TUI copy-paste polluted with box-drawing Unicode decorations (╎ ▎ ● │ ┃)

[eugenicum]

Problem

When selecting and copying text from the TUI terminal, decorative Unicode characters are included in the copied text. Example of what gets copied:

▎ Hi
… reasoning done · 0.9s
╎ The user just said "Hi"...
● Eugene, hi. What are we working on today?
┌─────────────────────────────────────────

These are NOT part of the model output — they are TUI rendering artifacts:

Char	Unicode	Source in code	Purpose
▎ ▏	U+258E/U+258F	history.rs:226 — '\u{258F}'	Conversation continuation rail
╎	U+254E	Thinking block rail	Thinking/reasoning prefix
●	U+25CF	history.rs:867 — tool status bullet	Tool success indicator
│	U+2502	history.rs:1758 — card rail	Tool card border
┌ └	U+250C/U+2514	Tool card borders	Card frame

Impact

Copy-paste from the TUI is unusable — every copied line contains visual noise that must be manually stripped. This affects anyone who wants to share or save model output.

Relevant source

crates/tui/src/tui/history.rs already tracks:

• rail_prefix_widths — display-column widths of rail prefixes
• copy_prefix_widths — visual prefixes that should be stripped from clipboard
• CopyLineSeparator — preserves hard newlines vs soft wraps

The plumbing for clean copy exists — it just needs to be wired into the clipboard/selection path.

Suggested fix

When copying selected text to clipboard, strip the rail prefix characters (based on rail_prefix_widths / copy_prefix_widths) so the clipboard gets plain text matching what the model actually output. Alternatively, add a rail_style = "ascii" / "off" setting for users who prefer plain rendering.

[github-actions]

Thanks @eugenicum for the report.

This issue is staying open for maintainer triage. CodeWhale gets better because people bring us real edge cases from real machines, providers, regions, and workflows.

If you can add a reproduction, logs, version output, screenshots, or the provider/model involved, that makes it much easier for us to verify and harvest the fix. Maintainers may comment /lgtmi to mark recurring issue reporters as approved so this intake note is skipped next time.

[Hmbown]

Fixed in the v0.8.68 candidate. In-app transcript selection now uses explicit copy-prefix metadata, so user/assistant/thinking/tool-card decorations (▎ ▏ ╎ │ ┃ ╭ ╰ ●) are removed without guessing from glyphs or damaging legitimate code/CJK text. The focused #4208 regression passes on current main. Terminal-native selection used with mouse capture disabled remains controlled by the terminal and cannot be rewritten by Codewhale; use the in-app drag selection and copy path for clean text. Thank you @eugenicum for identifying the existing copy metadata and listing the affected rails.