How the i-have-adhd Skill Differentiates Between Handling Destructive Actions and Normal Output

The i-have-adhd skill differentiates between handling destructive actions and normal output by checking the operation type against a hierarchy of rules defined in SKILL.md, applying a mandatory confirmation step for destructive operations while streaming standard ADHD-friendly formatting for safe commands.

The ayghri/i-have-adhd repository implements a specialized skill that adapts its response format based on operational risk. When processing requests, the skill must differentiate between handling destructive actions and normal output to balance cognitive accessibility with safety protocols. This differentiation occurs through a rule-based hierarchy defined in the canonical skill specification file.

The Rule Hierarchy in SKILL.md

The behavior is governed entirely by skills/i-have-adhd/SKILL.md, which establishes two distinct processing paths depending on whether the requested operation could cause data loss or system changes.

Standard Output Processing (Lines 31-41)

For non-destructive operations, the skill follows standard ADHD-focused formatting guidelines. These rules prioritize immediate actionability by leading with the next step, numbering multi-step tasks, and maintaining concise, structured output that accommodates attention management needs.

Destructive Action Detection (Lines 119-124)

When the skill detects operations like data deletion, force-push commands, schema migrations, or table drops, it shifts to the "When to break the rules" section. This section contains specific exemptions to standard formatting designed to prevent accidental execution of dangerous commands.

How Destructive Actions Trigger Confirmation

Destructive operations activate Rule 2 from the "When to break the rules" section: "Confirm before acting. Safety wins over brevity." This rule overrides the standard streaming output format and injects an explicit confirmation prompt that requires user verification before proceeding.

⚠️ **Destructive action detected**  
Are you sure you want to execute `rm -rf ./build`?  
Reply "yes" to confirm or "no" to cancel.

The skill pauses execution at this point, requiring affirmative user input before continuing with the operation. This confirmation layer ensures that potentially dangerous commands receive additional scrutiny regardless of the standard ADHD formatting preferences.

Standard Formatting for Non-Destructive Output

Normal output bypasses the confirmation requirement and proceeds directly under the regular formatting rules optimized for ADHD cognitive patterns. The skill structures these responses to minimize cognitive load while maximizing clarity and actionability.

1. Open `src/config.ts`  
2. Add the line `export const DEBUG = true;`  
3. Run `npm test` to verify the change.  

Next: commit the change with `git commit -m "Enable debug mode"` and push.

This format appears immediately without interruption, allowing users to execute safe, reversible operations with minimal friction.

Summary

• The skill checks operation type against the hierarchy in skills/i-have-adhd/SKILL.md to determine risk level.
• Destructive actions trigger Rule 2 (lines 119-124), requiring explicit confirmation before execution.
• Normal output follows standard ADHD-friendly formatting rules (lines 31-41) with immediate, structured responses.
• The differentiation ensures safety protocols take precedence over brevity only when operational risk is detected.

Frequently Asked Questions

What qualifies as a destructive action in the i-have-adhd skill?

Destructive actions include commands that could delete data, force-push to repositories, run schema migrations, or drop database tables. The skill identifies these operations by matching the requested command against patterns defined in the "When to break the rules" section of SKILL.md (lines 119-124).

Where is the differentiation logic implemented?

The differentiation logic resides entirely within skills/i-have-adhd/SKILL.md. Unlike traditional implementations that require compiled code, this skill uses a markdown-based specification file that defines both the standard formatting rules (lines 31-41) and the safety exceptions for destructive operations (lines 119-124).

Can the confirmation requirement for destructive actions be disabled?

No, the confirmation requirement is mandatory according to Rule 2: "Confirm before acting. Safety wins over brevity." This rule takes precedence over the standard ADHD formatting guidelines specifically to prevent accidental data loss or system damage, regardless of user preferences for brevity.

How does the skill format normal output differently from destructive action prompts?

Normal output uses numbered lists for multi-step tasks, leads with immediate next actions, and provides continuous streaming responses. Destructive action prompts interrupt this flow with a warning symbol (⚠️), explicit risk identification, and a binary yes/no confirmation requirement that halts execution until resolved.

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