# How the i-have-adhd Skill Modifies AI Coding Assistant Output

> Discover how the i-have-adhd skill transforms AI coding assistant output into clear, actionable steps. Get instant, optimized instructions for ADHD users.

- Repository: [Ayoub Ghriss/i-have-adhd](https://github.com/ayghri/i-have-adhd)
- Tags: how-to-guide
- Published: 2026-08-01

---

**The i-have-adhd skill is a response-shaping plug-in that rewrites AI assistant replies into instantly actionable, step-by-step instructions optimized for users with ADHD.**

This open-source skill transforms verbose LLM outputs into highly scannable formats by enforcing ten strict formatting rules. Originally developed by Ayoub Ghriss for Claude Code and Codex, the skill helps developers with limited working memory maintain context without losing track of complex technical tasks.

## What the i-have-adhd Skill Does

According to the [`skills/i-have-adhd/SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md) manifest in the ayghri/i-have-adhd repository, the skill acts as a **post-processor** that intercepts and restructures AI responses before they reach the user. Unlike standard prompting, this skill uses a `disable-model-invocation` flag—defined in [`skills/i-have-adhd/agents/openai.yaml`](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/agents/openai.yaml)—to prevent the LLM from generating its own response and instead applies deterministic formatting rules.

The transformation happens automatically once activated, without requiring manual prompt engineering for every request.

## The 10 Rules That Reshape AI Output

The skill enforces these specific constraints on every assistant reply:

- **Lead with the next action** — The first line is always a concrete command (CLI instruction, file path, or code snippet) with zero preamble.

- **Number multi-step tasks** — Solutions requiring multiple actions become numbered lists, each step being a single bounded operation.

- **End with one concrete next step** — Every response closes with a "Next:" prompt completing in under two minutes.

- **Suppress tangents** — Side issues wait until the main task finishes; no "by the way" digressions allowed.

- **Restate state every turn** — Current step number and overall progress repeat on each interaction.

- **Give specific time estimates** — Durations use concrete units (e.g., "≈ 12 minutes"), never vague phrasing.

- **Make completed work visible** — Successful outcomes display explicitly, such as "Login now works with `npm run dev`".

- **Matter-of-fact tone for errors** — Error reports state cause and fix plainly, without emotive language.

- **Cap lists at five items** — Longer lists split into "do now" vs. "later" or "must" vs. "nice-to-have" categories.

- **No preamble or closing pleasantries** — Replies start directly with answers and end immediately when complete.

## How to Activate the i-have-adhd Skill

### Explicit Activation in Claude Code

Install from the marketplace and enable:

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

```

During any session, type:

```

/i-have-adhd

```

The assistant maintains ADHD-friendly formatting until you type `stop adhd mode`.

### Implicit Activation in Codex

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

```

When requesting tasks matching the action-first pattern—such as "add JWT authentication to my Express app"—Codex automatically applies the skill without explicit invocation. The [`skills/i-have-adhd/agents/gemini.toml`](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/agents/gemini.toml) file defines this command integration for Gemini CLI compatibility.

### Deactivation

| Platform | Command |
|----------|---------|
| Claude Code | `stop adhd mode` |
| Codex | `normal mode` |

## Real Output Transformation Example

**Standard LLM response** (typical):

> To implement JWT authentication, you'll want to consider several approaches depending on your security requirements. First, let's think about the token structure and what claims you'll need. There are many libraries available, and each has trade-offs regarding bundle size and community support...

**i-have-adhd skill output**:

```

Run `npm install jsonwebtoken@latest`, then edit src/auth.ts:42.

1. Open src/auth.ts
2. Replace verifyToken (lines 42–58) with the snippet below
3. Run `npm test -- auth.spec.ts`

Next: run the tests and paste the first failing line.

```

The transformation removes 80%+ of structural overhead while preserving all actionable information.

## Key Implementation Files

Understanding these source files reveals how the skill modifies AI coding assistant output:

- **[`skills/i-have-adhd/SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md)** — Complete rule specification and behavioral contract (see reference ¹).

- **[`skills/i-have-adhd/agents/openai.yaml`](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/agents/openai.yaml)** — Defines `disable-model-invocation: true` and agent metadata that routes responses through the skill transformer (see reference ²).

- **[`skills/i-have-adhd/agents/gemini.toml`](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/agents/gemini.toml)** — Gemini CLI command definition for cross-platform skill invocation (see reference ³).

- **[`plugin.json`](https://github.com/ayghri/i-have-adhd/blob/main/plugin.json)** — Platform registration entry point specifying skill name, version, and activation hooks (see reference ⁴).

- **[`hooks/always-on.sh`](https://github.com/ayghri/i-have-adhd/blob/main/hooks/always-on.sh)** — Optional session-start hook auto-loading the skill when `~/.adhd-always-on` flag file exists (see reference ⁵).

- **[`README.md`](https://github.com/ayghri/i-have-adhd/blob/main/README.md)** — User documentation with before/after comparison examples (see reference ⁶).

## When This Skill Provides Maximum Value

The i-have-adhd skill modification triggers automatically for:

- Multi-file refactoring tasks with sequential dependencies
- Debugging workflows requiring precise reproduction steps
- Onboarding sequences with environment setup requirements
- API integration tasks with authentication prerequisites

The `disable-model-invocation` architecture guarantees consistent formatting regardless of underlying model temperature or context window behavior.

## Summary

- The **i-have-adhd skill** rewrites AI assistant responses through deterministic post-processing, not prompt variation.

- **Ten strict rules** govern every output: action-first structure, numbered steps, time estimates, suppressed tangents, and zero social friction.

- **Explicit activation** uses `/i-have-adhd` (Claude) or automatic pattern matching (Codex); **implicit activation** triggers on action-first task detection.

- Core files [`SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/SKILL.md), [`openai.yaml`](https://github.com/ayghri/i-have-adhd/blob/main/openai.yaml), and [`gemini.toml`](https://github.com/ayghri/i-have-adhd/blob/main/gemini.toml) define the transformation logic and platform integrations in the ayghri/i-have-adhd repository.

- The `disable-model-invocation` flag ensures responses follow rules regardless of base model behavior.

## Frequently Asked Questions

### Can I use the i-have-adhd skill with any AI coding assistant?

The skill currently supports **Claude Code** and **Codex** through official platform integrations, with **Gemini CLI** support via [`gemini.toml`](https://github.com/ayghri/i-have-adhd/blob/main/gemini.toml) configuration. Other assistants require manual implementation of the ten rules in their system prompts, as the `disable-model-invocation` mechanism is platform-specific.

### Does the skill change what the AI knows or only how it presents information?

Only presentation. The skill modifies AI coding assistant output format without altering underlying model knowledge, context window, or reasoning capabilities. The transformation happens after generation completes, ensuring factual accuracy remains unchanged while accessibility improves.

### Why disable model invocation instead of using a system prompt?

The `disable-model-invocation: true` flag in [`openai.yaml`](https://github.com/ayghri/i-have-adhd/blob/main/openai.yaml) guarantees **deterministic rule enforcement**. System prompts rely on model compliance; this architecture intercepts and restructures any response, eliminating variance from temperature settings or context length pressure.

### Can I customize the ten rules for my specific workflow?

The published skill enforces rules rigidly. However, [`SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/SKILL.md) uses standard markdown formatting—forking the ayghri/i-have-adhd repository and modifying your local [`SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/SKILL.md) creates a custom variant. Submitting pull requests with tested modifications back to the original repository is encouraged per the [`CONTRIBUTING.md`](https://github.com/ayghri/i-have-adhd/blob/main/CONTRIBUTING.md) guidelines.