[BUG] Skill tool incorrectly loads commands from .claude/commands/ as skills

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?

The Skill tool accepts and "successfully loads" files from .claude/commands/ directory, reporting them as skills even though they are commands.

Scenario:

• User has a command defined at .claude/commands/issue_create.md
• AI invokes Skill(issue_create) instead of recognizing it as a slash command
• Skill tool outputs: Successfully loaded skill · 1 tool allowed

The problem is twofold:

1. The AI model incorrectly uses the Skill tool for something that is a command
2. The Skill tool accepts and loads it without error, claiming success

What Should Happen?

Either:

1. The Skill tool should reject loading commands and return an error like "issue_create is a command, not a skill. Use /issue_create directly."
2. OR the AI should be instructed not to use the Skill tool for items in .claude/commands/

Commands and skills are distinct concepts:

• Commands (.claude/commands/): Slash commands invoked with /command_name
• Skills (.claude/skills/): Reusable capabilities invoked via the Skill tool

The current behavior conflates these, leading to confusion about what was actually executed.

Error Messages/Logs

● Skill(/issue_create)
  ⎿  Successfully loaded skill · 1 tool allowed

This message is misleading - it loaded a command, not a skill.

Steps to Reproduce

1. Create a command file at .claude/commands/my_command.md
2. Start a Claude Code session
3. Ask Claude to invoke the command in a way that triggers Skill tool usage (e.g., indirectly reference the action)
4. Observe: Skill tool says "Successfully loaded skill" for a command

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

1.0.33

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

Related to #19290 which discusses command/skill confusion, but this issue specifically addresses the Skill tool's acceptance of commands without validation or error.

The distinction matters because:

• Commands have different invocation semantics (slash prefix)
• Skills have tool restrictions and other metadata
• Mixing them creates unpredictable behavior and confusing output