Override Conditions for I-Have-ADHD Rules: 6 Critical Safety Exceptions

The I-Have-ADHD skill permits six specific override conditions that allow breaking its standard terse response rules when users request detailed walkthroughs, destructive actions require confirmation, debug loops spiral endlessly, requests remain ambiguous, rules conflict with explicit tasks, or system harness constraints demand compliance.

The ayghri/i-have-adhd repository defines a strict communication protocol designed for ADHD accessibility in skills/i-have-adhd/SKILL.md. While the default ten response-shaping rules enforce brevity and immediate actionability, the specification explicitly defines override conditions for I-Have-ADHD rules that prioritize safety, clarity, and task completion over rigid conciseness.

The Six Override Conditions Defined in SKILL.md

According to the source code in skills/i-have-adhd/SKILL.md (lines 17-27), the following six circumstances permit deviation from the standard response constraints:

1. User Requests a Full Explanation or Walkthrough

When the user explicitly asks for a walkthrough or detailed explanation, the agent may expand beyond the usual terse style. As implemented in skills/i-have-adhd/SKILL.md (lines 21-22), responses can add headers and detailed prose while still avoiding preambles and closings.

User: "Can you walk me through setting up the project?"

Agent (override #1):
1. Clone the repo: `git clone https://github.com/ayghri/i-have-adhd.git`
2. Install dependencies: `npm install`
3. Run the app: `npm start`

2. Destructive Actions Require Confirmation

For potentially dangerous operations such as rm -rf, force-push, or schema migrations, the agent must request confirmation before proceeding. According to lines 22-23 of skills/i-have-adhd/SKILL.md, this override prioritizes safety over brevity.

User: "Delete the old migrations folder."

Agent (override #2):
"Are you sure you want to delete `migrations/`? This operation cannot be undone."

3. Debug Spiral Detection

When the last three interaction turns indicate a "still broken" state, the agent must stop the iterative fix-loop. As specified in skills/i-have-adhd/SKILL.md (lines 23-24), the agent should name the most likely faulty assumption and pose a single diagnostic question rather than attempting another automatic fix.

User: "The test still fails after my last change."

Agent (override #3):
"The last three attempts all failed at `auth.spec.ts:42`. I suspect the token header is missing. Would you like to inspect the request configuration?"

4. Real Ambiguity in the Request

When the user's request contains genuine ambiguity, the agent must ask a concise clarification question instead of guessing. Lines 24-25 of skills/i-have-adhd/SKILL.md specify that a short clarification beats speculative rewrites.

User: "Fix the issue."

Agent (override #4):
"Could you specify which file or function is causing the error?"

5. Rule Conflicts with the Task Itself

If following a standard rule would delete or prevent the answer (e.g., enforcing "no preamble" when the user explicitly asked for a list of options), the task requirements take precedence. According to skills/i-have-adhd/SKILL.md (lines 25-27), the content must fulfill the request while the shape of the answer maintains consistency with the rule set where possible.

User: "List my deployment options with pros and cons."

Agent (override #5):
1. **Docker** – Portable, easy CI integration; slower startup.
2. **Kubernetes** – Scales automatically; higher operational overhead.

6. Rule Conflicts with the Harness

When system-level constraints (the "harness") conflict with skill rules—such as requiring an announcement before tool calls—the agent must follow the harness requirements. As noted in lines 26-27 of skills/i-have-adhd/SKILL.md, the overall response shape should remain consistent even when accommodating these mandatory interruptions.

User: "Run the linter."

Agent (override #6):
"Running `npm run lint` … (tool call announced as required by the harness)."

Implementation Context and Related Files

The override conditions integrate with the broader repository structure established in ayghri/i-have-adHD. The canonical definition resides in skills/i-have-adhd/SKILL.md, while AGENTS.md provides guidance on locating these manifests. Installation instructions appear in INSTALL.md, and the user-facing overview lives in README.md. All file paths follow the repository root structure, with line-specific references (17-27) marking the "When to break the rules" section.

Summary

• Override condition 1 permits detailed walkthroughs when explicitly requested, expanding beyond the default terse format while maintaining actionability.
• Override condition 2 mandates confirmation for destructive operations like rm -rf or force-push, prioritizing data safety over conciseness.
• Override condition 3 breaks debug spirals by stopping iterative fixes after three failures and switching to diagnostic questioning.
• Override condition 4 requires clarification questions when requests contain real ambiguity, preventing speculative errors.
• Override condition 5 resolves conflicts between rules and task requirements by allowing the task to win when rules would otherwise delete the answer.
• Override condition 6 accommodates system harness constraints that supersede individual skill rules, such as mandatory tool-call announcements.

Frequently Asked Questions

What file contains the canonical definition of I-Have-ADHD override conditions?

The canonical definition resides in skills/i-have-adhd/SKILL.md within the ayghri/i-have-adhd repository, specifically under the "When to break the rules" section spanning lines 17-27.

Does the destructive action override apply to all potentially dangerous commands?

Yes, override condition 2 applies to any destructive operation including rm -rf, force-push operations, database schema migrations, or any command that cannot be undone without data loss or significant system impact.

How does the debug spiral override prevent endless troubleshooting loops?

When the agent detects that the last three interaction turns have all resulted in "still broken" states, override condition 3 triggers a pattern break: the agent stops attempting automatic fixes, identifies the most probable faulty assumption in the current approach, and asks exactly one diagnostic question to redirect the troubleshooting strategy.

Can multiple override conditions activate simultaneously?

Yes, override conditions can compound. For example, a destructive action might also occur during a debug spiral, requiring both confirmation (condition 2) and a shift to diagnostic questioning (condition 3) while still respecting any system harness requirements (condition 6).

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