[BUG] Claude treats user instructions as tool output

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?

This issue only occurs in the new VSCode extension; the CLI works as expected.

Before calling a tool, Claude prompts the user either for permission or for instructions on what to change. However, when I enter text into the "what to do differently" field during a tool call, Claude mistakenly interprets my input as the tool’s output.

Instead of applying my input to adjust the tool call, it responds with something like:

“The tool returned ...my input...”

What Should Happen?

It should treat my input as further instructions, not as the output of the tool call it intended to make.

Error Messages/Logs

Steps to Reproduce

(In Claude VSCode extension)

1. Make Claude attempt to call a tool.
2. When asked by Claude if calling the tool is OK, write "don't call the tool" in the "Tell Claude what to do instead" text box.

You will see that "don't call the tool" is considered by Claude as the output of the tool call.

Claude Model

Sonnet (default)

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

v2.0.1

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Cursor

Additional Information

<img width="636" height="186" alt="Image" src="https://github.com/user-attachments/assets/76e7ffc2-e554-4ed1-90c4-08ee405fcc62" />