i-have-adhd — Stop Your Agent Burying the Answer

Your coding agent just opened with “Great question! I’d be happy to help with that.” The answer is in paragraph four. This is the problem i-have-adhd fixes — and about 57,000 GitHub accounts have starred the repo in the five months since it launched.

On 2026-10-06, the plugin manifests recorded commits for 0.4.0 and 0.4.1. The repository publishes no formal GitHub releases, so you are tracking the main branch rather than a release page — but it is the moment worth documenting, because the star count tells you something: formatting guardrails are now table-stakes for agent UX.

What Happened

i-have-adhd is a SKILL.md file — a formatting contract your coding agent reads at session start. It was created by Ayoub Ghriss (GitHub: ayghri, University of Colorado Boulder) on May 13, 2026, and has accumulated approximately 57,000 GitHub stars by mid-October. The repo publishes no GitHub releases; version numbers live in the plugin manifests only.

The ten rules are the entire product:

• Lead with the next action, not context
• Number multi-step tasks
• End every response with one concrete next step
• Suppress tangents
• Restate state every turn (“Step 3 of 5 done”)
• Time estimates in minutes, not “a bit” or “shortly”
• Make wins visible
• Report errors matter-of-factly
• Cap visible lists at 5 items — without dropping relevant items
• No preamble, no recap, no closers

The skill names five design assumptions: small working memory, the gap between knowing and doing, starting is the hardest step, vague time estimates all feel the same, visible progress matters. It is loosely based on The Adult ADHD Tool Kit by J. Russell Ramsay and Anthony L. Rostain, adapted for how an LLM responds rather than how a human organizes their day. The README is explicit: no ADHD diagnosis needed to use it.

It runs on Claude Code, Codex, Cursor, OpenCode, Gemini, Qwen, and Kimi. The star count suggests most people installing it do not have ADHD — they just want their agent to answer the question.

Why It Matters

The Hacker News thread from September 8 (542 points, 371 comments) captures the frustration bluntly. One commenter: “I literally tweeted yesterday that Claude has ADHD and nothing I do will keep him focused.” Another: “This is just an annoying thing for anyone. It gives a 10 page dissertation that sums up to, ‘it’s good, nothing to worry about’.” A third raised the sharper question: “I like the problem this is trying to solve. One thing I’m not sure about though is whether this kind of a preference should be a skill at all.”

That last comment is the one worth sitting with. The correct answer is probably “no, it shouldn’t need to be a skill” — but it does need to be one, because agent defaults are not changing fast enough. The skill’s existence is a workaround for a product failure, and 57,000 stars is the market telling you the failure is widespread.

The repo ran its own evaluation on August 2, 2026: 14 test prompts × 3 trials on Claude (claude-opus-4-8), scored blind by the same model on correctness, autonomy, actionability, safety, and concision. Weighted score moved from 4.045 without the skill to 4.473 with it. The skill won 10 of 14 cases, tied 2, lost 2. Blocking findings dropped from 7 to 3. The author’s own release gate still failed — any blocking finding fails it. The gains concentrate in status reports and error reports. Codex was not tested.

Same-model judging has obvious limits. But this is more rigorous than most plugin authors bother with, and the author published the failures rather than hiding them.

The Take

Install it. The install is two commands in Claude Code:

claude plugin marketplace add ayghri/i-have-adhd
claude plugin install i-have-adhd@i-have-adhd

Then type /i-have-adhd in a session to activate. Installing alone does nothing — disable-model-invocation: true means the skill is opt-in by design. It stays on until you say “stop adhd mode”. If you want it always-on:

touch ~/.claude/.i-have-adhd-always

In Codex, the commands are:

codex plugin marketplace add ayghri/i-have-adhd --ref main
codex plugin add i-have-adhd@i-have-adhd

Invoke with $i-have-adhd — Codex uses $, not /.

On claude.ai (web or desktop, including the Free plan), you can upload it manually: Customize → Skills → ”+” → Create skill → Upload a skill, using a ZIP of the skill folder. Also enable Settings → Capabilities → “Code execution and file creation”. The repo’s INSTALL.md does not document this path — it is the general Claude skill upload mechanism, not an official install route.

ChatGPT Business, Enterprise, Healthcare, and Edu seats support skill uploads. On Free, Go, or Plus, paste a condensed version of the ten rules into custom instructions. The full skill exceeds the Free/Go character limit of 1,500 characters — Plus gives you 5,000, which is enough for a tight condensed version.

Two things to watch after you install it. First: the two cases the eval lost. The skill suppresses context, and sometimes context is the answer — particularly when the agent is flagging something it has not confirmed yet. If it names a cause confidently, verify it did the check. Second: the 5-item list cap is a presentation rule, not a permission to drop items. If you are getting truncated output on a task with more than five steps, ask explicitly for the full list.

This is a formatting contract, not a productivity system. It makes answers easier to act on. It does not start the task for you.

The repo is MIT licensed, has commits in October 2026, and carries no GitHub releases — version numbers are in the manifests only. That is fine for a skill this simple. The lack of a release page means you are tracking the main branch, not pinned releases. Pull occasionally.