# Why Does the i-have-adhd Skill Generate Verbose Output? Rule Exceptions Explained > Discover why i-have-adhd generates verbose output by exploring rule exceptions like safety confirmations and debugging. Understand intentional deviations from its terse style. - Repository: [Ayoub Ghriss/i-have-adhd](https://github.com/ayghri/i-have-adhd) - Tags: deep-dive - Published: 2026-07-22 --- **The i-have-adhd skill produces verbose output only when specific override conditions are met—such as safety confirmations, explanation requests, or debugging dead-ends—allowing intentional, rule-driven deviations from its default terse style.** The `i-have-adhd` skill in the `ayghri/i-have-adhd` repository is engineered to deliver action-first, minimalist responses optimized for ADHD-friendly workflow. However, as implemented in [`skills/i-have-adhd/SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md), the skill contains explicit architectural provisions to break its own conciseness rules when context demands additional detail or safety measures. ## Built-in Rule Exceptions in SKILL.md The skill's architecture centers on ten strict output-shaping rules (lines 33-41 of [`SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/SKILL.md)) that enforce brevity, eliminate fluff, and prioritize scan-friendly formatting. Despite these constraints, the designers recognized that strict minimalism can hinder usability in specific scenarios. Consequently, [`SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/SKILL.md) defines a "when to break the rules" section (lines 119-129) that triggers **intentional verbose output** under five specific, well-defined conditions. ## Five Triggers for i-have-adhd Skill Verbose Output ### 1. Explicit Explanation Requests (SKILL.md L119-L124) When a user asks "why" or requests a detailed walkthrough, **Rule 10** ("No preamble, no recap…") is temporarily suspended. The skill expands to provide full exposition—still without unnecessary pleasantries but with the detail necessary for comprehension. ### 2. Safety Checks for Destructive Actions (SKILL.md L122-L124) Before executing destructive commands like `rm -rf` or database schema migrations, the skill overrides **Rule 1** (lead with the next action) to insert confirmation prompts. These prompts briefly explain the risk to prevent accidental data loss. ### 3. Debugging and Repeated Failures (SKILL.md L125-L126) After three consecutive "still broken" turns, the skill stops iterative attempts and emits a diagnostic question. This override inserts extra context to help the user break the deadlock, adding lines beyond the standard single-action response. ### 4. Ambiguous User Requests (SKILL.md L127-L128) Rather than guessing intent, the skill asks clarifying questions when requests are ambiguous. This adds explanatory sentences beyond the usual terse format to ensure the user receives relevant assistance. ### 5. Harness-Level Overrides (SKILL.md L126-L129) When integrated with agent harnesses like Claude, Codex, OpenAI, or Gemini (configured in [`agents/openai.yaml`](https://github.com/ayghri/i-have-adhd/blob/main/agents/openai.yaml) and [`agents/gemini.toml`](https://github.com/ayghri/i-have-adhd/blob/main/agents/gemini.toml)), the skill temporarily relaxes its constraints. This accommodates tool-call formats or harness-specific output requirements that would otherwise violate the "no preamble" rule. ## Code Examples: Concise vs. Verbose Modes ### Standard Concise Response When no override conditions are met, the skill obeys Rule 1 strictly: ```text Run `npm install jsonwebtoken`, then edit `src/auth.ts:42`. ``` ### Verbose Output Triggered by Explanation Request When the user asks for reasoning, the skill expands per lines 119-124: ```text Run `npm install jsonwebtoken`, then edit `src/auth.ts:42`. **Why:** The project depends on the `jsonwebtoken` library for token verification. Installing the latest version ensures you have recent security patches and API compatibility. ``` ### Safety Confirmation Before Destructive Action Before `rm -rf` or similar commands, the skill inserts protective verbosity: ```text ⚠️ Confirm: This command will permanently remove `dist/`. Type "yes" to proceed. ``` ### Harness Override for Tool Calls When the Claude or OpenAI harness requires tool invocation, the skill adds necessary preambles: ```text Running tool to fetch latest dependencies… [tool output] ``` ## Summary - The i-have-adhd skill defaults to strict, action-first brevity defined in [`SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/SKILL.md) lines 33-41. - **Verbose output is intentional**, occurring only through rule overrides defined in lines 119-129. - Five specific triggers enable expanded text: explanation requests, destructive action safety checks, debugging dead-ends, ambiguous inputs, and harness integration requirements. - Configuration files [`agents/openai.yaml`](https://github.com/ayghri/i-have-adhd/blob/main/agents/openai.yaml) and [`agents/gemini.toml`](https://github.com/ayghri/i-have-adhd/blob/main/agents/gemini.toml) demonstrate how external harnesses interact with the skill's output constraints. ## Frequently Asked Questions ### What file controls when the i-have-adhd skill breaks its own rules? The [`skills/i-have-adhd/SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/skills/i-have-adhd/SKILL.md) file contains the complete rule set and exception logic. Lines 119-129 specifically define the conditions under which the skill may produce verbose output despite Rule 10's "no preamble" constraint. ### Does verbose output indicate the i-have-adhd skill is malfunctioning? No. According to the source code in `ayghri/i-have-adhd`, verbosity is a controlled, intentional feature. The skill toggles between ADHD-friendly minimalism and richer explanatory modes only when safety protocols, debugging necessities, or user clarification requirements demand it. ### Which agent harnesses can trigger verbose output in i-have-adhd? The skill includes configuration files for OpenAI ([`agents/openai.yaml`](https://github.com/ayghri/i-have-adhd/blob/main/agents/openai.yaml)) and Gemini ([`agents/gemini.toml`](https://github.com/ayghri/i-have-adhd/blob/main/agents/gemini.toml)), and also responds to harness-level requirements from Claude and Codex. These integrations can force temporary rule relaxations to accommodate tool calls or specific formatting requirements. ### How does the skill handle destructive commands? Before executing destructive actions like `rm -rf` or database migrations, the skill overrides its standard brevity to insert safety confirmations. This override is documented in [`SKILL.md`](https://github.com/ayghri/i-have-adhd/blob/main/SKILL.md) lines 122-124 and prevents accidental data loss by requiring explicit user confirmation with a brief risk explanation.