How the 10 ADHD Rules in i‑have‑ADHD Shape Model Outputs for Better Focus

The 10 ADHD rules in ayghri/i-have-adhd transform model outputs by enforcing an action-first, low-cognitive-load format designed specifically for readers with ADHD.

The i-have-adhd skill is a behavioral prompt engineering framework that restructures how coding assistants generate responses. Defined in [skills/i-have-adhd/SKILL.md](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md) and loaded at runtime per [AGENTS.md](https://github.com/ayghri/i-have-adhd/blob/main/AGENTS.md), these rules activate when a user prefixes their request with /i-have-adhd. The model then applies post-processing guidelines that prioritize executability over explanation, keeping readers engaged and oriented throughout complex tasks.

How the Rule System Works

The skill operates through three pipeline stages when shaping model outputs:

• Prompt interception – The /i-have-adhd slash command triggers the skill loader to inject metadata from SKILL.md's header block
• Rule enforcement – System prompt injection guides the decoder to obey constraints while generating each token
• Post-generation sanitization – A lightweight formatter reorders sections, adds state markers, and prunes disallowed phrasing

This architecture ensures the 10 ADHD rules shape every aspect of the final output without requiring manual editing by the user.

The 10 Rules and Their Output Effects

Each rule targets a specific cognitive barrier associated with ADHD—working memory limits, task initiation difficulty, or distractibility—and translates into concrete formatting requirements.

Rule 1: Lead with the Next Action

The very first line must be a concrete, executable step.

Output transformation: Any preamble like "Let's think about…" moves to an immediate directive. The model restructures responses to open with commands, file paths, or code snippets.

# Before standard output

"To add authentication, we should consider which library fits your stack..."

# After Rule 1 enforcement

"Run `npm install jsonwebtoken`"

Source: [SKILL.md lines 33-40](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md#L33-L40)

Rule 2: Number Multi-Step Tasks

Tasks are expressed as a numbered list with bounded, independent steps.

Output transformation: Long prose collapses into 1. … 2. … 3. … format. This explicit structure reduces working memory load by giving each step clear boundaries.

Source: [SKILL.md lines 42-56](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md#L42-L56)

Rule 3: End with One Concrete Next Action

The final line gives a single action taking ≤2 minutes.

Output transformation: The model never terminates with "Hope this helps" or similar closures. Instead, outputs end with direct instructions like "Next: run npm test".

Source: [SKILL.md lines 57-63](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md#L57-L63)

Rule 4: Suppress Tangents

Secondary issues are postponed and presented separately.

Output transformation: The response stays focused on the primary task. Any side issue appears only after a clear "Separately:" prompt, preventing mid-task distraction.

Source: [SKILL.md lines 64-70](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md#L64-L70)

Rule 5: Restate State Every Turn

The model repeats the current step/count after each turn.

Output transformation: Cross-turn continuity markers appear automatically: "Step 2 of 4 done: … Next: …". This orientation mechanism compensates for context-switching costs.

Source: [SKILL.md lines 73-80](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md#L73-L80)

Rule 6: Give Specific Time Estimates

Vague durations convert to concrete minutes or hours.

Output transformation: "A bit of work" becomes "About 15 minutes". Time-boxing helps with temporal estimation challenges common in ADHD.

Source: [SKILL.md lines 82-88](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md#L82-L88)

Rule 7: Make Completed Work Visible

After a fix, the model shows the tangible result.

Output transformation: Rather than summarizing changes, outputs read "Login now works; try npm run dev". This provides immediate feedback to maintain motivation.

Source: [SKILL.md lines 89-95](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md#L89-L95)

Rule 8: Matter-of-Fact Tone for Errors

Errors are reported plainly, without "uh-oh" language.

Output transformation: Emotional framing is stripped. Outputs read "Test fails at …: expected 200, got 401". This prevents anxiety spirals from catastrophic language.

Source: [SKILL.md lines 96-101](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md#L96-L101)

Rule 9: Cap Lists at 5 Items

Long lists split into "do now" versus "later" sections.

Output transformation: The model never displays more than five bullet points before breaking the list. This respects the "magic number" of working memory capacity.

Source: [SKILL.md lines 103-106](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md#L103-L106)

Rule 10: No Preamble, No Recap, No Closing Pleasantries

The response starts directly with the answer and ends when finished.

Output transformation: Phrases like "Great question" and "Let me know if you need anything else" are automatically pruned. The output density increases substantially.

Source: [SKILL.md lines 107-114](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md#L107-L114)

Practical Example: Before and After

Activating the Skill

# Install in Claude Code

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

# Invoke in chat

/i-have-adhd Explain how to add a new route in Express

Model Output Following All 10 Rules

1. Create the file – src/routes/user.js.

2. Add the route – insert:

   router.get('/users', async (req, res) => {
     const users = await User.findAll();
     res.json(users);
   });

3. Register the router – in src/app.js add app.use('/api', require('./routes/user')).

4. Run the server – npm start.

5. Verify – open http://localhost:3000/api/users in a browser.

Next: open src/routes/user.js and paste the snippet above.

Notice: action-first opening (Rule 1), numbered steps (Rule 2), 5-item cap (Rule 9), concrete next action ending (Rule 3), no pleasantries (Rule 10), and implicit time estimate through step granularity (Rule 6).

Key Implementation Files

File	Purpose	Location
skills/i-have-adhd/SKILL.md	Core rule definitions, metadata header, and behavior specifications	[skills/i-have-adhd/SKILL.md](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md)
AGENTS.md	Runtime entry point mapping and skill loading protocol	[AGENTS.md](https://github.com/ayghri/i-have-adhd/blob/main/AGENTS.md)
README.md	Installation guide and rule summary for end users	[README.md](https://github.com/ayghri/i-have-adhd/blob/main/README.md)
.github/readme/README.zh-CN.md	Localized documentation demonstrating rule consistency across languages	Chinese localization

Summary

• The 10 ADHD rules shape model outputs through systematic prompt engineering defined in SKILL.md
• Three pipeline stages—prompt interception, rule enforcement, and post-generation sanitization—guarantee consistent application
• Each rule addresses a specific ADHD-related cognitive barrier: working memory limits (Rules 2, 5, 9), task initiation (Rules 1, 3), distractibility (Rule 4), temporal estimation (Rule 6), motivation maintenance (Rule 7), and emotional regulation (Rule 8)
• The output density increases dramatically by eliminating framing language while preserving all actionable information
• Activation requires only the /i-have-adhd prefix, with deactivation via "stop adhd mode"

Frequently Asked Questions

How do I activate the ADHD output mode in my coding assistant?

Install the skill through your runtime's plugin system (e.g., claude plugin install i-have-adhd@i-have-adhd), then prefix any request with /i-have-adhd. The rules apply immediately to that conversation thread until you say "stop adhd mode".

Can I use these rules with models other than Claude?

Yes. The SKILL.md format is runtime-agnostic. Any system that supports skill loading via AGENTS.md specifications can apply these rules. The repository includes documentation in multiple languages, indicating broad compatibility goals.

Why are the rules limited to 5 items per list?

Rule 9 reflects cognitive psychology research on working memory capacity. By capping lists at 5 items, the skill prevents the overwhelm that occurs when readers with ADHD encounter unbounded information streams. The model automatically splits longer sequences into "do now" and "later" sections.

Do the rules remove important explanatory content?

No. The rules restructure information rather than delete it. Explanations that would normally appear in preambles or recaps are either integrated into step descriptions or deferred to "Separately:" sections per Rule 4. The net information content stays equivalent while presentation becomes more scannable.

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