A Small Skill Called i-have-adhd Went Viral for Making Coding Agents Get to the Point
A free skill file that enforces ten rules on coding assistants topped Hacker News. It exists because frontier models keep burying the actual fix under three paragraphs of throat clearing, and the fix works for everyone.
A developer working under the handle ayghri published a small text file on GitHub called i-have-adhd, and it climbed to the front page of Hacker News with over 300 points in half a day. It does one thing: it forces a coding assistant to lead with the answer.
The file is a skill, which is a set of instructions an AI assistant loads and follows for the rest of a conversation. This one enforces roughly ten rules. Put the next action first. Number multi-step work. Restate where you are across turns so the reader does not have to scroll back. Suppress tangents. Give specific time estimates rather than vague ones. Make finished steps visible. In practice it rewrites the kind of answer everyone recognises, something like “you might want to look at your dependency versions, there could be a mismatch there, let me know if you want to dig deeper,” into a command: run npm install jsonwebtoken@latest, then edit src/auth.ts line 42. It installs across Claude, Cursor and Gemini through a short setup file, and you turn it on with a slash command and off by saying stop adhd mode.
What is behind this
The name is honest about who it was written for, and it works because the constraint is a good one for everyone. Models hedge for a reason. They are trained on text where hedging reads as careful, and reinforced by human raters who tend to prefer answers that sound thorough. The result is an assistant that knows the fix and spends four sentences approaching it sideways.
There is a second reason this resonated. As assistants move from chat windows into agents that act, the cost of a buried answer goes up. If you are skimming an agent’s output to decide whether to approve a change, a wall of qualified prose is not caution, it is friction. The traction of a 300 line text file also says something about where the real leverage sits right now: not in bigger models, but in telling the model you already have how to behave.
What this means for you: You do not need this specific file to get the benefit. If your assistant waffles, tell it so, in the same words you would use with a colleague: answer first, then the reasoning, no preamble, name the file and line. Most tools let you save that as a standing instruction, a custom instruction or a project rule, so you say it once rather than every time. If you do use Claude, Cursor or Gemini for code, the repo is free and takes a minute to install. And if the output of an AI tool has been annoying you in a specific way, this is your reminder that the fix is usually a sentence, not a different model.
Sources
A Diffusion Model Now Writes 1,100 Words a Second, and It Costs Almost Nothing
Inception Labs released Mercury 2.5, a language model built the way image generators are built rather than the way ChatGPT is. It produces 1,107 tokens per second on ordinary Nvidia cards at 20 cents per million input tokens.
