[BUG] Tool use rejection feedback entered in the permission prompt does not appear to influence the model's next turn

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?

When the CLI prompts for permission to use a tool, the third option labelled along the lines of "No, and tell Claude what to do differently" lets the user type free-text feedback before confirming the rejection. The expectation is that this feedback is delivered to the model as part of the rejected tool's tool_result so it can adjust its next turn. In practice the model behaves as if the rejection had been a bare "no" with no feedback attached.

What Should Happen?

The user-supplied feedback should reach the model in the next request as the content of the rejected tool's tool_result block and the model's next turn should visibly take it into account.

Error Messages/Logs

Steps to Reproduce

1. Ask Claude to do something that triggers a tool that isn't auto-allowed (e.g. Bash, an Edit, etc.).
2. At the permission prompt, focus the "No" option and press Tab to enter feedback mode.
3. Type something specific the model could act on, e.g. "don't use grep, use ripgrep instead".
4. Submit.
5. Observe the model's next assistant turn: it does not reflect the feedback; typically, it retries the exact same tool call, or moves on as if no instructions were given.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.148

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Other

Additional Information

_No response_