# How "Lead with the Next Action" Works in i-have-adhd: The ADHD-Friendly Response Rule Explained > Understand the i-have-adhd 'Lead with the next action' rule. Discover how this ADHD-friendly response strategy provides immediate, executable commands for instant action and clarity. - Repository: [Ayoub Ghriss/i-have-adhd](https://github.com/ayghri/i-have-adhd) - Tags: deep-dive - Published: 2026-09-01 --- **The "Lead with the next action" rule requires that the very first line of every response must be a concrete, runnable command, file path, or code snippet that the reader can execute immediately—no context, preamble, or multi-step plans allowed in that first line.** The *i-have-adhd* open-source skill is designed to reshape how AI assistants communicate with people who have ADHD. According to the project source code, the first and most important rule—**"Lead with the next action"**—eliminates the cognitive friction of parsing through explanatory text before knowing what to do. This article breaks down exactly how this rule is defined, enforced, and implemented across all runtimes. --- ## What "Lead with the Next Action" Means The rule is brutally simple: **the first line must be actionable**. Context, reasoning, and follow-up steps come *after*, never before. In [`skills/i-have-adhd/SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md) (lines 33-41), the canonical definition states: > "The first line is something the reader can do. Not context. Not a plan. The action." This means responses that start with phrases like "Let's think about..." or "First, you should understand..." are automatically invalid. The rule forces the model to cut to what matters most for ADHD readers: **the immediate next step** they can take. --- ## Where the Rule Is Defined The *i-have-adhd* skill uses a two-layer definition system to ensure consistency across all AI runtimes: | Layer | File | What It Contains | |-------|------|----------------| | Canonical skill definition | [`skills/i-have-adhd/SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md) | The master ruleset with examples of "bad" vs. "good" responses | | Runtime-specific prompts | [`skills/i-have-adhd/agents/gemini.toml`](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/agents/gemini.toml) | The embedded rule for the Gemini CLI agent (line 10) | In [`agents/gemini.toml`](https://github.com/ayghri/i-have-adhd/blob/main/agents/gemini.toml), the rule appears as: ``` 1. Lead with the next action. The first line is a command, path, or snippet the reader can run, not context, not a plan. ``` This dual-definition approach ensures that whether you're using Gemini, Claude, Codex, or another interface, the same "action-first" constraint applies. --- ## How the Rule Is Enforced The enforcement happens through a **three-stage pipeline** implemented in the source code: 1. **Prompt injection** – The runtime-specific agent file (like [`gemini.toml`](https://github.com/ayghri/i-have-adhd/blob/main/gemini.toml)) bakes the rule into the system prompt, conditioning the model to generate compliant responses. 2. **Response generation** – The model produces output with the action line first, followed by any necessary context. 3. **Pre-send validation** – As documented in the "Pre-send check" section of [`SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/SKILL.md), responses are verified before delivery. If the first line fails the action test, the system edits or rejects the output. This pipeline guarantees **runtime-agnostic enforcement**—the same constraint applies whether you're in a CLI, web interface, or IDE plugin. --- ## Correct vs. Incorrect Usage Examples The source code provides concrete examples that illustrate what passes and fails the "Lead with the next action" rule. ### ✅ Correct: Action-First Response ```text npm install jsonwebtoken ``` > Run the command above, then edit `src/auth.ts:42` as instructed in the next line. The first line is immediately executable. Context follows on line two. ### ❌ Incorrect: Preamble-First Response ```text Let's think about the auth flow first… ``` This violates the rule. The pre-send check would flag or rewrite this response because the opening line provides context rather than action. ### ✅ Correct: Multi-Step Task Still Leads with Action ```text npm install jsonwebtoken ``` 1. Open [`src/auth.ts`](https://github.com/ayghri/i-have-adhd/blob/main/src/auth.ts). 2. Replace the `verifyToken` function with the snippet below. 3. Run `npm test -- auth.spec.ts`. Even complex workflows comply—the first line is runnable, and the numbered list (Rule 2) organizes what follows. --- ## Persistence and Session Behavior The "Lead with the next action" rule is not a one-time setting. According to the "Persistence" section of [`SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/SKILL.md), once activated, the skill stays active for the entire session until explicitly disabled with "stop adhd mode." This ensures **consistent ADHD-friendly formatting** across all interactions without requiring repeated activation. --- ## Summary - **"Lead with the next action"** is the foundational rule of the *i-have-adhd* skill, requiring executable content in the first line of every response. - The rule is **canonically defined** in [`SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/SKILL.md) and **propagated to runtimes** through agent-specific files like [`agents/gemini.toml`](https://github.com/ayghri/i-have-adhd/blob/main/agents/gemini.toml). - **Pre-send validation** enforces compliance, rejecting or rewriting responses that start with context instead of action. - The rule **persists across sessions** until manually disabled, ensuring consistent behavior for ADHD users. --- ## Frequently Asked Questions ### What happens if a response violates the "Lead with the next action" rule? The pre-send check mechanism documented in [`SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/SKILL.md) intercepts non-compliant responses before they reach the user. Depending on implementation, the system either rejects the response for regeneration or automatically rewrites it to move actionable content to the first line. ### Can the rule handle complex, multi-step instructions? Yes. As shown in the source examples, complex workflows simply place the immediate first step as line one, then use numbered lists ( complying with Rule 2) for subsequent steps. The constraint is on *position*, not *complexity*. ### Is "Lead with the next action" specific to the Gemini runtime? No. While [`agents/gemini.toml`](https://github.com/ayghri/i-have-adhd/blob/main/agents/gemini.toml) shows the Gemini implementation, the architecture is runtime-agnostic. The canonical definition in [`SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/SKILL.md) serves as the single source of truth, with each agent file adapting the rule for its specific model and interface. ### How do I disable the "Lead with the next action" rule? According to the persistence documentation in [`SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/SKILL.md), users can deactivate the entire skill—including this rule—by issuing the command "stop adhd mode". This returns responses to standard formatting for the remainder of the session.