claude-code - 💡(How to fix) Fix Session-summary text renders garbled (box-drawing/Cyrillic glyphs) in Cursor terminal — Claude Code 2.1.142/2.1.143

Preflight Checklist

• I have searched existing issues and this hasn't been reported yet
• This is a single bug report (please file separate reports for different bugs)
• I am using the latest version of Claude Code

What's Wrong?

Random letters in the session-summary block (the "Worked for Ns / Pushed N commits" status text shown after a turn) render as box-drawing and Cyrillic-looking glyphs instead of the intended ASCII characters. Same word renders differently in different positions, so it isn't a simple 1-to-1 font fallback — looks like corruption in the emitted byte stream or in the styled-text rendering path.

First seen: Week of 11–16 May 2026. Same machine + terminal setup has run cleanly for months. Correlates with the rapid release cadence this week (2.1.139 → 2.1.143 in five days).

Affects multiple users: Confirmed on at least two independent machines (myself plus another member of my group), so it isn't local config.

I also submitted this via the in-app /bug channel earlier in the week. Still seeing it on latest, so filing publicly so it's tracked.

See screenshot attached for the visible corruption.

What Should Happen?

Plain ASCII text in the session-summary block, like this:

Published. Pushed 4 commits (821914e19..d7329554c) + isbox hint (655215f08) to origin/main. Pushed 2 handoffs (#1r1, #1r — both Phase b local-LLM migrations; Y only). Total P: 16 files. Updating files left for now: .

Instead, letters u, s, e, o (inconsistently) are replaced with г, └, ┘, ┐, etc. The same letter in the same word renders differently between occurrences, which rules out a clean font-substitution.

Error Messages/Logs

Steps to Reproduce

1. Launch claude inside Cursor's integrated terminal (TERM_PROGRAM=vscode).
2. Run any task that produces the end-of-turn summary block — e.g. a /ship or /h2c style command, anything that triggers the "Pushed N commits / Worked for Ns" output.
3. Observe garbled letters in the summary text.

The corruption is in the rendered summary text only. Other output (tool results, regular agent prose) renders cleanly.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

2.1.138

Claude Code Version

2.1.143

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Cursor

Additional Information

• Claude Code: 2.1.143 (also reproduced on 2.1.142)
• Terminal: Cursor integrated terminal — TERM_PROGRAM=vscode, TERM_PROGRAM_VERSION=3.4.17
• TERM: xterm-256color
• COLORTERM: truecolor
• Locale: en_US.UTF-8 across all LC_*
• OS: macOS 15.7.7 (Darwin 24.6.0)
• Hardware: Apple Silicon — M3 Ultra (Mac Studio)
• Node: installed via npm-global

Hypothesis / extra context: Reproduces in Cursor's xterm.js terminal renderer. I haven't yet confirmed whether iTerm2 / Apple Terminal show the same corruption — if they don't, that suggests Claude Code is emitting Unicode codepoints (e.g. styled / mathematical-italic variants) that xterm.js doesn't fall back gracefully on. If iTerm2 also shows it, the cause is in the byte stream itself.

Workaround: Pinning back to @anthropic-ai/[email protected] appears to avoid it (needs confirmation).