10 Rules Enforced by the i-have-adhd Skill: Complete Technical Guide

The i-have-adhd skill enforces ten strict communication rules—including leading with executable actions, numbering multi-step tasks, capping lists at five items, and eliminating conversational filler—to create a lean, dopamine-friendly communication style designed for readers with ADHD.

The i-have-adhd skill in the ayghri/i-have-adhd repository is a specialized AI behavior modifier that reshapes every response according to a strict manifest defined in skills/i-have-adhd/SKILL.md. These ten rules persist for the duration of a session (until the user says "stop adhd mode") and are designed to minimize cognitive load while maximizing executive function support.

The Ten Core Rules

1. Lead with the Next Action

According to skills/i-have-adhd/SKILL.md (lines 35-42), the very first line of any response must be a concrete, immediately executable step—such as a command, file path, or code snippet. No introductory prose precedes this action.

2. Number Multi-Step Tasks

When a solution requires more than one step, the skill presents a numbered list where each item constitutes a single, bounded action. As implemented in SKILL.md (lines 44-52), trivial steps are combined to keep the list concise and scannable.

3. End with One Concrete Next Action

Every explanation must conclude with a single, doable action that can be completed in under two minutes (e.g., "run npm test"). This rule, defined in SKILL.md (lines 59-66), ensures the user always has a clear, immediate path forward.

4. Suppress Tangents

If a secondary issue appears during execution, the skill finishes the current task first before presenting the new issue as a separate question. This prevents context switching, as specified in SKILL.md (lines 66-73).

5. Restate State Every Turn

Because working memory is limited, each response must restate current progress (e.g., "Step 3 of 5 done..."). This requirement in SKILL.md (lines 75-82) ensures the reader always knows exactly where they are in a workflow.

6. Give Specific Time Estimates

Vague durations are prohibited; every estimate must use concrete units such as minutes or hours. The enforcement in SKILL.md (lines 84-90) eliminates ambiguous language like "soon" or "a while."

7. Make Completed Work Visible

The skill must show the result of any change immediately (e.g., displaying a command that now works properly). This visibility, mandated in SKILL.md (lines 91-97), provides positive feedback and confirms success.

8. Matter-of-Fact Tone for Errors

Errors are reported plainly without emotive language (no "Uh oh" or "Oops"). As defined in SKILL.md (lines 98-104), the skill states the cause and fix directly to reduce anxiety and ambiguity.

9. Cap Lists at 5 Items

If a list would exceed five items, it must be split into categories (e.g., "do now" vs. "do later"). This boundary, set in SKILL.md (lines 105-108), keeps the immediate workload manageable and prevents overwhelm.

10. No Preamble, Recap, or Closing Pleasantries

The response starts immediately with the answer and ends as soon as the answer is complete. According to SKILL.md (lines 109-116), there are no greetings, "let me..." statements, or sign-offs like "Happy coding!"

How to Invoke the Skill

Activate the mode by typing:

/i-have-adhd

The skill remains active across all subsequent interactions until explicitly disabled. The behavior modifications are applied persistently for the duration of the session.

Example Output Following All Rules

Below is a sample response demonstrating the action-first structure:

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

1. Open `src/auth.ts`.
2. Replace the `verifyToken` function (lines 42-58) with the snippet below.
3. Run `npm test -- auth.spec.ts`.

Next: run `npm test` and paste the first failing line.

Step 3 of 3 done: tests updated. About 5 minutes total.

This output demonstrates leading with the next action, numbered steps, a concrete final action, state restatement, specific time estimates, and zero pleasantries.

Model Configuration

The skill supports multiple AI backends through configuration files:

• skills/i-have-adhd/agents/openai.yaml — OpenAI model parameters and prompting templates
• skills/i-have-adhd/agents/gemini.toml — Gemini model configuration and safety settings

These files bind the ten rules to specific model instances while maintaining consistent behavior across providers.

Disabling the Skill

To return to normal conversational mode, type:

stop adhd mode

The skill confirms the deactivation in a single line and immediately resumes standard chat behavior.

Summary

• Action-first structure: Every response begins with an executable step and ends with a sub-two-minute next action.
• Working memory support: State is restated every turn, lists are capped at five items, and tangents are suppressed until the current task completes.
• Error neutrality: Failures are reported plainly without emotive language, stating only cause and solution.
• Zero filler: No greetings, apologies, or sign-offs appear anywhere in the output.
• Session persistence: Rules remain active from /i-have-adhd invocation until stop adhd mode is issued.

Frequently Asked Questions

How do I activate the i-have-adhd skill?

Type /i-have-adhd to enable the skill. It remains active for the duration of your session across all subsequent messages until you explicitly disable it.

Where are the i-have-adhd skill rules defined?

The complete rule set is defined in skills/i-have-adhd/SKILL.md in the repository root, with specific line references for each behavior constraint (e.g., lines 35-42 for the "lead with action" rule).

Can I use the skill with different AI models?

Yes, the repository includes configuration files for both OpenAI (skills/i-have-adhd/agents/openai.yaml) and Gemini (skills/i-have-adhd/agents/gemini.toml), allowing the ten rules to be enforced across different model providers.

What happens if a task requires more than five steps?

The cap lists at 5 items rule requires splitting long sequences into categorized groups (such as "do now" vs. "do later") to maintain manageable cognitive chunks and prevent executive function overload.

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