i-have-adhd: a skill that stops coding agents from burying the answer
MIT-licensed, 10 output rules, one command to install across Claude Code, Cursor and Codex — plus Copperhead designs real PCBs from a prompt.
Copy markdown
One command, and your agent leads with the answer
The skill loads 10 output rules — lead with the next action, number multi-step tasks, cap lists at 5 items, and cut every preamble, recap and closer — so the fix lands on line one instead of after three paragraphs of context. In Claude Code: `claude plugin marketplace add ayghri/i-have-adhd`, then `claude plugin install i-have-adhd@i-have-adhd`, and type `/i-have-adhd`.
One skill, six agents — Cursor, Codex, Gemini, Kimi, Qwen
Cursor installs with `npx skills add ayghri/i-have-adhd -a cursor -y`; Codex adds it via marketplace and triggers with `$i-have-adhd`. The rules live in a plain-text, forkable SKILL.md — no code — which is how it pulled ~25.6k stars and 1.6k forks within a day of hitting the Hacker News front page (368 points, 275 comments).
Copperhead turns a prompt into a real circuit board
Also trending on HN: an open-source (Apache 2.0) agent that designs, documents and validates actual PCBs from a prompt — it edits KiCad files, runs eight verification stages, and leans on Claude to keep docs and schematics from drifting apart. `npm i -g copperhead` (Node 20+, kicad-cli, an API key) and you can spec a board this weekend; it is already validated on real ESP32-S3 hardware.
Elsewhere: watch your model think, in the browser
A free, open-source LLM attention visualizer (published Sept 7) runs a 600M model fully in-browser via Transformers.js — hover any generated token to see which earlier tokens it pulled from. A fast way to build intuition for why a model copied a variable name, summarized cleanly, or drifted.