[BUG] Blank screen / rendering freeze when a slash command auto-expands into a long input

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?

Environment:

Claude Code VSCode native extension
OS: macOS (Darwin 25.5.0)
Extension version: (please fill in — Extensions panel → Claude Code → version)
Description:
I type a short message — typically 20-30 lines of requirements plus a pull request link. I'm invoking a custom slash command (/review) that auto-injects its full skill definition (several thousand lines of markdown) into the actual prompt sent to the model. It's this auto-expansion, not anything I personally paste, that produces the very long input. When that expanded input goes through, the UI streams a few words of the response and then goes completely blank. The session seems to still be alive underneath — it looks like a rendering/display bug in the extension's output panel, not a crash of the underlying process.

Steps to reproduce:

Set up a custom slash command (e.g. /review) whose skill file is large (thousands of lines).
Type a short invocation — e.g. 20-30 lines of requirements plus a PR URL — and submit.
The harness expands this into a full prompt that includes the entire skill body (thousands of lines) plus my short input.
Observe: response starts streaming (a few words appear), then the view goes blank.
This happens consistently whenever a command expands to a long input — not on plain short messages with no large auto-injected content.
Expected behavior: The response should stream and render normally regardless of how long the effective prompt is after slash-command expansion, or at minimum degrade gracefully rather than going blank.

Workaround found: I have not found or tried anything yet.

Additional context: Reproduced multiple times in the same session, each time correlated with a /review invocation whose expanded skill content ran to several thousand lines, even though my own typed input was short (20-30 lines + a link).

What Should Happen?

the input text window show shrink to a standard size and have a scroll bar for me to see the prompt.

Error Messages/Logs

Steps to Reproduce

Description:
I type a short message — typically 20-30 lines of requirements plus a pull request link. I'm invoking a custom slash command (/review) that auto-injects its full skill definition (several thousand lines of markdown) into the actual prompt sent to the model. It's this auto-expansion, not anything I personally paste, that produces the very long input. When that expanded input goes through, the UI streams a few words of the response and then goes completely blank. The session seems to still be alive underneath — it looks like a rendering/display bug in the extension's output panel, not a crash of the underlying process.

Claude Model

Sonnet (default)

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

anthropic.claude-code-2.1.263-darwin-arm64

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

VS Code integrated terminal

Additional Information

_No response_