[BUG] "Word (By Anthropic)" extension: replace_text always fails with AppleScript syntax error

### 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 replace_text function of the "Word (By Anthropic)" desktop extension always fails. It generates malformed AppleScript regardless of the search text or the "all" parameter — Word rejects it with a syntax error because the "replace" parameter is set twice in the execute command.

### What Should Happen?

replace_text should find and replace the given text in the active Word document without error.

### Error Messages/Logs

```shell
execute find findObject replace replace all replace replace all
→ syntax error: The «class 5642» parameter is specified more than once. (-2750)
```

### Steps to Reproduce

1. Open a .docx file with the "Word (By Anthropic)" extension (open_document)
2. Call replace_text with any find/replace string, e.g. find: "test", replace: "Test"
3. Observe AppleScript syntax error every time, even for a single plain word

### Claude Model

Sonnet (default)

### Is this a regression?

Yes, this worked in a previous version

### Last Working Version

_No response_

### Claude Code Version

Claude Desktop 1.52386.3

### Platform

Other

### Operating System

macOS

### Terminal/Shell

Other

### Additional Information

This occurs via the "Word (By Anthropic)" local MCP extension in Claude Desktop, not the Claude Code CLI. Device: macOS, arm64.

Start with the Word (By Anthropic) extension's replace_text entry point and reproduce the call with find "test" and replace "Test" in an active Word document. Inspect the generated AppleScript command; done means it no longer specifies the replace parameter twice and the text replacement succeeds without a syntax error.

Written by the indexing model from the issue text.