i-have-adhd: 5.6K stars this week for a skill that puts the answer on line one (2026)

ayghri/i-have-adhd is one Markdown file that changes how your coding agent answers. The first line of every reply becomes something you can act on, multi-step work arrives as a numbered list, and the reply stops once the answer is complete. The repository ships it as a plugin for Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Grok, OpenCode, Kimi Code, Qwen Code, Zed, Pi, Hermes, Antigravity, AstronClaw, and Oh My Pi. 1

The project states the goal in its README: "A skill for your coding assistant that stops it from burying the answer. Action first. Steps numbered." 2

The adoption signals for this week are concrete. GitHub's weekly Trending board listed ayghri/i-have-adhd at rank 16, with 5,589 stars gained in the seven-day window, when the board was read on 20 September 2026. 3 The repository itself stands at 49.1k stars, 2.8k forks, and 245 commits, the most recent dated 19 September 2026; it carries an MIT licence, and updates arrive as commits rather than tagged releases. 2 4 The skills directory at skills.sh counts 19.7K installs of the skill. 5 On 8 September 2026 the project reached the Hacker News front page, where the thread closed at 541 points and 371 comments. 6

What the skill changes

The instruction layer is a 140-line file, skills/i-have-adhd/SKILL.md, alongside plugin manifests and one hook per host. That layer is prose; the only code involved is each host's own plugin machinery. Its front matter sets disable-model-invocation: true, which keeps the skill dormant until you invoke it. 7

SKILL.md opens with five claims about reading that the rules are derived from: working memory holds only what is on screen, understanding an answer and doing it are different events, starting is the hardest step, vague time estimates all feel alike, and visible progress registers while buried progress does not. From those it sets ten rules. 7

Lead with the next action. A command, path, or snippet occupies line one; prose follows if it earns the space.
Number multi-step tasks. One bounded action per step, and no step contains "and then" twice.
End with one concrete next action that takes under two minutes.
Suppress tangents. Finish the open issue, then offer the second one as a separate question.
Restate state every turn, because you cannot hold "step 3 of 5" between messages.
Give specific time estimates — minutes or afternoons, never "a bit".
Make completed work visible by naming what now works and how to try it.
Use a matter-of-fact tone for errors. Location, cause, fix.
Cap lists at five items in the final reply, ranked, with the rest kept back.
Cut preamble, recap, and closers — no "Great question", no "Hope this helps", no "Let me know if you need anything else".

Rule 10 is the one readers notice first, and it is the one the README demonstrates. The project publishes this pair: 2

Great question! Let me think about this. Your auth flow has a few moving pieces: the middleware, the token verification, and the cookie handling. Looking at src/auth.ts, the verifyToken function (around lines 42-58) seems to be using an older jsonwebtoken API. One approach would be to update the package and rewrite that function. After making the change, you'd want to run the auth tests to confirm nothing breaks. By the way, you might also want to look at your dependency versions overall. Hope this helps! Let me know if you want to dig deeper.

Run npm install jsonwebtoken@latest, then edit src/auth.ts:42.
Open src/auth.ts
Replace verifyToken (lines 42–58) with the snippet below
Run npm test -- auth.spec.ts
Next: paste the first failing line if any test fails.

The rules relax in six named cases, and these are what make the skill safe to leave running. 7

You ask for an explanation. You get the full explanation, with headers so you can skim back; the opening and closing pleasantries still stay out.
A destructive action is next — rm -rf, a force push, a schema migration. It confirms before acting, and safety outranks brevity.
The last three turns have been "still broken". It stops patching code, names the assumption that might be wrong, and asks one diagnostic question.
The request is genuinely ambiguous. One short question beats a guess.
A rule would delete the answer. "What are my options" returns two to four ranked options with one-line trade-offs.
A rule fights the harness. The harness's own system prompt wins, and the response shape is preserved inside it.

A pre-send checklist closes the file: delete the first sentence if it announces what you are about to do, delete the last sentence if it recaps or asks "anything else?", drop "by the way" sidebars and hedges that carry no uncertainty, and check that reading only the first and last lines tells you what to do next and what just happened.

Install it on the host you actually use

Every documented route installs the same skill and leaves it switched off until you invoke it. 1

Host	Install	Turned on by
Claude Code	`claude plugin marketplace add ayghri/i-have-adhd` then `claude plugin install i-have-adhd@i-have-adhd`	`/i-have-adhd`
Codex	`codex plugin marketplace add ayghri/i-have-adhd --ref main` then `codex plugin add i-have-adhd@i-have-adhd`	`$i-have-adhd`
Cursor, Amp, other skills harnesses	`npx skills add ayghri/i-have-adhd` (add `-g` for every project)	`/i-have-adhd`
GitHub Copilot	`npx skills add ayghri/i-have-adhd -a github-copilot -g`, or copy the skill folder into `~/.copilot/skills/`	`/i-have-adhd`
Gemini CLI	`curl` the command file into `~/.gemini/commands/`, or `gemini extensions install https://github.com/ayghri/i-have-adhd`	`/i-have-adhd`, or from message one on the extension route
Zed	Agent Panel → Skills → Create skill from URL, pasting the raw `SKILL.md` address	`/i-have-adhd`

Each row follows the installation guide. 1 Cline appears in the skills.sh host list and has no instructions of its own in the project's guide, so treat Cline as the generic route and confirm that it discovers the skill. 1 5

Restart the agent after installing. Skills are indexed at startup, so a missing /i-have-adhd in autocomplete means you have not restarted yet.

If the shape suits you, make it the default with one empty flag file:

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

A SessionStart hook then loads the full ruleset before your first message in every session. On Codex, Grok, Zed, Copilot, or a generic harness there is no hook; you paste the ten rules into the host's persistent rules file instead — ~/.codex/AGENTS.md, ~/.grok/AGENTS.md, or Cursor's Settings → Rules → User Rules. Saying "stop adhd mode" or "normal mode" still switches it off for the session. 1

First test: hand it a failing build

Run the test in a scratch checkout, and hand the agent a real failure to work on. In Claude Code:

/i-have-adhd My production build fails with "Invalid environment variable" and I do not know which one. Give me the first command to run, then at most three next steps.

Four things should be visible in the reply: line one is a command you can paste, the next steps are numbered, the last line names one thing to do next, and the response opens without an acknowledgement. If you get a paragraph of framing instead, the skill is installed and inactive — reopen the session and invoke it again. 1

The test is worth running twice in the same project: once with /i-have-adhd and once without, on the same prompt. The comparison is the evidence, and it costs nothing but two replies.

Second test: check the escape hatches

A formatting skill you leave switched on has to give way when the work needs room. Two prompts confirm it does. 7

/i-have-adhd Explain how the auth middleware in this repository works, end to end.

You should get a long, structured explanation with headers. The explanation override is written for exactly this request, so a reply that shortened it has misread its own rules.

/i-have-adhd Delete the unused migration files in db/migrate and drop the stale tables.

Rule 10 says to cut the preamble, and the destructive-action override outranks it: you should see the plan and a confirmation request before anything runs. If the agent acts immediately, stop the session and check that you are on a current install.

What users and issue reports found

The Hacker News thread is the richest feedback available on this project, and it splits in a useful way. Several commenters describe the skill as a shorter way to write something they already type — "at this point, I just write 'I have adhd, can you explain that again'", said one, while another reported adding a standing instruction to end every summary with a plain-language takeaway. 6

Two practical observations from that thread are worth carrying into your own test. 6

One commenter uses it on demand — invoked when a wall of text arrives — and notes that instructions parked in AGENTS.md "tend to lose weight as context grows", which is the argument for the skill file over a rules line.
Fidelity depends on the model behind the harness. The same commenter reports that GLM and DeepSeek largely ignore this kind of instruction while GPT Luna follows it hundreds of thousands of tokens into a session.

The open issues are where the install-time friction lives. 8 9 10 11

Issue	Opened	What it means for you
#221 — the always-on hook needs Node, and recent Claude Code installs ship without it	16 Sept 2026	On a native-binary install the `SessionStart` hook fails with `node: command not found`. Your flag file sits there looking correct while always-on quietly does nothing. 8
#217 — Copilot docs list an unsupported personal skill directory	16 Sept 2026	`~/.claude/skills/` is a project-level path for Copilot, and its personal registries are `~/.copilot/skills/` and `~/.agents/skills/`. Install into the wrong one and new Copilot sessions report the skill missing. 9
#187 — "A lot of overhead, couldn't this be an output-style instead of a skill?"	11 Sept 2026	The repository is far larger than its 140-line payload because it carries an evaluation harness and per-host packaging. If you want the behaviour with less machinery, the ten rules paste into a rules file on their own. 10
#202 — an unpinned `npm` install example triggers supply-chain warnings	13 Sept 2026	Read the install commands before running them, and pin what you install. 11

A related open request, #193, asks for one reversible multi-agent installation path across the fifteen supported hosts. Until something like it lands, the per-host instructions are the only route, and updating means re-running that host's own update command. 12

Adopt it, or use what you already have

Install it when your sessions are mostly execution — debugging, installing, shipping — and you keep scrolling past the first paragraphs to find the command. It pays off fastest when you switch between harnesses, because the same ten rules travel to all of them.

Look first at what your tool already ships. Claude Code has a built-in Concise output style that "leads with the result, skips preamble and narration, and keeps responses short by default", while keeping complete error reports and confirmation prompts for destructive actions; it needs Claude Code v2.1.237 or later, and you select it with /output-style. That is a zero-install comparison worth making before you add a plugin. 13

Where the skill fits poorly:

Exploratory work. Architecture reviews, research, and brainstorming want long prose, and these rules push against it.
A rules file you already maintain. If your CLAUDE.md or AGENTS.md already sets output format, you are running a simpler version of this one. The skill's advantages are the ten written-out rules, the six overrides, and a single source of truth across hosts.
Always-on on a native-binary Claude Code install. Confirm the hook actually fires; issue #221 is the case where it looks enabled and is not.
Harnesses whose documentation is still catching up. Gemini CLI, Kimi Code, Qwen Code, and Copilot have documented steps, and the Copilot path carries an open correction against it.

If you leave one step for later, make it the second test. Confirming that a destructive request still stops for confirmation is what tells you whether the mode can stay on while you work.

Quick reference

Field	Details
Pick	`ayghri/i-have-adhd`, a ten-rule output skill for coding agents. 2
Best for	Debugging, installs, and shipping, where you want the next command on line one.
Hosts	Claude Code, Codex, Cursor, Copilot, Gemini CLI, Grok, OpenCode, Kimi Code, Qwen Code, Zed, Pi, Hermes, Antigravity, AstronClaw, Oh My Pi. 1
Install (Claude Code)	`claude plugin marketplace add ayghri/i-have-adhd`, then `claude plugin install i-have-adhd@i-have-adhd`; restart. 1
Activation	Off until `/i-have-adhd`; `touch ~/.claude/.i-have-adhd-always` makes it permanent. 1
Size	A 140-line `SKILL.md` with ten rules, six overrides, and a pre-send checklist. 7
Licence	MIT, declared in both the repository and the skill's front matter. 7
First test	Ask for the first command on a real build failure, then run the same prompt without the skill and compare.
Adoption signal	Rank 16 on GitHub's weekly Trending board with 5,589 stars in the seven-day window; 49.1k stars and 2.8k forks overall; 19.7K skills.sh installs; 541 points and 371 comments on Hacker News. 3 5 6
Main caveats	The always-on hook needs Node on PATH; models differ in how closely they follow it; Claude Code's built-in Concise style covers part of the same ground. 8 13