When Can the i‑have‑adhd Skill Rules Be Overridden? 6 Override Conditions Explained

The i‑have‑adhd skill's strict output‑formatting rules can be overridden in six specific situations: user explanations, destructive actions, debug spirals, real ambiguity, rule‑task conflicts, and harness conflicts.

The i‑have‑adhd skill is an open‑source AI conversation framework designed to keep interactions concise, structured, and action‑oriented. While its ten default rules normally persist for an entire session, the skill definition explicitly carves out exceptions where brevity and rigid formatting must give way to safety, clarity, or task requirements. This article breaks down exactly when and how those overrides apply, straight from the source code in ayghri/i-have-adhd.

Where the Override Rules Live

All override conditions are defined in skills/i-have-adhd/SKILL.md, specifically the "When to break the rules" section. The rules remain active until the user explicitly exits with commands like "stop adhd mode" or "normal mode" (SKILL.md – Persistence).

The 6 Override Conditions

1. User Requests an Explanation or Walkthrough

When the user asks for conceptual understanding rather than immediate action, provide a full explanation—still without preambles or closers. Use headers to support skimmability.

Explain the OAuth2 authorization code flow step‑by‑step, using headings so the reader can skim.

The formatting relaxes to accommodate educational structure, but the prohibition against conversational fluff remains.

2. Destructive Actions Are Required

Any operation that cannot be undone—rm -rf, forced Git pushes, schema migrations, table drops—requires confirmation. Safety and clarity override brevity.

User: "Delete the old migrations folder."
Assistant: "Are you sure you want to permanently delete `migrations/`? This cannot be undone."
[await confirmation] → then run `rm -rf migrations/`

Pause the action sequence to secure explicit user consent.

3. Debug Spiral Detected

If the last three conversation turns are "still broken," stop iterating on the same code. Identify a likely wrong assumption and pose one targeted diagnostic question.

User: "The test still fails after three attempts."
Assistant: "It looks like the failure may be due to an outdated mock. Which version of the library are you using?"

This prevents infinite loops of ineffective fixes.

4. Real Ambiguity in the Request

When the user's intent is genuinely unclear, ask a short clarifying question rather than guessing and potentially rewriting code incorrectly.

User: "Fix the bug in the auth module."
Assistant: "Do you mean the token verification error or the cookie handling issue?"

Ambiguity overrides force brevity with a precise, minimal question.

5. Rule Conflicts with the Task

If following a rule would delete the answer itself, let the task win. Retain the answer's shape but ignore the conflicting rule.

User: "List all possible deployment strategies."
Assistant: "1️⃣ Blue‑Green – quick rollback, … (full list) …"

Task completeness takes precedence over strict formatting constraints.

6. Rule Conflicts with the Harness

When system prompts or tool‑call requirements clash with skill rules, announce the tool call, execute the work, and adapt time estimates to the actor that will perform the steps.

User: "Run the lint command."
Assistant: "I'm calling the tool to run `npm run lint`. It will take about 30 seconds for you."

Integration with external systems overrides internal formatting preferences.

How to Exit ADHD Mode Entirely

Beyond situational overrides, users can terminate the skill's behavior completely. According to SKILL.md, saying "stop adhd mode" or "normal mode" ends session persistence and restores default assistant behavior without overrides.

Summary

• Six defined conditions allow i‑have‑adhd rule overrides: explanations, destructive actions, debug spirals, ambiguity, rule‑task conflicts, and harness conflicts.
• Source authority: All conditions are documented in skills/i-have-adhd/SKILL.md under "When to break the rules."
• Safety first: Destructive operations always require confirmation, regardless of brevity goals.
• Manual exit: Users can leave the mode entirely with "stop adhd mode" or "normal mode."

Frequently Asked Questions

Can I override i‑have‑adhd rules for any reason I want?

No. Overrides are restricted to the six documented conditions in SKILL.md. Outside these cases, the default rules (lead with next action, number steps, no preambles) remain mandatory until the user exits the mode entirely.

Does asking for an explanation always trigger an override?

Only when the request is genuinely explanatory or a walkthrough. Pure action requests still follow standard formatting. The override applies to the structure (headers allowed) but maintains the prohibition on conversational padding.

What happens in a debug spiral if I don't want to answer a diagnostic question?

The skill mandates stopping iteration and asking one diagnostic question after three failed attempts. This is non‑negotiable—the framework prioritizes breaking ineffective loops over user preference for continued guessing.

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