and is ready to apply it as needed. ...

In my description of my current working practices, I forgot the most important point: Tip ...

Written for a friend who’s getting started with the LLMs, Claude Code specifically. In no particular order. Superstition Everyone’s got their own superstitions / rituals / incantations / whatever related to using the magic robots. I’m no exception. My approach is to talk to Claude the way I want Claude to talk to me. Intelligently, articulately, precisely, clearly. I use full sentences and proper grammar. Even periods most of the time! I don’t swear, I don’t complain, I don’t make jokes, I just tell it what I want. And I never get any weird behavior back, none of the crazy stuff that so many other people seem to run into. ...