How to Make AI Give Actionable Commands First: The 9.1K‑Star “i‑have‑ADHD” Skill

Problem: Verbose AI output

Strong LLMs tend to start with background, list possibilities, and drift into unrelated topics, hiding the critical execution steps at the end of a long paragraph. This makes developers who frequently use AI for code debugging or task breakdown lose focus and forget the core action.

Solution: i‑have‑ADHD skill

The GitHub project ayghri/i-have-adhd (over 9,100 stars) transforms AI responses from “explanation‑first” to “action‑first” by applying ten strict output rules. It works with Claude Code and Codex without modifying the underlying model.

Core rules (adapted from The Adult ADHD Tool Kit )

Action pre‑placement : the first line must state the immediate operation, file path, and target.

Step numbering : any response with multiple actions must use a numeric list; large text blocks are prohibited.

Single terminal instruction : the conclusion contains only one concrete, executable next step.

No off‑topic digression : the AI must not expand into unrelated business or optimization before the current issue is closed.

Continuous progress sync : for long tasks, each turn repeats the current completed node to avoid context loss.

Quantified time estimate : time is expressed precisely in minutes; vague terms like “soon” are forbidden.

Result visualization : the response must clearly indicate the observable effect after the modification.

Objective error handling : errors follow the fixed format “error location – root cause – fix”.

List length limit : a list may contain at most five items; excess items are split into “now do / later plan” sections.

Remove all boilerplate : delete generic greetings, polite closings, and any non‑essential chatter.

Installation guide

Claude Code

Add the plugin from the marketplace:

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

Enable per session with /i-have-adhd or enable globally by creating a marker file: touch ~/.claude/.i-have-adhd-always Codex

Add the plugin:

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

Activate with $i-have-adhd; the tool will invoke the rules automatically.

Safety verification & common pitfalls

The repository includes nine standardized unit tests that pass 100%, confirming the rule scripts load correctly. However, the tests do not guarantee every model will obey the constraints in complex, long‑text scenarios.

Frequent issues observed:

When globally enabled, AI may over‑compress explanations for teaching or creative tasks; temporarily disable the constraint by requesting “full explanation”.

High‑risk destructive commands (e.g., file deletion) may bypass the confirmation dialog if the global flag is active; manually disable the flag for such operations.

Lists longer than five items stop being split; refresh the rule cache by re‑issuing the enable command.

Older plugin versions can error; delete the local cache and pull the latest main branch.

To quickly validate effectiveness, compare a verbose AI reply with the same query after enabling the skill. Check three criteria: first line contains an executable action, steps are numbered, and only one two‑minute actionable item remains at the end.

Target audience and cautions

Programmers, DevOps, and testers who frequently use AI for code, error diagnosis, or task breakdown.

Developers whose attention is easily fragmented by long texts.

Teams that need standardized AI output to reduce communication overhead.

Fast‑paced users who dislike hunting for core actions in AI chatter.

Not recommended for:

Scenarios requiring deep theoretical explanations, extensive brainstorming, or creative writing.

Beginners needing step‑by‑step teaching.

Literary or marketing content creation that benefits from rich context.

Limitations

The skill only reshapes the output format; it does not improve the model’s factual correctness. All generated code or solutions must still be verified and tested by the user.

https://github.com/ayghri/i-have-adhd