How the disable-model-invocation Feature Works in SKILL.md

The disable-model-invocation flag in SKILL.md prevents the skill from automatically applying formatting rules to every chat interaction, requiring explicit user invocation via commands like /i-have-adhd to activate the ADHD-focused output style.

The i-have-adhd repository provides a declarative skill configuration for Copilot and Claude Code that reshapes AI responses into ADHD-friendly formats. Within skills/i-have-adhd/SKILL.md, the disable-model-invocation field serves as the primary control mechanism that determines whether the skill’s aggressive formatting rules apply automatically or wait for manual activation.

Understanding the disable-model-invocation Flag

The disable-model-invocation field is a top-level boolean configuration in SKILL.md that defaults to true in the i-have-adhd skill. When enabled, this flag instructs the underlying harness—whether Copilot, Claude Code, or the AGY plugin—to load but not enforce the skill’s rules during standard chat interactions.

According to the installation documentation in [INSTALL.md](https://github.com/ayghri/i-have-adhd/blob/main/INSTALL.md#L404-L407), this behavior ensures that "nothing applies until you invoke the skill," giving developers explicit control over when the ADHD formatting mode is active.

The Three-Phase Activation Lifecycle

Phase 1: Skill Loading with Automatic Blocking

When the repository is installed, the harness reads SKILL.md and recognizes the skill’s presence. However, because disable-model-invocation: true is set, the harness does not apply any output-shaping rules—such as "lead-with-action" or numbered steps—to regular chat messages. The skill exists in a dormant state where it is available but inactive.

Phase 2: Explicit User Invocation

The skill activates only when the user types a specific slash command matching the skill name. In practice, entering /i-have-adhd triggers the harness to detect the invocation and immediately apply all defined rules from SKILL.md.

Once invoked, the assistant switches to ADHD-mode formatting: responses lead with concrete actions, use numbered steps, and follow the specific structural rules defined in the skill configuration.

Phase 3: Session Deactivation

The activated skill remains in effect for the duration of the session until the user explicitly signals completion. The harness recognizes deactivation phrases such as "stop adhd mode" or "normal mode", at which point it unloads the skill and returns to the default output style without the specialized formatting constraints.

Practical Examples of disable-model-invocation Behavior

The following interactions demonstrate how the flag controls skill activation in real usage scenarios.

Example 1: Normal Query Without Invocation (Skill Disabled)

User: How do I run the test suite?
Assistant: [Standard response format without ADHD-specific structuring]

Because disable-model-invocation blocks automatic activation, Copilot or Claude Code returns a conventional answer despite the skill being technically installed.

Example 2: Explicit Invocation (Skill Activated)

User: /i-have-adhd
User: How do I run the test suite?
Assistant: 1. Run `npm install` to install dependencies
        2. Execute `npm test` to run the suite
        3. Review results in `test-output.log`

After the slash command, the harness activates the skill and applies all formatting rules defined in SKILL.md, transforming the response into actionable, structured steps.

Example 3: Deactivating the Skill

User: stop adhd mode
Assistant: [Returns to standard conversational formatting]

The harness detects the deactivation phrase and disables the skill's rules for subsequent interactions.

Implementation Files and Configuration

The disable-model-invocation feature spans multiple configuration files within the repository:

• skills/i-have-adhd/SKILL.md: Defines the skill metadata and sets disable-model-invocation: true at the top level of the declarative configuration.
• INSTALL.md (lines 404-407): Documents how Copilot and Claude Code respect this flag, noting that the skill remains inactive until explicitly invoked.
• GEMINI.md: Contains optional configuration that could bypass the disable-model-invocation restriction if added, creating an "always-on" override for specific use cases.

Summary

• The disable-model-invocation flag in SKILL.md controls whether a skill applies automatically or requires explicit activation.
• When set to true, the skill loads but remains dormant until the user invokes it via the /i-have-adhd slash command.
• The harness respects this setting across Copilot, Claude Code, and compatible plugins, ensuring consistent behavior.
• Users can deactivate the skill mid-session using phrases like "stop adhd mode" or "normal mode".
• Optional bypass configurations in GEMINI.md can override this behavior for specialized deployment scenarios.

Frequently Asked Questions

What happens if I set disable-model-invocation to false?

Removing or setting this flag to false would cause the harness to automatically apply the skill's rules to every chat message without requiring explicit invocation. While this creates an "always-on" experience, it may surprise users who prefer standard formatting for general queries.

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

According to the INSTALL.md documentation, both Copilot and Claude Code respect the disable-model-invocation directive. The behavior is implemented by the surrounding harness (whether GitHub Copilot, Claude Code, or the AGY plugin), making it compatible across supported platforms that read SKILL.md configurations.

How can I tell if the ADHD skill is currently active?

The skill is active when you have recently invoked it with /i-have-adhd and have not yet deactivated it with "stop adhd mode" or "normal mode". When active, responses will exhibit the distinct formatting patterns defined in SKILL.md, such as leading with actionable commands and using numbered steps instead of prose paragraphs.

Can I bypass the manual invocation requirement?

Yes. The repository includes optional configuration in GEMINI.md that demonstrates how to create an "always-on" setup that bypasses the disable-model-invocation restriction. This configuration is designed for users who want the ADHD formatting applied to every interaction without requiring the /i-have-adhd prefix.

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