Copying from terminal output silently drops a chunk of characters starting at a non-ASCII character

Description

When copying text from Claude Code's output in the terminal (both via select-to-copy and manual Cmd+C copy), a contiguous chunk of characters is sometimes silently dropped from the copied text. The user has hit this multiple times; this report documents one fully verified occurrence.

Notably, the dropped span starts exactly at a non-ASCII character (Polish diacritic ą, U+0105), which suggests the bug may be related to multibyte/wide character handling in the TUI rendering or the selection/copy path.

Environment

• Claude Code: 2.1.224
• OS: macOS 15.6.1 (Darwin 24.6.0, arm64)
• Terminal: Apple Terminal (TERM_PROGRAM=Apple_Terminal, TERM=xterm-256color)

What happened

Claude Code rendered a Polish-language text inside a fenced code block. The user copied it from the terminal and pasted it elsewhere. The pasted text was missing exactly 26 characters in the middle of a line.

Original line as rendered by Claude Code:

zaplanować na lawetę. Oglądanie i zakup w Kiekrzu k. Poznania — na
miejscu jest pracownik upoważniony do wszystkich formalności.

Pasted result from the clipboard:

zaplanować na lawetę. Ogl. Poznania — na
miejscu jest pracownik upoważniony do wszystkich formalności.

The dropped span is:

ądanie i zakup w Kiekrzu k

Verification

A character-level diff (Python difflib.SequenceMatcher over the original file vs. the pasted clipboard content) shows the two texts (2081 vs. 2055 chars) are identical except for a single contiguous deletion of 26 characters at position 1977–2003. The first lost character is ą = U+0105 (LATIN SMALL LETTER A WITH OGONEK).

The rest of the ~2 kB text — including many other Polish diacritics (ł, ś, ż, ę, ó, č), em dashes, en dashes and typographic quotes — survived the copy intact, so this is not a general encoding problem but a localized drop.

Why this is nasty

The truncation is silent and mid-sentence, so it's easy to miss — the user only noticed after pasting the text into a real document. With non-English content this can corrupt meaning without any visual error indication.

Steps to reproduce (best effort)

1. Have Claude Code output a longer (~2 kB) block of text containing non-ASCII characters (e.g. Polish) inside a fenced code block, in Apple Terminal on macOS.
2. Select the output text in the terminal and copy it (the user reports it happens both with select-to-copy and manual copy).
3. Paste and diff against the original — occasionally a chunk of characters beginning at a non-ASCII character is missing.

The issue is intermittent — it does not reproduce on every copy, but the user has observed it repeatedly across sessions.