How the "Lead with Action" Rule Works in i-have-adhd: A Technical Breakdown

The "lead with action" rule in i-have-adhd requires that every response start with a concrete, executable step the reader can perform immediately, with no preamble, context, or planning language allowed beforehand.

The ayghri/i-have-adhd repository defines a Claude skill designed specifically for readers with ADHD. The core philosophy is simple: eliminate friction between understanding and doing by front-loading every response with something actionable. This article examines how the "lead with action" rule is defined, why it matters for ADHD cognition, and how it functions within the broader response structure.

Rule Definition and Location

The formal rule appears in skills/i-have-adhd/SKILL.md at line 33, where it is presented as the first and foundational directive for all generated content.

The rule explicitly states that the first line of every response must be an actionable command—not background, not a plan, not context-setting language. The skill file provides direct before-and-after examples to illustrate this constraint:

• Bad: "Let's think about the authentication flow before we modify anything."
• Good: "Run npm install jsonwebtoken, then edit src/auth.ts:42."

These examples appear in lines 33-40 of SKILL.md and serve as the canonical reference for correct implementation [1].

Why Action Must Come First

The rule's placement is deliberate and grounded in ADHD-specific cognitive research documented within the same skill file.

According to the "Five facts" section (lines 25-30 of SKILL.md), people with ADHD experience "small working memory" [1]. This means intermediate context, explanatory framing, or multi-step planning sequences overload cognitive resources before any actual progress occurs.

By leading with the action:

• The reader receives a singular focal point immediately
• No working memory is consumed holding "why" before "what"
• The barrier between reading and doing collapses to zero

How the Rule Interacts with Other Guidelines

The "lead with action" rule does not operate in isolation. It forms the foundation of a cohesive response structure defined across the skill.

Other rules that build upon this foundation include:

• Numbered steps – Breaking subsequent guidance into ordered, bite-sized units
• Restating state – Periodically clarifying what has been accomplished so far
• Time estimates – Attaching duration predictions to every action

Together, these rules enforce a terse, concrete, progress-oriented response style throughout any session using the i-have-adhd skill [1].

Implementation in the Plugin Pipeline

The i-have-adhd plugin enforces this rule through its processing pipeline, configured across multiple files.

The plugin.json file provides the high-level description that shapes output according to skill-defined rules. Various runtime-specific manifests—including .claude-plugin/plugin.json and kimi.plugin.json—propagate these constraints to each supported model environment [1].

When formatting a response, the pipeline executes this sequence:

1. Detect any actionable command or code snippet in the generated content
2. Elevate that snippet to the first line of the reply
3. Append explanatory prose only after the action (and only if necessary)

Practical Examples

Correct Usage (Action First)

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

The `jsonwebtoken` library provides `sign()` and `verify()` methods you'll need for token handling.

Incorrect Usage (Action Buried)

Authentication in Node.js typically involves several libraries that each handle different aspects of the OAuth2 flow. Before implementing anything, let's think about the tradeoffs between session-based and token-based approaches...

Run `npm install jsonwebtoken`.

The second example fails the rule because the actionable command appears only after cognitive load has already been imposed.

Code Block Placement

The rule applies regardless of response type. Even when providing shell commands:

git checkout -b feature/auth-refactor

This single line must appear first, with any branch naming rationale or merge strategy discussion following below.

Key Files Reference

File	Purpose
skills/i-have-adhd/SKILL.md	Contains the formal rule definition at lines 33-40
skills/i-have-adhd/SKILL.md	Documents ADHD cognitive constraints ("Five facts") at lines 25-30
plugin.json	High-level plugin configuration enforcing skill rules
.claude-plugin/plugin.json	Claude-specific runtime manifest
kimi.plugin.json	Kimi-specific runtime manifest
README.md / GEMINI.md	Summarize overall response style guidelines

Summary

• The "lead with action" rule mandates that every response begin with an executable command, not preamble
• Defined in skills/i-have-adhd/SKILL.md at line 33, with examples at lines 33-40
• Designed specifically for small working memory constraints common in ADHD (documented lines 25-30)
• Enforced by the plugin pipeline via plugin.json and runtime-specific manifests
• Works in concert with numbered steps, state restatement, and time estimates for complete response structure

Frequently Asked Questions

What happens if the response doesn't contain an obvious action?

The skill still requires starting with the most concrete available step. Even diagnostic or exploratory responses must lead with something executable—such as "Run ls -la to check current file permissions" rather than "First, we need to understand your directory structure."

Does the rule apply to multi-step explanations?

Yes. Each major response or continuation must re-lead with action. The rule operates at the message level, not just the conversation level, ensuring that a reader returning to any point in a thread can immediately resume progress.

Why is "lead with action" the first rule rather than another guideline?

According to the i-have-adhd source, action-first ordering is prioritized because it addresses the most critical friction point for ADHD readers: the gap between comprehension and execution. Other rules optimize subsequent processing, but this rule ensures processing can begin at all.

Are there exceptions for purely informational responses?

The skill documentation provides no exceptions. Even informational content must be reframed with an action orientation—such as "Read the following three facts about JWT expiration" rather than "Here are some things to know about JWTs."

---

[1] Based on source analysis of ayghri/i-have-adhd repository, specifically skills/i-have-adhd/SKILL.md and plugin configuration files.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}