Purpose of disable-model-invocation: true in the i-have-adhd Skill Manifest

Setting disable-model-invocation: true in the skill manifest's front-matter forces AI assistants to wait for explicit user commands before applying ADHD-friendly formatting rules.

The ayghri/i-have-adhd repository provides a specialized skill for AI coding assistants that reshapes responses to support users with ADHD. The disable-model-invocation: true entry in the YAML front-matter serves as a critical gatekeeper, ensuring these behavioral modifications never activate without direct user consent via the /i-have-adhd slash command.

How disable-model-invocation: true Controls Skill Activation

This flag determines whether the runtime harness eagerly loads the skill's behavior or waits for a direct signal. When set to true, the parser skips automatic rule injection during session initialization.

Many modern AI assistants—including Claude Code, GitHub Copilot, and Qwen Code—automatically scan and apply skill descriptions at startup. Without this flag, the ADHD-specific formatting rules would shape every response immediately. The setting ensures the skill remains dormant until the user explicitly opts in.

Manifest Declaration in SKILL.md

The flag appears in the YAML front-matter of the skill definition file at the repository root.

In skills/i-have-adhd/SKILL.md (lines 1‑5), the front-matter declares:

---
disable-model-invocation: true
---

This single line instructs the harness parser to bypass automatic loading. The skill's ten behavioral rules—such as leading with the next action, numbering steps, and providing concrete time estimates—remain inactive until the slash command triggers the session state change.

Explicit Invocation Flow

The runtime implements a two-phase activation process controlled by this flag.

Before invocation, the model operates normally:

User: "How do I add a new route in Express?"

Model: Provides a generic answer without ADHD-style formatting.

After explicit invocation, the rules activate:

User: `/i-have-adhd`

Model (now active):
1. Open `src/routes.js`.
2. Insert `router.post('/new', handler);`.
3. Run `npm test` to verify.

Next: `npm run dev` and open `http://localhost:3000/new`.

To deactivate, the user issues:

stop adhd mode

Cross-Platform Implementation Standards

The repository aligns with emerging standards across AI agent harnesses documented in INSTALL.md (lines 62‑66).

According to the installation guide, the flag corresponds to Claude Code's native disable-model-invocation setting. The repository also provides an OpenAI-specific configuration in agents/openai.yaml that uses the analogous allow_implicit_invocation: false policy.

As noted in the documentation:

"Copilot respects disable-model-invocation: nothing applies until you invoke the skill, same as Claude Code (tested in #60)."

This consistency ensures users experience identical opt-in behavior regardless of whether they run GitHub Copilot, Claude Code, or custom OpenAI agents.

Summary

• disable-model-invocation: true prevents automatic application of ADHD-friendly formatting rules at session startup.
• The flag is declared in the YAML front-matter of skills/i-have-adhd/SKILL.md at lines 1‑5.
• AI assistants must wait for the /i-have-adhd slash command before activating the skill's ten behavioral rules.
• The repository maintains compatibility with Claude Code and OpenAI harnesses through equivalent configuration options documented in INSTALL.md and agents/openai.yaml.

Frequently Asked Questions

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

If the flag is set to false or omitted, the AI assistant applies the ADHD-friendly formatting rules to every response automatically, without requiring the /i-have-adhd command. This eager activation forces the ten specific rules—including numbered steps and concrete time estimates—on all output immediately upon session start.

Where is the disable-model-invocation flag documented?

The primary documentation appears in INSTALL.md at lines 62‑66, which explains the flag's effect on Copilot and other runtimes. The declaration itself lives in the YAML front-matter of skills/i-have-adhd/SKILL.md at lines 1‑5. The behavior was validated in pull request #60.

How do I temporarily disable the skill after activation?

Users can deactivate the skill by issuing the command stop adhd mode. This returns the model to its default response style for the remainder of the session, effectively pausing the ADHD-specific formatting rules until the next explicit invocation.

Does this flag work with all AI coding assistants?

The flag works with any harness that respects the skill manifest specification, including Claude Code and GitHub Copilot. For OpenAI agents, the repository provides agents/openai.yaml which implements the equivalent allow_implicit_invocation: false policy to achieve identical behavior across platforms.

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