Korean IME composition window appears at wrong position (bottom-left of terminal)

[subinium]

Korean IME composition window appears at wrong position (bottom-left of terminal)

Summary

When typing Korean (or other CJK languages) in Claude Code, the IME composition/candidate window renders at the bottom-left corner of the terminal instead of at the cursor position. This makes Korean input extremely difficult — the composing character appears far from where text is being typed.

Environment

• OS: macOS (Darwin 25.2.0)
• Claude Code model: Opus 4.6
• Input method: macOS default Korean input (2-set / du-beol-sik)
• Terminal: iTerm2 / Terminal.app (reproducible in both)

Steps to Reproduce

1. Launch Claude Code in a terminal on macOS.
2. Switch to Korean input method.
3. Type Korean text in the input prompt (e.g., keystrokes for "한글").
4. Observe where the IME composition window appears.

Expected Behavior

The IME composition window should appear directly adjacent to the text cursor in the input field.

Actual Behavior

The IME composition window appears at the bottom-left corner of the terminal, completely detached from the input area:

1. Split attention: The composing syllable (ㅎ → 하 → 한) appears far from where text is inserted.
2. Disorienting feel: Characters feel "pushed away" or "lagging" due to spatial disconnect.
3. Affects all Korean input: Korean users all experience this.

Screenshot

Technical Analysis

Root Cause

Claude Code uses React Ink, which hides the real terminal cursor and renders a fake cursor via chalk.inverse(). macOS IME relies on the real terminal cursor position to place the composition window. Since the real cursor is hidden, the IME window falls back to position (0,0).

The Fix: Ink's useCursor API

Ink 6.7.0 introduced useCursor (PR #866) which moves the real terminal cursor to specified coordinates after each render:

import { useCursor } from 'ink';
import stringWidth from 'string-width';

const { setCursorPosition } = useCursor();
setCursorPosition({
  x: stringWidth(prompt + textBeforeCursor),
  y: inputRow,
});

Local verification: Tested with ink@6.7.0 + useCursor — IME composition window appears at the correct position. Korean input works as expected.

Related Issues

• [BUG] IME input causes performance issues and duplicate conversion candidates #1547, IME Composition Failure for Korean Character Input #2620, Investigation: Fixing IME Issues in Claude Code by Patching React Ink / React Inkのパッチによる Claude Code のIME問題修正について #3045, [BUG] CJK IME (Chinese/Japanese/Korean) preedit & candidate window anchored to wrong position on macOS (real cursor not moved) #16372, Feature Request: IME cursor position support for CJK input (Japanese/Chinese/Korean) #19207, [BUG] Korean IME composition fails in Quick Launcher (Option+Space) on macOS #21382

Proposed Fix

1. Integrate useCursor from ink >=6.7.0 into Claude Code's text input component
2. Use string-width for CJK-aware cursor column calculation
3. Show the real terminal cursor alongside the existing visual cursor

Upstream PRs

• ink-text-input #93: Add IME cursor positioning via useCursor
• @inkjs/ui #24: Add IME cursor positioning to TextInput
• ink #876: Improve cursor-ime example with full TextInput
• ink #866: useCursor hook (merged)
• ink #872: <Cursor> component (open)

Reproduction

Available at korean-terminal:

Script	Description	IME behavior
npm run test1	Fake cursor (current behavior)	IME at bottom-left (broken)
npm run test2	useCursor with real cursor	IME at correct position (fixed)
npm run test3	Full TextInput + useCursor	IME at correct position (fixed)

[github-actions]

Found 3 possible duplicate issues:

1. [BUG] CJK IME (Chinese/Japanese/Korean) preedit & candidate window anchored to wrong position on macOS (real cursor not moved) #16372
2. Feature Request: IME cursor position support for CJK input (Japanese/Chinese/Korean) #19207
3. [BUG] Chinese Input Method Candidate Window Not Following Cursor Position #6342

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

[subinium]

This is not a duplicate — it is a consolidation with a concrete fix.

The referenced issues (#16372, #19207, #6342) were all closed, but the underlying problem still exists as of today. The IME composition window still appears at the bottom-left corner of the terminal when typing Korean in Claude Code.

What makes this issue different

1. Reproduction repo: subinium/korean-terminal — three runnable demos (npm run test1/test2/test3) showing the broken vs fixed behavior
2. Root cause analysis: Traced to Ink hiding the real terminal cursor; IME falls back to position (0,0)
3. Upstream PRs submitted to fix this at the source:
   • ink-text-input#93 — useCursor integration
   • @inkjs/ui#24 — useCursor integration
   • ink#875 — improved cursor-ime example
4. Verified fix: useCursor (ink >=6.7.0) + stringWidth correctly positions the IME window

Please keep this open as the canonical tracking issue for CJK IME positioning.

[myriky]

Reproduction video (macOS, iTerm2, Korean IME)

I recorded a short screen capture demonstrating this issue.

korean.character.IME.issue.mov

Environment:

• macOS Sequoia 15.5
• iTerm2
• Claude Code v2.1.42
• Korean (2-Set) IME

What's happening:

• While composing Hangul characters, the IME candidate window appears at the bottom-left corner of
  the terminal instead of next to the actual cursor position.
• The composing jamo (ㄱ, ㅏ, etc.) flicker and split apart during input, making it very
  disorienting — almost like audio feedback/howling on a speakerphone call.
• This makes Korean input practically unusable without resorting to copy-paste workarounds.

Looking forward to the useCursor fix landing. Thanks for tracking this!

[gonnector]

Cross-comparison: Gemini CLI and Codex CLI handle Korean IME correctly

Adding a data point from Windows (MSYS2/bash, Windows Terminal).

Environment:

• Windows 11 (26200)
• Windows Terminal + bash (MSYS2)
• Claude Code v2.1.50
• Korean (2-Set) IME

Comparison with other AI CLI tools:

CLI Tool	IME Inline Composition	Jamo visible at cursor
Claude Code	No — composition window at bottom-left (0,0)	No
Gemini CLI	Yes — inline at cursor position	Yes
OpenAI Codex CLI	Yes — inline at cursor position	Yes

Both Gemini CLI and Codex CLI show real-time jamo composition (ㅎ → 하 → 한) directly at the input cursor. Claude Code is the only one among the three where the composition falls to the bottom-left corner.

Why this matters for prioritization:

This is not an edge case — it affects every Korean-speaking user on every session. Competing CLI tools have already solved this, which makes the UX gap very noticeable in daily use. Gemini CLI addressed this by forking Ink and adding terminalCursorPosition support (google-gemini/gemini-cli#17711). The upstream useCursor hook (ink >=6.7.0) provides a non-fork path to the same fix.

Looking forward to this being resolved. Thanks to @subinium for the thorough root cause analysis and upstream PRs.

[bibitto]

Japanese IME too

[moonward]

I can confirm this issue also affects Japanese IME input on WezTerm on linux — a terminal emulator with first-class IME support.
Vim and other TUI apps handle inline Japanese input correctly in WezTerm, but Claude Code does not. This makes the problem clearly Claude Code-specific, not a terminal limitation.

I want to also flag a concerning pattern in how this has been handled:

• Feature Request: IME cursor position support for CJK input (Japanese/Chinese/Korean) #19207 (the primary feature request with a working fix proposal) was closed with a reference to Feature Request: Add /reload command to refresh configuration without restarting session #17127 as "fixed" — but Feature Request: Add /reload command to refresh configuration without restarting session #17127 is about adding a /reload command for config files. It has absolutely nothing to do with IME input. This was either a mistake or the issue was closed without being read.

The fix is well-understood and already available: useCursor from ink >= 6.7.0 has been merged upstream. Competing CLI tools (Gemini CLI, OpenAI Codex CLI, even Open Code) have already solved this. Claude Code is the outlier.

Please don't close this as a duplicate of already-closed issues that were themselves closed incorrectly. This bug is still present and affects all CJK users daily.

[HIHOOOOO]

Reproduction on Windows 11 + Windows Terminal

Environment:

• OS: Windows 11 Home
• Claude Code: 2.1.55
• Terminal: Windows Terminal 1.23.x
• Shell: PowerShell 7
• Input Method: Windows Korean IME (2-set)

Observed behavior:

Korean IME composition characters appear at the bottom of the Claude Code UI, outside the input area — identical to the behavior described in this issue. The composing syllable is spatially disconnected from the cursor position, making Korean input disorienting.

Screenshot showing Korean characters leaking to the bottom of the UI:

The fix proposed here (integrating useCursor from ink 6.7.0) would resolve this on Windows as well. Upvoting for prioritization.

[ilaikim99]

Claude Code's React Ink TUI hides the real terminal cursor and replaces it with a styled fake one — so the OS IME has no cursor position to anchor to and falls back to (0,0). The dev fix is the useCursor hook from Ink 6.7.0. User workarounds + upstream PR links: https://cacheoverflow.dev/blog/M5yrbtvz