# When to Break the i-have-adhd Rules: 6 Exceptions for Safety and Clarity > Discover the 6 specific safety and clarity exceptions to i-have-adhd rules. Learn when to break formatting rules for critical interventions and productivity. - Repository: [Ayoub Ghriss/i-have-adhd](https://github.com/ayghri/i-have-adhd) - Tags: best-practices - Published: 2026-08-21 --- **The i-have-adhd skill allows breaking its ten formatting rules only in six specific scenarios—including destructive action confirmation, debug spiral intervention, and explain-style requests—to prevent harm and maintain productivity.** The `i-have-adhd` repository by **ayghri** enforces a rigid structure designed for ADHD-friendly interactions: numbered steps, concrete next actions, suppressed tangents, and strict time estimates. However, the [[`skills/i-have-adhd/SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md)](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md) file recognizes that blind adherence can hinder user goals. Lines 19-30 define **six explicit exceptions** where rule-breaking preserves safety, clarity, and task integrity. ## When to Break the Rules: The Six Exceptions in SKILL.md According to the source code in [`SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/SKILL.md), the following six conditions override the standard ten-rule protocol: ### 1. Explain-Style Requests When users request **full explanations** or **walkthroughs**, the skill expands beyond its usual brevity. While preambles and closers remain forbidden, the response provides comprehensive numbered steps rather than a single next action. ```text User: "Can you walk me through setting up a virtual environment?" Assistant: 1. Install Python 3.10 or newer. 2. Run `python -m venv .venv`. 3. Activate the environment (`source .venv/bin/activate` on Unix, `.venv\Scripts\activate` on Windows). 4. Install dependencies with `pip install -r requirements.txt`. 5. Verify the setup by running `python -c "import sys; print(sys.version)"`. ``` ### 2. Destructive Actions **Data-loss operations** require explicit confirmation, breaking the "next action first" rule to insert a safety gate. Commands like `rm -rf`, forced Git pushes, or database schema migrations trigger a confirmation prompt. ```text User: "Delete the `tmp` folder." Assistant: ⚠️ This action will remove all files in `tmp`. Confirm? (If the user replies "yes", proceed with `rm -rf tmp`.) ``` ### 3. Debug Spirals After **three consecutive failed iterations**, the skill stops suggesting new fixes. Instead of proposing a fourth speculative solution, the response poses a single diagnostic question to break the ineffective loop. ```text User: "My tests still fail after the last change." Assistant: We've tried three fixes without success. What error message do you see when you run `pytest`? (Only one diagnostic question is asked, halting the endless loop.) ``` ### 4. Real Ambiguity When the request is **genuinely unclear**, the skill prioritizes a single clarifying question over probabilistic guessing. This prevents wasting time on incorrect assumptions. ```text User: "Update the config." Assistant: Which configuration file should be updated – `settings.yaml` or `config.json`? ``` ### 5. Rule-Task Conflict If applying a rule would **delete the actual answer** required by the task, the content takes precedence. The response shape (numbered steps, no preambles) remains intact, but the rule causing content deletion is suspended. ### 6. Harness Conflict When the **surrounding agent harness** demands specific behaviors—such as tool calls or alternative time-estimate formatting—the harness requirements override skill constraints. The i-have-adhd structural style is preserved where possible, but system integration takes priority. ```text User: "Run the evaluation script." Assistant: Running the script now… *(tool call executed by the surrounding harness)* ``` ## Summary - The `i-have-adhd` skill enforces ten strict formatting rules defined in [`SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/SKILL.md) but allows **six specific exceptions**. - **Destructive action** and **debug spiral** exceptions prevent data loss and endless iteration loops. - **Real ambiguity** and **explain-style** exceptions prioritize clarity over brevity. - **Rule-task conflicts** and **harness conflicts** ensure content integrity and system compatibility. - All exceptions preserve the skill's core structure (no preambles/closers) while allowing necessary flexibility. ## Frequently Asked Questions ### When does i-have-adhd require confirmation for destructive operations? The skill mandates confirmation for any operation capable of causing **data loss**, such as `rm -rf`, forced Git pushes, or schema migrations. According to the [`SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/SKILL.md) exceptions (lines 19‑30), the standard "next action first" rule breaks to insert a warning and await explicit user confirmation before execution. ### How does i-have-adhd prevent infinite debug loops? After **three consecutive unsuccessful fix attempts**, the skill invokes the **debug spiral** exception. Instead of proposing a fourth speculative solution, the response halts iteration and asks a single targeted diagnostic question, such as requesting specific error messages or logs. ### What happens when the user's request is ambiguous? Under the **real ambiguity** exception, the skill asks **one clarifying question** rather than guessing. This applies when the user provides insufficient context (e.g., "Update the config") and prevents wasted effort on incorrect assumptions while maintaining the skill's direct, numbered format. ### Can the surrounding agent system override i-have-adhd rules? Yes. The **harness conflict** exception states that when the parent agent system requires specific tool calls or formatting, those requirements take precedence over strict skill rules. The response maintains the i-have-adhd structural style (numbered steps, no fluff) where possible, but integrates mandatory system behaviors.