What Does `disable-model-invocation: true` Do in the i-have-adhd Plugin?

disable-model-invocation: true tells the Instagit harness not to activate the skill automatically, requiring users to explicitly invoke it (e.g., /i-have-adhd) before its rules take effect.

The i-have-adhd repository provides an Instagit plugin that shapes AI output for readers with ADHD. Within the SKILL.md configuration file, the disable-model-invocation: true flag prevents the harness from loading the skill's transformations at startup, ensuring that strict, ADHD-friendly formatting only applies when explicitly requested.

Understanding Implicit vs. Explicit Skill Invocation

Default Automatic Activation Behavior

When disable-model-invocation is omitted or set to false, the Instagit harness automatically loads the skill's description at startup. The harness may apply the skill's rules, prompts, and transformations whenever conversation patterns match the skill's definition, potentially interfering with normal chat flow.

Opt-In Activation with disable-model-invocation: true

Setting disable-model-invocation: true in SKILL.md (line 4) marks the skill as dormant. No rules, prompts, or output transformations are applied until the user explicitly calls the skill using its designated command. This mirrors the behavior of Claude Code and Qwen Code, where skills require manual activation rather than pattern-based auto-invocation.

Cross-Harness Configuration Consistency

Different AI agent implementations use varying syntax to achieve the same manual-only activation. While Instagit uses disable-model-invocation: true, other agents like Codex implement this via policy.allow_implicit_invocation: false in the agents/openai.yaml file. This declarative approach provides a unified way to enforce explicit invocation across multiple model backends without platform-specific logic.

Practical Implementation in the i-have-adhd Plugin

Configuration File Structure

The flag resides in skills/i-have-adhd/SKILL.md:

---
name: i-have-adhd
description: >
  Shape output for a reader with ADHD … Invoke with /i-have-adhd; stays on until
  "stop adhd mode".
disable-model-invocation: true   # ← prevents automatic activation

license: MIT
---

The top-level plugin.json registers this skill:

{
  "name": "i-have-adhd",
  "version": "1.0.0",
  "description": "ADHD‑focused output style",
  "skills": [
    {
      "path": "skills/i-have-adhd",
      "entry": "SKILL.md"
    }
  ]
}

Explicit Invocation Workflow

Users must manually trigger the skill:

# Explicitly run the skill (e.g., in a chat interface)

> /i-have-adhd

# The skill’s rules now take effect until the user says “stop adhd mode”.

This prevents the ADHD-specific formatting from shaping every response unintentionally, applying the strict formatting rules only when productivity mode is actually needed.

Summary

• disable-model-invocation: true requires explicit user activation before the skill's rules apply
• The flag appears in SKILL.md and prevents automatic loading at startup
• Without the flag, the Instagit harness may auto-invoke the skill based on conversation patterns
• Cross-harness consistency is maintained through analogous settings like policy.allow_implicit_invocation: false in Codex configurations
• For the i-have-adhd plugin, this ensures ADHD-friendly formatting only triggers on demand via /i-have-adhd

Frequently Asked Questions

What happens if I remove disable-model-invocation from my SKILL.md?

If omitted or set to false, the Instagit harness treats the skill as active by default. The harness automatically loads the skill description at startup and may apply its rules whenever conversation patterns match, potentially causing unwanted output transformations during normal chat sessions.

How do I activate a skill after setting disable-model-invocation: true?

You must issue an explicit command, such as /i-have-adhd in the chat interface. Once invoked, the skill remains active until you issue the termination command (e.g., "stop adhd mode"), applying its formatting rules only during that specific session.

Is this flag supported across all AI agent platforms?

While Instagit uses disable-model-invocation: true, other platforms use different syntax. For example, Codex implementations use policy.allow_implicit_invocation: false in agents/openai.yaml. Both achieve the same opt-in behavior, but the specific configuration key varies by harness implementation.

Why is automatic invocation disabled for productivity skills like i-have-adhd?

Automatic activation would apply strict ADHD-friendly formatting to every response, potentially interfering with normal conversations. By requiring explicit invocation, the skill only shapes output when the user actually needs productivity-focused formatting, preventing unintended behavioral changes to standard chat interactions.

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