AI Doesn’t Work For ADHD Brains. One Skill Fixed It.

TL;DR: Virginia Tech researchers studied 61 neurodivergent communities (ADHD, autism, and others) on Reddit and found that AI chatbot answers are, in the study’s own words, “overly neurotypical.” Translation: built for everyone except the brains that need the point first. A developer named ayghri built a free, open source Claude Code skill called i-have-adhd that fixes this directly, it puts the answer first, numbers every step, and cuts the filler. It installed with two commands, passed 3,000+ GitHub stars within weeks, and went viral on X after one post thanking whoever shared it hit 27,700+ likes in a day. No diagnosis needed to use it, and no medical claims attached, it is a formatting fix that happens to help everyone.

The problem: AI answers bury the point

Ask Claude, ChatGPT, or any large language model a direct question and you often get a small essay back: context, caveats, a walkthrough, and then, somewhere near the bottom, the actual answer. For most people that is a minor annoyance. For a brain that struggles to hold a wall of text in working memory, it is the difference between using the tool and abandoning it.

That is not a hunch. It is a documented finding. A team of researchers at Virginia Tech, Carik, Ping, Ding, and Rho, published a study titled “Exploring Large Language Models Through a Neurodivergent Lens: Use, Challenges, Community-Driven Workarounds, and Concerns” (arXiv:2410.06336). Their method was a qualitative analysis of discussions across 61 neurodivergent communities on Reddit, spanning ADHD, autism, and other conditions, not just one diagnosis.

Two of their named findings are worth quoting directly, because they describe the exact failure mode a lot of AI users have felt but never had language for:

“Overly neurotypical LLM responses”

“The limitations of text-based interactions”

Plain-language translation: the models default to a communication style built around neurotypical norms, long-form, hedge-heavy, context-first, and that style actively works against people who process information differently. The study also documented that affected communities were already building their own manual workarounds, custom prompts, personal formatting rules, browser extensions, because nobody had shipped a real fix.

The fix: a Claude Code skill called i-have-adhd

That gap is exactly what a developer known as ayghri closed. Their repository, i-have-adhd, is a skill for Claude Code (and Codex) that changes how the AI structures its replies. The README states its purpose without hedging: “A Claude Code skill that stops burying the answer. Action first. Steps numbered. No ‘Hope this helps!’”

The repo is equally direct about who it is for: “ADHD-friendly outputs. No ADHD diagnosis needed!” That line matters. This is not a medical tool, it makes no claims about treating or diagnosing anything. It is an output-formatting change, and the developer built it to be useful to anyone, whether or not they identify with the condition in the name.

What actually changes

Before the skill, a typical Claude reply to a direct question reads like a short essay: background, reasoning, maybe a caveat or two, then the answer folded in near the end. After installing the skill, the same question gets a reply that leads with the action, breaks any multi-part process into numbered steps, and drops throwaway phrases like “hope this helps” that add words without adding information.

The change is not about making Claude smarter. It is about making Claude’s existing intelligence easier to actually use under time pressure or with limited working memory, which describes a much larger audience than any single diagnosis.

How to install it (two commands)

The entire setup is three commands, run once inside any project folder:

git clone https://github.com/ayghri/i-have-adhd ./i-have-adhd
claude plugin marketplace add ./i-have-adhd
claude plugin install i-have-adhd@i-have-adhd

Once installed, Claude Code picks it up automatically for that project. If you want to invoke it explicitly rather than rely on auto-detection, type /i-have-adhd before your question.

The same repository also works with Codex, where it is invoked as $i-have-adhd. Same underlying formatting logic, different terminal.

If you decide you don’t want it anymore, removing it is one command: claude plugin uninstall i-have-adhd@i-have-adhd.

The viral moment

The skill itself was released in mid-May 2026 and sat quietly at a few hundred stars for weeks, standard trajectory for a useful but unmarketed developer tool. That changed after a post on X from user @jjacky, thanking whoever had originally shared the skill, started circulating. That single post passed 27,700 likes within a day. In the days that followed, the repository’s star count climbed past 3,000.

The spike is a useful case study in its own right: the tool didn’t change, the awareness did. A real accessibility gap, once someone puts a name and a number on it, spreads fast because so many people recognize the problem the moment they see it named.

Is this only useful if you have ADHD?

No, and the repository is explicit about that. “No ADHD diagnosis needed” is not a disclaimer, it is the design intent. Answer-first, numbered, filler-free output is a better default for most people using AI to actually get something done rather than to explore a topic at leisure.

That said, it is a trade-off, not a strict upgrade. If you use Claude for open-ended research, brainstorming, or want it to reason out loud through ambiguity, the longer, more exploratory default format has real value. The skill is built for task-mode: “what do I do and in what order,” not “help me think this through.”

What this is not

To be clear about the boundaries here: this is not a medical device, it does not diagnose ADHD or any other condition, and it makes no claim to treat symptoms of anything. It is a formatting and interaction-design fix for a documented communication mismatch between how large language models default to writing and how a large share of users, neurodivergent and neurotypical alike, actually need information delivered. Nothing here should be read as health advice.

FAQ

Does this require an ADHD diagnosis to use? No. The developer’s own README says explicitly: “No ADHD diagnosis needed!” It is a free, open tool available to anyone using Claude Code or Codex.

What study is this based on? Carik, Ping, Ding, and Rho, “Exploring Large Language Models Through a Neurodivergent Lens: Use, Challenges, Community-Driven Workarounds, and Concerns,” Virginia Tech, arXiv:2410.06336. The method was a qualitative analysis of discussions from 61 neurodivergent communities on Reddit.

Does it work with tools other than Claude Code? Yes. The same repository supports Codex, invoked as $i-have-adhd instead of the Claude Code plugin command.

Is this an official Anthropic feature? No. It is an independent, community-built skill hosted on GitHub by developer ayghri, installed the same way any third-party Claude Code plugin is installed.

Will this make Claude worse at complex reasoning? No, it changes formatting and ordering, not the underlying model. For deep research or brainstorming tasks where you want Claude to reason out loud, you can simply not invoke the skill for that session.

How do I remove it if I don’t like it? Run claude plugin uninstall i-have-adhd@i-have-adhd. That’s the whole process.

Want the full setup

The full breakdown, install commands, the exact study quotes, and how to disable it, is packaged in a free one-page guide: The Answer-First Claude Setup. If you want more builds like this one delivered every week, the Actionable AI community walks through exactly this kind of setup live.