How to Stop AI from Waffling and Get It to Give Clear Instructions

You ask a coding assistant where a test failed or how to fix authorization, and get a wall of text in return. The AI first praises your question, then digresses into architecture, recounts file contents, and only somewhere in the middle delivers the actual line of code you need. It always ends with something like "Hope this helps! Feel free to reach out if you have questions."

To get to the point, you have to scroll through and hunt for commands amid the polite noise. Developer ayghri solved this problem radically by creating a plugin for Claude Code and Codex called i-have-adhd.

Why change AI response style

You don't need an ADHD diagnosis to use this tool. The name reflects the core idea: people with attention deficits struggle to wade through blocks of extra text when they need to quickly perform a specific action. However, this format saves time and focus for any developer.

The plugin intercepts the assistant's system prompt and forces it to respond with maximum brevity. The action comes first, steps are numbered, and all lyrical digressions are cut entirely.

What rules the plugin follows

The plugin has 10 rules hardcoded inside. The full list is in the SKILL.md file in the repository. Here's how they change the neural network's behavior:

1. Response always starts with an action. Terminal command or needed line of code first, then explanations.
2. Multi-step tasks become numbered lists.
3. Response always ends with one specific next step. No vague "what do you think?"
4. Any tangents and side advice are blocked.
5. At each step, the assistant briefly reminds you of the current task state.
6. Time estimates are given only in exact minutes—no "this will take a bit."
7. Successful actions are recorded clearly and visibly.
8. Errors are described factually, without apologies or "sorry this happened."
9. Lists are limited to a maximum of five items.
10. Introductions, concluding summaries, and polite sign-offs are forbidden.

Difference in a real example

Let's see what the same answer looks like before and after connecting the plugin.

Standard Claude response:

Great question! Let's think about this. Your authorization system has several parts: middleware, token validation, and cookie handling. Looking at src/auth.ts, the verifyToken function (around lines 42–58) uses a deprecated API from the jsonwebtoken library. One option would be to update the package and rewrite this function. After making changes, you should run the authorization tests. By the way, you might want to check other dependencies too. Hope this helps! Let me know if you'd like to dig deeper.

Response with i-have-adhd connected:

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 step: debug the first failing log line if tests don't pass.

The difference is immediately visible. In the first case, you have to fish out the essence from the text; in the second, you can copy commands straight to the terminal.

Installation and setup

You can connect the plugin in literally two commands.

For Claude Code:

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

After that, just type the /i-have-adhd command in the chat with the assistant. No need to clone the repository locally—Claude will fetch the instructions itself.

For Codex:

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

Invoked with the $i-have-adhd command. Additionally, Codex can connect this skill automatically if it sees a task where this format is appropriate.

Customizing for yourself

The author honestly states that they based it on the book The Adult ADHD Tool Kit (by Russell Ramsay and Anthony Rostain). However, they reworked ADHD daily organization techniques into text formatting rules for language models.

If the default rule set seems too strict, or you want to add something of your own, the project is easy to customize. Just fork the repository, edit the skills/i-have-adhd/SKILL.md file to your preferences, and install the plugin from your own account.

The repository has already gathered over 6.7k stars on GitHub, and this interest is quite understandable. The tool solves a very specific problem: the hyper-verbosity of modern AI.

The project will be useful for anyone who actively uses Claude Code or Codex in daily development and is tired of scrolling through the screen looking for the point. If you want to spend less time reading AI introductions, it's definitely worth a try.