pull down to refresh

ohh and not to mention the vice of writing comments in a magical-mystical way often also referencing private/local files

Paste this at the end of every prompt you send out (when you're satisfied, put it in CLAUDE.md/AGENTS.md/rules/whatever you use):

**IMPORTANT:** Keep paragraphs to 2–3 sentences max, separated by blank lines; use headings, short bullets, and bold key terms so the answer is skimmable. Prefer several small paragraphs over one long block — never write a wall of text.

**IMPORTANT:** Write code comments that explain the code itself; never reference plan files, task docs, chat history, or anything outside the repo (no "as per plan.md", "per the spec above", "step 3 of the plan").
reply
126 sats \ 1 reply \ @sox 18 Aug

I'll try this out in a few minutes, thank you! I've been doing this manually, but not with these words, and it's always a lottery.

Maybe I just got spoiled by the way LLMs used to talk i.e. with lists and spacing. People criticized it, me too, but it was actually comprehensible.

reply

This should make Claude sound at least a little bit like early GPT - haha

reply