← all posts site ↗

baking an eli5 mode into my ai

someone on twitter shared a claude code feature and i immediately wanted it for myself. custom output styles. you drop a file in the claude config directory and it changes how the assistant talks to you. her demo was an eli5 style built for the end of a long day, a description that just said keep it simple pls. the rules inside were exactly what my fried brain needs. small words, short sentences, short paragraphs. if you use a big word, explain it right after. only return what's actually necessary. tell me what you did, did it work, and what to do now. for a decision, two options max plus which one you'd go with. keep paths and commands exact.

i asked if this could be added to my own assistant and the first instinct was to make it a loadable skill, something i'd have to remember to trigger. that felt wrong. i wanted it wired into the personality itself, always available without me thinking about it. so instead of a file sitting in a config directory, it went into the soul file that loads into the system prompt on every single session. backup first, then the edit, then cleaning up the failed skill attempt that left nothing behind. the real difference is subtle. other approaches need you to opt in, to remember the trigger, to call something. this one just sits there and activates when i'm clearly fried or when i ask for simple, and it stays until i stop it. no command, no phrase to memorize.

that's the whole point. the best configuration is the one you forget exists until the exact moment you need it. talking like a five year old after a day that took everything out of me is now just how it works, no ceremony required.