[Bug] Vietnamese IME input drops characters in terminal (Unikey/ibus-unikey)

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?

---
Title:
[Bug] Vietnamese IME input drops characters in terminal (Unikey/ibus-unikey)

---
Body:

## Bug Description

When typing Vietnamese using an IME (e.g., Unikey on Windows/macOS, ibus-unikey on Linux),
characters are randomly dropped from the input. The IME uses backspace (0x7F / DEL) to
replace raw keystrokes with composed Vietnamese characters, but Claude Code's input handler
processes the backspaces without inserting the replacement text.

## Steps to Reproduce

1. Install a Vietnamese IME (Unikey, ibus-unikey, or similar) (I am using OpenKey app)
2. Set input method to Telex or VNI
3. Open Claude Code in terminal
4. Type any Vietnamese sentence, e.g.: check dùm tôi các lỗi sau

## Expected Behavior

check dùm tôi các lỗi sau

## Actual Behavior

check d tôi c l sau

Characters like ùm, ác, ỗi are dropped — only the first keystroke survives.

## Root Cause

Vietnamese IMEs compose characters by sending:

1. Raw keystrokes (e.g., d, u, m)
2. Multiple DEL/backspace bytes (0x7F) to erase them
3. The final composed character (e.g., dùm)

Claude Code's input handler processes the DEL bytes (deleting characters from the buffer)
but does not insert the replacement text that follows, causing characters to disappear.

The affected code pattern in cli.js:
```js
if(input.includes("\x7f")) {
// handles backspaces but discards the replacement text
}

Environment

• OS: macOS / Windows / Linux
• IME: Unikey (macOS/Windows), ibus-unikey (Linux)
• Input method: Telex / VNI
• Claude Code version: latest (@anthropic-ai/claude-code)
• Node.js version: (run node --version)
• Terminal: iTerm2 / Terminal.app / Windows Terminal / GNOME Terminal

Workaround

A community patch exists that fixes this by applying both the backspaces and
the replacement text:
https://github.com/manhit96/claude-code-vietnamese-fix

Fix Suggestion

When the input buffer contains DEL bytes (0x7F), the handler should:

1. Apply N backspaces to remove N DEL chars
2. Insert the remaining characters in the buffer as replacement text

if (input.includes("\x7f")) {
const n = (input.match(/\x7f/g) || []).length;
const replacement = input.replace(/\x7f/g, "");
let state = curState;
for (let i = 0; i < n; i++) state = state.backspace();
for (const c of replacement) state = state.insert(c);
if (!curState.equals(state)) {
if (curState.text !== state.text) updateText(state.text);
updateOffset(state.offset);
}
return;
}

This affects all Vietnamese users of Claude Code (~100M Vietnamese speakers).

What Should Happen?

When typing Vietnamese using an IME (Telex/VNI), the full composed text should appear correctly in the input buffer.

Example: typing "check dùm tôi các lỗi sau" should display exactly:
check dùm tôi các lỗi sau

Error Messages/Logs

Steps to Reproduce

1. Install a Vietnamese IME (Unikey on macOS/Windows, or ibus-unikey on Linux)
2. Enable Telex or VNI input method
3. Open Claude Code in terminal: claude
4. Type any Vietnamese sentence using the IME, e.g.: check dùm tôi các lỗi sau

No additional code or files needed — the bug occurs directly in the Claude Code
input prompt.

Claude Model

Sonnet (default)

Is this a regression?

Yes, this worked in a previous version

Last Working Version

Claude Code v2.1.63

Claude Code Version

Claude Code v2.1.63

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_