i-Have-ADHD Rules 8-10 Exceptions: When to Break the Formatting Guidelines

The i-have-adhd skill allows six specific exceptions to rules 8-10 when the user requests detailed explanations, destructive operations require confirmation, conversations enter debug spirals, requests contain ambiguity, rules conflict with the task itself, or the system harness overrides them.

The ayghri/i-have-adhd repository defines a concise communication framework designed to minimize cognitive load for users with ADHD. Rules 8 through 10 enforce matter-of-fact error tones, 5-item list limits, and zero preamble/closing text. However, the skill explicitly recognizes that rigid adherence can sometimes harm clarity or safety. This guide explains when and how to override these rules based on the canonical source in skills/i-have-adhd/SKILL.md.

Where the Exceptions Are Documented

The complete exception list lives in the "When to break the rules" section of the main skill file:

• Source file: skills/i-have-adhd/SKILL.md (lines 119-133)
• Direct URL: https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md

This section establishes six override conditions that apply universally—including to rules 8, 9, and 10.

The Six Override Conditions for Rules 8-10

1. User Requests Full Explanation or Walk-Through

When the user explicitly asks for step-by-step guidance, the assistant may exceed normal brevity constraints while maintaining the no-preamble/closing style.

User: "Explain why my test is failing and show me the fix step-by-step."

Assistant (Rule 8 override):
1. Open tests/auth.spec.ts and locate line 42.
2. The failure is "expected 200, got 401" because the request lacks an Authorization header.
3. Add Authorization: Bearer ${token} to the request.

The detailed explanation is permitted despite Rule 8's default matter-of-fact error tone because the user requested depth.

2. Destructive Operations Require Confirmation

Any operation that could cause irreversible damage—rm -rf, force-push, schema migrations—triggers a mandatory confirmation prompt, overriding Rule 10's no-preamble rule.

User: "Delete the old temp/ folder."

Assistant (Rule 10 override):
"Are you sure you want to delete temp/ and all its contents? This action cannot be undone."

The safety confirmation takes precedence over concision. Found in SKILL.md lines 122-123.

3. Debug Spiral Detection

When the last three conversation turns consist of "still broken" iterations, the assistant must stop and ask a diagnostic question rather than continue guessing. This overrides all formatting rules temporarily to reset context. Documented at lines 124-125.

4. Real Ambiguity in Requests

If the user's intent is unclear, a brief clarifying question beats a confident but wrong answer. Lines 126-127 explicitly prioritize accuracy over rule adherence.

5. Rule Conflicts With Task Requirements

When following a rule would literally prevent completing the request—such as capping a list at 5 items when the user asked for all items—the task's shape wins.

User: "List all the npm scripts in my package.json."

Assistant (Rule 9 override):
1. test – Runs the test suite.
2. build – Compiles the project.
3. start – Starts the server.
4. lint – Lints the code.
5. dev – Runs the dev server.
6. coverage – Generates test coverage.

The six-item list is correct because "all" in the request supersedes Rule 9's cap. Lines 128-130.

6. Harness Instruction Overrides

System-level prompts, tooling constraints, or runtime-specific requirements (Claude, Codex, OpenCode, etc.) take precedence over skill rules. The harness always wins. Lines 131-133.

What Rules 8-10 Actually Require

Understanding the defaults clarifies when exceptions matter:

Rule	Requirement	Default Behavior
Rule 8	Matter-of-fact tone for errors	Brief, neutral statements without cushioning
Rule 9	Cap lists at 5 items	Truncate with "..." or use sub-grouping
Rule 10	No preamble/recap/closing pleasantries	Immediate answer, no "Here's what you need to know" or "Hope this helps!"

All three appear in skills/i-have-adhd/SKILL.md alongside their exception conditions.

Key Files in the Repository

• skills/i-have-adhd/SKILL.md — Canonical source for all rules and exceptions
• README.md — Repository overview and usage instructions
• AGENTS.md — Runtime-specific entry points and integration mappings

Summary

• Six universal exceptions apply to rules 8-10: user-requested depth, destructive operations, debug spirals, ambiguity, task conflicts, and harness overrides
• Exception 1 permits longer explanations when explicitly requested
• Exception 2 mandates confirmation prompts for irreversible actions
• Exception 5 allows list expansion beyond 5 items when "all" or complete enumeration is required
• All exceptions derive from skills/i-have-adhd/SKILL.md lines 119-133
• The harness (system/tooling) always takes ultimate precedence

Frequently Asked Questions

Can I combine multiple exceptions at once?

Yes. If a user requests a full walk-through (exception 1) of a destructive operation (exception 2), the assistant should provide detailed steps and request confirmation before executing. The skill logic permits stacking when conditions independently apply.

Do exceptions 3 and 4 require me to break Rule 10's no-preamble rule?

Not necessarily. A diagnostic question or clarification request can still be direct: "What's the error message?" rather than "I hope you don't mind, but could you please share..." The tone stays crisp; only the structure changes from answer-seeking to question-asking.

How does the harness exception (6) interact with runtime-specific files?

AGENTS.md maps different runtimes (Claude, Codex, OpenCode) to entry points. If a runtime's system prompt includes constraints—such as mandatory safety disclaimers—that component of the harness overrides Rule 10 automatically. The skill defers to the deployment environment.

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