[BUG] Skills/commands with disable-model-invocation: true appear in autocomplete but fail when invoked

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Summary

Skills and slash commands that have disable-model-invocation: true in their frontmatter appear in the autocomplete menu when users type /. However, when selected, they fail with an error stating the skill cannot be invoked due to disable-model-invocation. This creates a confusing UX where users are presented with non-functional options.

Environment

• Claude Code version: 2.1.23
• Platform: Linux
• OS: Linux 6.6.87.2-microsoft-standard-WSL2

The Problem

Example: /brainstorm command from superpowers plugin

The command file (commands/brainstorm.md) has:

---
description: "You MUST use this before any creative work..."
disable-model-invocation: true
---

When typing /, the autocomplete shows:

• /brainstorm - Appears in menu
• superpowers:brainstorming - Also appears as a skill

Attempting to invoke:

User types: /superpowers:brainstorm

Result:

Skill superpowers:brainstorm cannot be used with Skill tool due to disable-model-invocation

Impact

• User confusion: Why is this option presented if it doesn't work?
• Broken discovery flow: Users exploring available functionality hit errors
• Inconsistent with intent: disable-model-invocation should mean "model can't auto-invoke" not "users can't select from menu"

What Should Happen?

Option A: Items with disable-model-invocation: true should NOT appear in autocomplete

• Prevents confusion
• Only shows invokable options

Option B: If they do appear, they should work when selected

• The command invocation path should work regardless of the flag
• The flag should only prevent autonomous Skill tool invocation by the model

Proposed Solution

When building the autocomplete menu for /:

1. Filter out items with disable-model-invocation: true at the command level, OR
2. Allow user selection of these items to work (respect the intent - user explicitly chose it)

The disable-model-invocation flag should prevent:

• Model from autonomously invoking via Skill tool
• Auto-detection/inclusion in context

But should NOT prevent:

• User from explicitly selecting from autocomplete menu
• User from typing the command directly

Error Messages/Logs

Steps to Reproduce

1. Install a plugin with a command that has disable-model-invocation: true

``bash
/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers@superpowers-marketplace
``

1. Type / to open autocomplete menu
2. Observe that commands/skills with disable-model-invocation: true appear in the list
3. Select one of these items
4. Observe error: "Skill X cannot be used with Skill tool due to disable-model-invocation"

Claude Model

Opus

Is this a regression?

No, this never worked

Last Working Version

_No response_

Claude Code Version

2.1.23

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

WSL (Windows Subsystem for Linux)

Additional Information

_No response_