i-have-adhd: Optimizing Coding Agent Outputs for Executive Dysfunction

Overview

i-have-adhd is a specialized skill and plugin set designed to prevent coding agents from "burying the answer" under layers of conversational filler and verbosity. By implementing a strict set of output constraints, it transforms AI responses from narrative explanations into action-oriented instructions, making them more accessible for users with ADHD or those who simply prefer high-density, low-noise technical communication.

The Core Problem: AI Verbosity

Modern coding agents, particularly those powered by recent Claude models, often exhibit a tendency toward extreme verbosity. This behavior typically manifests as:

Preamble and Closers: Starting responses with "Great question!" or ending with "Hope this helps!"
Narrative Explanations: Providing long-winded context before stating the actual fix.
Negative Constraints: Explaining what the AI did not do in addition to what it did.
Circular Reasoning: Adding caveats at the end of a response that may invalidate the primary solution.

The 10 Rules of ADHD-Friendly Output

The i-have-adhd skill enforces ten specific rules to ensure the AI remains focused on the task and the user's cognitive load is minimized:

Lead with the next action: The very first sentence must be the immediate step the user needs to take.
Number multi-step tasks: Use numbered lists for sequences to ensure clarity and trackability.
End with one concrete next step: Every response must conclude with a single, unambiguous action item.
Suppress tangents: Remove any information not directly required to complete the current task.
Restate state every turn: Keep the current status of the project visible to prevent disorientation.
Specific time estimates: Use actual minutes or hours (e.g., "15 minutes") rather than vague terms like "a bit."
Make wins visible: Clearly highlight when a step has been successfully completed.
Matter-of-fact errors: Report errors plainly without apologetic language.
Cap lists at 5 items: Prevent cognitive overload by limiting the length of lists.
No preamble, recap, or closers: Eliminate all conversational filler.

Implementation and Compatibility

The project is implemented as a "skill" that can be integrated into various AI agents. It provides specific adapters and plugins for several platforms:

Claude Code: Supports plugins and hooks for "always-on" functionality.
OpenCode: Includes a server plugin and dedicated commands.
Gemini CLI: Provides native routes for custom commands and extensions.
Cursor: Includes portable skill metadata for integration.
Other Platforms: Compatibility layers for Kimi and Qwen.

Installation

Users can install the skill by directing their CLI prompt to: Install the i-have-adhd skill/plugin from https://github.com/ayghri/i-have-adhd, refer to the repo's AGENTS.md for instructions.

Community Insights and Critiques

While the project has gained significant traction (nearly 30k stars), the community has raised several technical and philosophical points regarding its efficacy and implementation.

Model-Specific Behavior

Many users noted that the need for this skill is most acute with Anthropic's Claude models. Some users reported that even with the skill active, Claude tends to revert to its verbose "Claudian" style after a few turns.

"Claude models are the ones that need this the most and in my experience with this specific skill only maintain the conciseness for a few turns at most before they completely forget and are back to their unfathomable verbosity."

Technical Overhead

Some contributors questioned the repository's size, noting that while the core prompt is relatively small (around 140 lines in SKILL.md), the repository contains thousands of lines of code across dozens of files to support various agent plugins.

Alternative Approaches

Users suggested several alternatives to formal skills for achieving similar results:

Prompting: Using keywords like "BLUF" (Bottom Line Up Front), "tersely," or "Assume ADHD."
Output Styles: Using "output style" configurations which may remind the model more frequently than a global skill.
Model Selection: Some users found that downgrading from Opus 5 to Opus 4.8 significantly improved readability and reduced verbosity.

Accessibility Concerns

There is a debate regarding the naming and framing of the project. While intended to help those with executive dysfunction, some users found the name "i-have-adhd" potentially offensive or a trivialization of a clinical diagnosis used for "10xing productivity."

Comparison: Before vs. After

Feature	Standard AI Response	i-have-adhd Response
Opening	"Great question! Let me think..."	"Run `npm install jsonwebtoken@latest`..."
Structure	Narrative paragraphs	Numbered action steps
Closing	"Hope this helps! Let me know..."	"Next: paste the first failing line..."
Focus	Context and explanation	Immediate execution