Text output is centered AND truncated on Windows

[m13v]

Text in responses is being centered instead of left-aligned, AND long lines are truncated/cut off instead of wrapping properly.

Problems

1. Centered text: Content appears centered in the terminal instead of left-aligned
2. Truncation: Long lines are cut off rather than wrapping to the next line

Environment

• Windows 11
• Claude Code v2.0.73
• Tested on: PowerShell, CMD, Windows Terminal, Git Bash
• Same behavior across all terminals

What I see

• Left side has UI components properly aligned
• Text descriptions/content are centered in the middle of the terminal
• Long lines get truncated at some arbitrary width instead of wrapping
• Makes output hard to read and information is lost

Expected

• All text should be left-aligned like normal CLI output
• Long lines should wrap to the next line, not be truncated

Screenshot attached shows the centering issue - notice how the text is floating in the center while the UI chrome stays left.

[rdfitted]

I'm experiencing this too!

There is also missing text if you look at the screenshots

[rdfitted]

I've noticed that it's only on final output. Not on code snippets, or reasoning steps.

[jenko]

Same here, Windows 11, git bash

[SherlockianAsh]

I'm having the same issue too. Already trouble shoot with PowerShell, VS Code terminal, Claude Code extension, and CMD

[bahayonghang]

Same here on Windows (Claude Code v2.0.73): the main response text is centered with large blank margins, and parts of the output become invisible or clipped. As a workaround, I removed the npm-based installation and reinstalled Claude Code using the official native installer scripts, pinning to version 2.0.65. For Windows PowerShell, I used the documented install.ps1 entry point to install that specific version, and after installing v2.0.65 via the native installer the output renders normally again in Windows Terminal with PowerShell 7.​

[m13v]

my workaround is a claude md rule to always output response to a specific text file, haha

On Thu, Dec 18, 2025 at 10:33 PM 李永航 @.***> wrote:

bahayonghang left a comment (anthropics/claude-code#14581) https://github.com/anthropics/claude-code/issues/14581#issuecomment-3673742876 image.png (view on web) https://github.com/user-attachments/assets/27facd91-e17c-4f76-b2a2-a2159816b6af

Same here on Windows (Claude Code v2.0.73): the main response text is centered with large blank margins, and parts of the output become invisible or clipped. As a workaround, I removed the npm-based installation and reinstalled Claude Code using the official native installer scripts, pinning to version 2.0.65. For Windows PowerShell, I used the documented install.ps1 entry point to install that specific version, and after installing v2.0.65 via the native installer the output renders normally again in Windows Terminal with PowerShell 7.​

— Reply to this email directly, view it on GitHub https://github.com/anthropics/claude-code/issues/14581#issuecomment-3673742876, or unsubscribe https://github.com/notifications/unsubscribe-auth/AY62CDFGPMIGA6GS2IQB43D4COL3JAVCNFSM6AAAAACPP625KGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTMNZTG42DEOBXGY . You are receiving this because you authored the thread.Message ID: @.***>

[bahayonghang]

my workaround is a claude md rule to always output response to a specific text file, haha …

Because I don’t need the features from the latest few versions for now, I’ve decided to roll back to version 65 after checking the changelog, and I’ll update again once the issue is fixed.

[solosza]

im having the same issue. it started about 6pmish pacific.

[ilkerarslan]

Same here

[solosza]

Same here

a commenter above suggested a workaround where you rollback to v2.0.65 for now until a fix comes out. i just did it and i no longer see the issue.

[charmenlondon-cmd]

Had the same issue when my instance auto updated to v2.0.73. Rolled back to v2.0.72 and it's working ok.

[deadeuzesse]

Same, in 2.0.73 also Downgraded to 2.0.71 as a workaround

[david-cilluffo]

I had the same issue. I rolled back to 2.0.72 and it's working again:

npm install -g @anthropic-ai/[email protected]

[Niels-LNS-Research]

Before you downgrade, verify if you run Claude Code natively by running:

claude doctor

If you are running npm, then following the instructions above that other users have recommended. If you are running natively, then follow these instructions:

# 1) Close any running Claude Code sessions first.

# 2) Install a specific version (2.0.72)
& ([scriptblock]::Create((irm https://claude.ai/install.ps1))) 2.0.72

# 3) Verify
claude -v
claude doctor

You should see something like this:

 Diagnostics
 └ Currently running: native (2.0.72)
 └ Path: native
 └ Invoked: C:\Users\xxxxxxxxxx\.local\bin\claude.exe
 └ Config install method: native
 └ Auto-updates: enabled
 └ Search: OK (bundled)

You then have to disable auto-updates.

Option A — disable for this PowerShell session only

$env:DISABLE_AUTOUPDATER = "1"
claude

Option B — disable persistently for your Windows user

[Environment]::SetEnvironmentVariable("DISABLE_AUTOUPDATER","1","User")
# restart PowerShell after this

Option C — disable via Claude Code user settings (recommended if you already manage settings there)

User settings file is ~/.claude/settings.json (Windows: $env:USERPROFILE.claude\settings.json). Claude

Add/merge:

{
  "env": {
    "DISABLE_AUTOUPDATER": "1"
  }
}

[nsxdavid]

I truly don't mean to be obnoxious, and I know we are all AI vibin' these days, but some old-school software engineering techniques such as unit tests, regression tests and such still hold their same value. And I know it's not the cool Silicon Value way of doing things, but maybe have at least some team members rock Windows instead of a Mac or Linux so as to catch these things before the rest of the world.

[jenko]

I truly don't mean to be obnoxious, and I know we are all AI vibin' these days, but some old-school software engineering techniques such as unit tests, regression tests and such still hold their same value. And I know it's not the cool Silicon Value way of doing things, but maybe have at least some team members rock Windows instead of a Mac or Linux so as to catch these things before the rest of the world.

Yeah, I had the same thought, kind of shocked that a version with such an intrusive error got through testing.

[Niels-LNS-Research]

I truly don't mean to be obnoxious, and I know we are all AI vibin' these days, but some old-school software engineering techniques such as unit tests, regression tests and such still hold their same value. And I know it's not the cool Silicon Value way of doing things, but maybe have at least some team members rock Windows instead of a Mac or Linux so as to catch these things before the rest of the world.

There must have been 0 testing on Windows prior to release. Quite shocking, given that there is even a native Windows build. The developers wouldn't even need to run Windows to catch this; it could just be done as an automated build test.