July 24, 2026
i-have-adhd: A Skill That Makes AI Answers Easier to Follow
By Synthex
An AI agent can give you the right information and still make the answer unnecessarily difficult to use.
The actual action may be buried under an introduction. Three separate steps may be squeezed into one paragraph. A small fix may arrive with five unrelated suggestions. Time estimates may be so vague that you still have no idea whether to start now or leave it for tomorrow.
i-have-adhd (opens in new tab) is a free, open-source agent skill that changes how an AI agent presents its answers. It asks the agent to lead with the next action, split work into bounded steps, keep tangents out of the way, show progress clearly, and use concrete language when something fails.
You do not need an ADHD diagnosis to use it. It is simply an output preference for people who find long, indirect, or loosely structured AI responses difficult to act on.
What you'll learn
- What
i-have-adhdis and what an agent skill does. - How the skill changes an AI agent's responses.
- What it can help with and what it cannot fix.
- How to install it in Codex.
- How to install it in Claude Code or Gemini CLI.
- How to activate it, use it for real work, and turn it off.
What is i-have-adhd?
i-have-adhd is a small package of written instructions for AI agents such as Codex and Claude Code.
The main file is called SKILL.md. It describes:
- When the skill should be used.
- How answers should be structured.
- How multi-step work should be presented.
- How the agent should report errors and progress.
- When the normal rules should be relaxed for safety or clarity.
This makes it an agent skill, not a separate AI model or chatbot.
OpenAI describes skills as reusable workflows that can contain instructions, references, scripts, and assets. Codex can discover a skill from its name and description, then load the full instructions only when the skill is needed. This is called progressive disclosure: the complete skill does not have to fill the conversation context from the beginning.
The project is also packaged as a plugin. The distinction is useful:
| Term | Plain-language meaning |
|---|---|
| Skill | The reusable instructions that tell the agent how to handle a task |
| Plugin | The installable package that can distribute one or more skills and other components |
| AI model | The underlying system producing the response |
Installing i-have-adhd does not replace the model. It gives the model a different set of communication instructions.
What the skill changes
The skill is built around a practical problem: knowing the answer is different from being able to start the work.
Its rules change the shape and order of a response:
| Normal response problem | What the skill asks the agent to do |
|---|---|
| The answer begins with background information | Begin with the answer or next action |
| Several actions are buried in one paragraph | Use numbered, bounded steps |
| The conversation moves on before the current issue is finished | Finish the current issue before raising another one |
| Progress is easy to lose between messages | Restate the current step and what comes next |
| Time estimates are vague | Use minutes or hours when an estimate is useful |
| Errors sound dramatic or unclear | State the location, cause, and fix plainly |
| Long lists become difficult to scan | Keep the active list to five items or split it into priorities |
| The response ends with generic pleasantries | End when the work is complete, or give one concrete next action |
This is more specific than asking an agent to "be concise."
A short answer can still be hard to use. It may assume you remember the earlier steps, combine multiple actions, or omit the one thing you should do first. i-have-adhd focuses on action order, working memory, and visible progress, not word count alone.
A simple before-and-after example
Imagine you ask:
A typical answer may begin by explaining several possible causes, discuss dependency management in general, suggest updating other packages, and eventually mention the command that reveals the real error.
With the skill active, the answer should look more like this:
The agent has not solved the problem yet. It has made the first diagnostic step obvious and reduced the amount of information you need to hold in your head.
Who may find it useful
The name is direct, but the format can help more than one kind of reader.
It may be useful when:
- You have ADHD or struggle with working memory.
- You feel overwhelmed by long technical answers.
- You are learning a tool and need one step at a time.
- You return to a task after interruptions.
- You want an agent to show progress during longer work.
- You prefer direct error messages without extra emotional language.
It can also help experienced users who simply dislike padded answers.
What it does not do
This skill changes presentation. It does not make every answer correct.
It does not:
- Improve the model's factual knowledge by itself.
- Give the agent new tools or permissions.
- Guarantee that a time estimate will be accurate.
- Replace testing, source checking, or human review.
- Diagnose ADHD or provide medical treatment.
The agent can still misunderstand the task. A command can still be wrong. A concise answer can still omit important context.
Use the skill to reduce reading and starting friction, then keep the normal checks that matter for the work.
Worth knowing: concrete time estimates are still estimates. They can make a task easier to judge, but they are not promises.
How the skill handles longer explanations
The rules are not meant to make every answer tiny.
If you ask the agent to explain something or walk you through it, the skill allows a longer answer. The explanation should still avoid a padded introduction and should use headings that make it easy to return to a section.
This matters for topics where shortening the answer would remove the explanation itself.
For example:
The skill also makes room for:
- Confirmation before destructive actions.
- One clear diagnostic question when repeated fixes are failing.
- A clarification question when the request is genuinely ambiguous.
- System or safety instructions that override the preferred response style.
The output preference should never remove a necessary warning or skip confirmation before risky work.
How to use i-have-adhd in Codex
After the skill is installed, start a new Codex task and invoke it explicitly:
You can activate it and give the task in the same message:
Or use it for work that has nothing to do with coding:
The bundled OpenAI metadata allows Codex to select the skill when a task matches it, but explicit invocation is easier to verify. If you want this exact response style, name the skill.
Once active, the skill instructs the agent to keep using the rules for the rest of the current session. A new task may require activating it again, depending on how the skill was installed and how your Codex environment loads skills.
To stop using it in the current conversation, say:
or:
The easiest Codex installation method
If your Codex environment can install skills from GitHub, paste this into a Codex task:
Codex should inspect the skill, install it into your personal skills directory, and tell you when it will become available.
Start a new task after installation, then type:
If your Codex environment cannot perform the installation directly, use the plugin method below.
Install i-have-adhd with Codex CLI
The project currently documents these commands:
Then verify the installation:
Start a new Codex session and activate the skill:
OpenAI's plugin documentation also provides a visual route in Codex CLI:
- Run
codex. - Enter
/plugins. - Open the relevant marketplace.
- Select the plugin and install it.
- Start a new session before using the bundled skill.
In the ChatGPT desktop app, plugins can be installed through the Plugins directory when the plugin is available in one of your configured marketplaces. Open the plugin details, select the plus button, then start a new task.
If the skill does not appear immediately: close the current task and start a new one. Codex loads newly installed plugin skills at the beginning of a new task or CLI session.
Install i-have-adhd in Claude Code
The project provides a Claude Code plugin:
Verify it:
Start a new Claude Code session, then activate it with:
Claude Code uses the slash command. Codex uses the dollar-prefixed skill name. Mixing those two commands is a common reason the skill appears not to work.
Install i-have-adhd in Gemini CLI
Gemini CLI can use the custom command included in the repository.
Copy this file:
into:
Start a new Gemini CLI session and run:
The project also offers an always-on Gemini extension:
Use the custom command when you want the format only in selected sessions. Use the extension when you want it loaded automatically.
The repository's complete installation guide also covers Zed, GitHub Copilot, Cursor, OpenCode, Pi, Hermes, and Antigravity: Install i-have-adhd (opens in new tab).
How to check that it is working
Do not test it with a vague question such as "Is the skill active?" Give it a small task where the formatting is visible.
Try:
A working response should:
- Put the first action near the top.
- Break the work into numbered steps.
- Keep the active list short.
- State the current progress when the conversation continues.
- End with one concrete next action if work remains.
If the answer returns to long introductions after several turns, invoke the skill again or start a new task. Agent instructions can drift during long conversations, especially when many other instructions are competing for attention.
A useful way to customize it
You may like most of the skill and dislike one rule.
For example:
- You may want lists to allow seven items instead of five.
- You may prefer a one-sentence explanation before a command.
- You may not want time estimates unless you ask for them.
- You may want completed tasks to end without another suggested action.
The skill is plain Markdown and uses the MIT license, so you can fork the repository and edit skills/i-have-adhd/SKILL.md.
Keep the rules that remove friction for you. Change the ones that create new friction.
A sensible way to use it
Start with on-demand activation rather than making it permanent immediately.
Use it for a few real tasks:
- A small technical problem.
- A longer task with several stages.
- A non-coding task such as research or planning.
- A task where you need a full explanation.
- A task that fails and needs diagnosis.
That will show you whether the structure helps across different situations.
If it does, keep using the explicit command or add the core rules to your personal agent instructions. If one part feels too rigid, edit your copy.
The useful test is ordinary: after reading the response, do you know what to do next without rereading the whole thing?