[BUG] Bare skill-only slash command as first message in a new session sometimes hangs indefinitely (never reaches "Thinking...")

### Preflight Checklist

- [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) 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 client cycles through filler/loading words ("Clauding...", "Envisioning...", etc.) but never progresses to "Thinking..." or actually starts the skill's work. It appears to hang indefinitely. Might be particular to the "/code-review" skill, though that is the skill I most frequently use.

### What Should Happen?

The skill should invoke and begin work within a few seconds, as it does when referenced in a normal prose message.

### Error Messages/Logs

```shell

```

### Steps to Reproduce

1. Start a brand new session (no prior messages).
2. As the very first message, type only a skill invocation, e.g. /code-review, with no surrounding text.
3. Observe the output.

### Claude Model

Sonnet (default)

### Is this a regression?

I don't know

### Last Working Version

_No response_

### Claude Code Version

2.1.112

### Platform

Anthropic API

### Operating System

Windows

### Terminal/Shell

Other

### Additional Information

In VS Code "panel", not terminal

Start by reproducing a bare /code-review invocation as the first message in a new session, specifically in the VS Code panel on Windows, and compare it with the same skill referenced in normal prose. Trace the slash-command entry point and determine why the request never reaches "Thinking..."; done means the skill begins work within a few seconds without requiring surrounding text.

Written by the indexing model from the issue text.