AI Credit Usage for uninstalling a plugin.

### Describe the bug

Today I removed the superpowers plugin from the superpowers marketplace using `/plugin rm superpowers@superpowers-marketplace` and in order to do so, it needed to read the plugin help, and convert rm alias to the uninstall command.

I don't understand why this is not just an alias that performs the exact same command.

Instead it used 8.83 AI Credits. 0.126% of my 7000 allowance.

### Affected version

GitHub Copilot CLI 1.0.68

### Steps to reproduce the behavior

With the plugin installed:
/plugin rm superpowers@superpowers-marketplace

Watch it think:

```
Thought for 1s
The user is trying to uninstall a plugin called "superpowers@superpowers-marketplace" and I need to figure out what CLI command handles plugin removal in this copilot system.

Shell Remove superpowers plugin 2 lines…
copilot plugin rm superpowers@superpowers-marketplace 2>&1

Thought for 1s
I'm wondering if "plugin rm" is actually a subcommand within the Copilot CLI's interactive session rather than something to execute directly through PowerShell. Let me check the help documentation to clarify.

$ Shell Check plugin command help 28 lines…
copilot plugin --help 2>&1
$ Shell Uninstall superpowers plugin 2 lines…
copilot plugin uninstall superpowers@superpowers-marketplace 2>&1
```

Be shocked and amazed at the ridiculousness of this.

### Expected behavior

Use NO input tokens. It is the /plugin command, why does it need to send that to the LLM?

### Additional context

Claude Sonnet 5 (medium) was the selected model.

I had used /plugin list to allow me to copy-paste the plugin name, so it is possible some of the credit usage was from this.

No source file or test is named. Reproduce `/plugin rm superpowers@superpowers-marketplace` with the plugin installed, compare it with `/plugin uninstall`, and trace where the alias is interpreted and where model calls are made. Done means the removal completes without unnecessary help lookup or LLM input, with tests covering the behavior.

Written by the indexing model from the issue text.