What Is `disable-model-invocation: true` in the i-have-adhd Skill?

The disable-model-invocation: true flag in skills/i-have-adhd/SKILL.md prevents the ADHD-focused skill from activating automatically, requiring an explicit user command such as /i-have-adhd to enable its behavioral transformations.

In the ayghri/i-have-adhd repository, this YAML frontmatter setting functions as a gatekeeper that keeps the skill dormant until deliberately invoked. Unlike skills that apply their rules immediately upon loading, this configuration ensures the ADHD-specific response modifications remain inactive until you explicitly request them. The repository implements this pattern to give users complete control over when the specialized formatting and cognitive accessibility features take effect.

How disable-model-invocation Controls Skill Activation

The frontmatter directive resides in the skill's configuration file and signals to compliant runtimes that automatic activation is prohibited.

When the runtime loads the skill, it parses the metadata in skills/i-have-adhd/SKILL.md and recognizes the disable-model-invocation: true declaration. This instructs the system to load the skill's description into memory but refrain from applying any behavioral transformations to the model's output. The skill effectively exists in a passive state, monitoring for specific invocation strings but not interfering with standard model responses.

Only after you issue an explicit command—typically /i-have-adhd—does the runtime activate the skill's rule set. Once activated, the ADHD-focused formatting guidelines, conciseness constraints, and structural preferences modify how the model generates responses for the remainder of the session.

Why Explicit Invocation Matters for ADHD-Focused Interactions

The intentional dormancy prevents accidental or unwanted rule enforcement that could disrupt normal workflows.

Users with ADHD may need specialized response formatting—such as bullet-point summaries, chunked information delivery, or specific tonal adjustments—only during certain tasks or focus sessions. By requiring explicit activation via the /i-have-adhd command, the skill respects the user's contextual needs without permanently altering the AI's default behavior. This design philosophy acknowledges that accessibility features are most helpful when user-controlled rather than universally applied.

Runtime Compatibility and Equivalent Configurations

The disable-model-invocation pattern appears across multiple AI coding assistants, though implementations vary by platform.

Claude Code and Qwen Code both respect this flag natively, treating skills with disable-model-invocation: true as opt-in rather than opt-out features. In these environments, the skill definition loads at startup but remains inactive until explicitly called.

OpenAI Codex achieves the same behavioral outcome through a different configuration syntax. In agents/openai.yaml, setting policy.allow_implicit_invocation: false produces identical results—preventing the agent from automatically applying skill behaviors without user consent.

The repository also maintains a Cursor-specific mirror at .cursor/skills/i-have-adhd/SKILL.md that includes the same disable-model-invocation: true flag, ensuring consistent behavior across different editor environments.

Activating and Deactivating the Skill Manually

You control the skill's state through explicit conversational commands. The runtime responds to these specific interaction patterns:

# Explicitly enable the skill

/i-have-adhd

After issuing this command, the model immediately applies ADHD-friendly rules—such as structured formatting and concise paragraph lengths—to all subsequent responses.

# The skill stays inactive without invocation

# Standard questions receive normal, unmodified responses

Without the /i-have-adhd trigger, the runtime ignores the skill's transformation rules entirely, even though the skill definition remains loaded in the background.

# Disable the skill and return to standard behavior

stop adhd mode

This command deactivates the skill's influence, causing the model to revert to its default response patterns for the remainder of the conversation.

Summary

• disable-model-invocation: true in skills/i-have-adhd/SKILL.md blocks automatic skill activation upon load.
• The skill remains dormant until you explicitly invoke it with /i-have-adhd or equivalent commands.
• This explicit opt-in model prevents unwanted behavioral changes and gives users complete control over when ADHD-focused formatting applies.
• Compatible runtimes include Copilot, Claude Code, and Qwen Code; Codex achieves similar behavior via policy.allow_implicit_invocation: false in agents/openai.yaml.
• A Cursor-specific copy exists at .cursor/skills/i-have-adhd/SKILL.md with identical frontmatter settings.

Frequently Asked Questions

What happens if I don't explicitly invoke the i-have-adhd skill?

If you do not issue the /i-have-adhd command, the skill remains completely inactive. According to the documentation in INSTALL.md, "nothing applies until you invoke the skill"—meaning you receive standard model responses without any ADHD-specific formatting, bullet constraints, or structural modifications.

How do I know if the skill is currently active?

The skill activates immediately after you send the /i-have-adhd command in the chat interface. You will observe immediate changes in the model's response style, including shorter paragraphs, structured bullet points, and concise summaries. The skill remains active until you explicitly disable it with "stop adhd mode" or start a new session.

Does the disable-model-invocation flag work in all AI coding assistants?

No, support varies by runtime. Copilot, Claude Code, and Qwen Code respect the disable-model-invocation: true flag directly. However, OpenAI Codex uses a different configuration approach—requiring policy.allow_implicit_invocation: false in agents/openai.yaml to achieve the same opt-in behavior. Always verify your specific runtime's documentation for skill frontmatter compatibility.

Where can I find the skill configuration files?

The primary configuration resides in skills/i-have-adhd/SKILL.md at the repository root. A Cursor-specific implementation exists at .cursor/skills/i-have-adhd/SKILL.md. The installation and usage guidelines are documented in INSTALL.md, which explains how various runtimes interpret the disable-invocation setting.

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