Who i-have-adhd is for#
Developers with ADHD using Claude Code for daily coding work
The ten rules map directly to ADHD-specific challenges in knowledge work: difficulty filtering relevant information, losing the thread in long responses, and the cognitive cost of parsing before acting. Action-first output and numbered steps reduce the pre-processing required to start a task.
Skip if:
You need the AI to walk you through unfamiliar systems or explain architectural decisions in depth. The rule set is tuned for task execution, not exploration, and suppresses the contextual explanations those situations require.
Developers who prefer distraction-free, focused coding sessions
Even without an ADHD diagnosis, many developers find AI assistant verbosity disruptive to flow. The README explicitly notes no ADHD diagnosis is needed. If you have turned off an AI coding tool because responses required too much scanning, the rule set addresses that friction directly.
Skip if:
You regularly ask broad questions where the tangents and context are genuinely useful. For research-style queries or unfamiliar problem domains, default AI behavior provides more signal, not less.
Teams standardizing AI output format across a Claude Code workflow
Because the skill is forkable, teams can maintain a shared fork of SKILL.md with rules tuned for their specific workflow. One fork becomes the team's shared standard for how the AI responds during coding sessions, replacing individually configured setups.
Skip if:
Your team uses a mix of AI coding tools. i-have-adhd only affects Claude Code sessions, so a shared fork helps only the subset of the team on Claude Code.
The problem it solves#
Coding agents produce verbose output by default. Ask where a bug is and the response opens with context-setting, summarizes the problem you already described, walks through multiple possible approaches, and closes with a cheerful filler line before reaching the fix. For developers who need to stay focused, every sentence before the answer is friction.
The pain is not just cognitive load. Verbose responses break the working rhythm. You asked a specific question; you got a performance of helpfulness. For developers with ADHD or anyone who works in focused sprints, the mismatch between how an AI responds and how you need to receive information is a real productivity drain.
How it solves it#
Action-first output
Every response leads with the next action you need to take. No context-setting paragraph, no restatement of your question, no preamble. The README's before/after example contrasts a default response opening with acknowledgment and context-setting prose against the ADHD-friendly version opening directly with the command and the file path to edit.
Numbered multi-step tasks
Any task requiring more than one step is numbered sequentially. Steps are listed in the order you perform them, not in the order the model reasoned through them. This removes the need to parse a prose block for the sequence before you can start working.
Concrete next-step closer
Every response ends with one specific next step, not a filler closing line. The closer is always actionable: a command to run, a file to open, or a failing line to paste back. This keeps the session moving without requiring you to decide what to do next after each response.
Tangent suppression
Rule 4 suppresses tangents: information the model considers relevant but which you did not ask for. If you asked about auth.ts, you do not get a note about your dependency versions. This keeps responses scoped to the question asked.
State restatement per turn
Rule 5 restates the current state at each turn so you never have to scroll up to remember where the session was. If a test failed and you are iterating, the model restates what failed before telling you what to do next. No more losing your place in a long session.
Capped lists and specific time estimates
Lists are capped at five items to prevent unbounded bullet rundowns. Time estimates use specific units such as minutes, not vague qualifiers like 'a bit.' Both rules reduce the cognitive cost of parsing a response before you can act on it.
Strengths and trade-offs#
Strengths
- MIT licensed with no usage restrictionsThe skill is MIT licensed, so you can fork it, modify it, use it commercially, and redistribute it without any restriction. Unlike proprietary AI assistant configurations or locked-down plugin systems, you own every rule you run and can change any of them.
- Tunable by forking SKILL.mdThe ten rules live in a single SKILL.md file. Fork the repository, edit any rule, and install your fork via Claude Code's plugin system instead of the upstream version. The README includes the exact plugin commands to swap between the upstream copy and a personal fork.
- Strong community validationThe repository crossed 30,000 GitHub stars and 1,800 forks in roughly four months since its May 2026 release. That adoption rate reflects how widely developers feel the gap between default AI assistant verbosity and what focused coding work actually requires.
- Zero infrastructure overheadThe skill runs as a plugin inside your existing Claude Code session. No server to deploy, no API to call, no configuration file beyond the plugin install. It works on any machine where Claude Code is installed.
Trade-offs
- -Claude Code-specific, no cross-tool supporti-have-adhd uses Claude Code's plugin and skill system. It does not work with Cursor, GitHub Copilot, Windsurf, or any other AI coding assistant. If your team uses multiple AI tools, each developer needs to be on Claude Code for the rules to apply.
- -Tangent suppression may remove useful contextThe rules are designed for task-driven sessions. In exploratory conversations such as architectural discussions, onboarding to a new codebase, or documentation writing, suppressing tangents removes context you may actually need. The rules optimize for 'fix this bug' scenarios, not 'help me understand this system' scenarios. Forking lets you relax any rule that causes problems.
i-have-adhd vs alternatives#
i-have-adhd vs Default AI Coding Assistant Behavior
The core difference is structural, not cosmetic. AI coding assistants like Cursor and GitHub Copilot deliver thorough responses by default: context first, answer in the middle, polite closer at the end. i-have-adhd changes that order for Claude Code users by enforcing ten rules at the prompt level.
| Feature | i-have-adhd | Default AI assistant behavior |
|---|---|---|
| Response structure | Action first, then steps | Context first, answer later |
| Preamble | None | Common |
| List length | Capped at 5 items | Unlimited |
| Closers | None | Common |
| State restatement | Every turn | Inconsistent |
| Customizable rules | Yes, via fork | No |
| Works across AI tools | Claude Code only | Depends on the tool |
i-have-adhd is the better choice when you use Claude Code and find that default AI responses require too much scanning before you can act. The rule set is particularly effective for longer sessions where verbose responses accumulate and force you to re-read context on every turn.
The default AI assistant style is still the better fit for exploratory work: onboarding to an unfamiliar codebase, architectural discussions, or documentation writing. The rules suppress tangents by design, which strips out genuinely useful context in those scenarios. The fork path lets you tune specific rules rather than running the full set when that tradeoff matters.
Install and self-host#
Install the skill by cloning the repository, then follow the setup instructions in AGENTS.md.
```bash
git clone https://github.com/ayghri/i-have-adhd
```What it's built on#
- Languages
- JavaScriptPythonTypeScript
FAQ#
Does i-have-adhd work with Cursor or GitHub Copilot?
No. i-have-adhd is a plugin for Claude Code specifically. It uses Claude Code's plugin and skill system to inject response rules into the session. Cursor, GitHub Copilot, and other AI coding assistants have different architectures and are not compatible with this plugin.
Is i-have-adhd free to use?
Yes. The plugin is MIT licensed and free to install, use, fork, and modify without restriction. You need a working Claude Code installation to use it, which has its own separate licensing and pricing terms.
Can I customize which rules are active?
Yes. Fork the repository, edit skills/i-have-adhd/SKILL.md to adjust or remove any of the ten rules, then install your fork via the Claude Code plugin system instead of the upstream version. The README includes the exact plugin commands for swapping between the upstream and a personal fork.
Will the rules suppress useful context I actually need?
They can. Tangent suppression (Rule 4) and list capping (Rule 9) are designed to reduce noise in task-driven sessions. For exploratory work such as understanding a new codebase, architectural discussions, or writing documentation, the suppressed context may be useful. The rules work best for focused, task-specific sessions. Forking lets you relax any rule that causes problems for your workflow.
Do I need an ADHD diagnosis to benefit from this plugin?
No. The README states explicitly that no ADHD diagnosis is needed. The ten rules describe a tighter response format: action first, numbered steps, no preamble, concrete closers. Any developer who prefers that structure over verbose, context-heavy AI output will benefit from the plugin regardless of diagnosis.
Similar open-source tools#
OpenCode
OpenCode is an open-source AI coding agent that assists developers in
escrcpy
Graphical Android device mirroring and control with scrcpy
JellyBoxPlayer
Native Jellyfin and Emby music player for every platform
browser
Headless browser for AI agents: 9x faster than Chrome, built in Zig
context-mode
Cuts LLM context use by up to 98% per session via local MCP sandbox
no_human
From ticket to reviewed pull request, on your machine.