How the i-have-adhd AI Coding Assistant Shapes Agent Output for ADHD Readers

The i-have-adhd AI coding assistant is a lightweight plug-in that rewrites raw LLM agent responses to follow a concise, action-first format with numbered steps and capped list lengths, making technical instructions easier for ADHD readers to scan and execute.

The ayghri/i-have-adhd repository provides a runtime-agnostic skill that transforms verbose AI outputs into ADHD-friendly formats. By enforcing a strict post-processing pipeline across multiple agent platforms, the i-have-adhd AI coding assistant ensures every response leads with concrete actions rather than explanatory fluff. Detailed platform-specific installation instructions reside in AGENTS.md, while the general behavior specification is documented in README.md.

Rule Definition and Enforcement Architecture

The Immutable Rules in SKILL.md

The authoritative rule set resides in skills/i-have-adhd/SKILL.md, which defines ten non-negotiable formatting constraints. According to the source code, these rules include instructions to lead with the next action, number multi-step tasks, cap lists at five items, and eliminate preambles and recaps. This markdown file acts as the single source of truth for all output transformations, making the behavior explicit and auditable without buried configuration logic.

The Always-On Post-Processing Hook

Enforcement occurs in hooks/always-on.mjs, which registers a global post-processing hook that intercepts every raw agent response before it reaches the user interface. The hook imports applySkill from the skill processor and invokes it on the message content:

// Example of the always-on hook (hooks/always-on.mjs)
import { applySkill } from '../skills/i-have-adhd/processor.js';
export async function onMessage(message) {
  const raw = await next(message);
  return applySkill(raw);   // rewrites according to the 10 rules
}

This ensures zero-config compliance; developers do not need to manually invoke formatting logic on each query. The onMessage function waits for the raw LLM output via next(message), then returns the transformed result after applying the SKILL.md constraints.

Multi-Runtime Manifest Architecture

The skill maintains compatibility across diverse AI coding environments—Claude Code, OpenCode, Pi, OMP, Gemini, Qwen, and Kimi—through runtime-specific manifest files. Each manifest, such as plugin.json, .claude-plugin/plugin.json, or opencode.json, points to the identical core implementation. This architecture guarantees consistent ADHD-friendly output formatting regardless of which LLM agent a developer uses, as the same SKILL.md rules and hooks/always-on.mjs logic execute in every environment.

Customizing the Output Rules Without Code Changes

Because the rules live in plain markdown rather than compiled code, teams can fork the ayghri/i-have-adhd repository and edit skills/i-have-adhd/SKILL.md to tune the output style. After modifying the rule set, reinstall the customized plug-in using the CLI:

# Install the skill (any runtime)

claude plugin install i-have-adhd@i-have-adhd

# Or from a forked marketplace

claude plugin marketplace add <your-username>/i-have-adhd

This workflow allows organizations to enforce domain-specific formatting standards—such as adjusting the list cap or mandating specific action verbs—while retaining the underlying enforcement infrastructure.

Summary

• The i-have-adhd AI coding assistant processes every agent response through hooks/always-on.mjs to enforce compliance with skills/i-have-adhd/SKILL.md.
• Ten immutable rules govern the transformation, including action-first ordering, numbered steps, and a five-item list cap.
• Multi-runtime support via plugin.json and platform-specific manifests ensures consistent behavior across Claude Code, OpenCode, Gemini, and other platforms.
• Users can customize output styles by editing the markdown rule file and reinstalling the plug-in without modifying source code.

Frequently Asked Questions

Which file contains the formatting rules for the i-have-adhd AI coding assistant?

The skills/i-have-adhd/SKILL.md file stores the authoritative list of ten immutable rules that define how agent output is reshaped, including constraints like capping lists at five items and eliminating preambles.

How does the assistant intercept and modify LLM responses?

The hooks/always-on.mjs file registers a post-processing hook that calls applySkill() on every raw response before it reaches the user, automatically rewriting the content to comply with the SKILL.md rule set.

What AI coding platforms support this plug-in?

The repository provides manifest files—plugin.json, .claude-plugin/plugin.json, opencode.json, and others—that enable the skill to run on Claude Code, OpenCode, Pi, OMP, Gemini, Qwen, and Kimi.

Can I customize the output rules for my team?

Yes. Because the rules are defined in plain markdown within SKILL.md, you can fork the repository, edit the rules, and reinstall the plug-in via claude plugin marketplace add <your-username>/i-have-adhd to enforce custom formatting standards.

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