Actual ncurses output in T3 loses the underline beneath the focused block cursor. Native Ghostty 1.3.1 keeps it. A wide glyph loses underline across both columns covered by the cursor. Moving the cursor away restores the underline.
Independently confirmed by Claude Opus 5.5 at medium effort in a fresh read-only worktree on current main 834eb51301a447e6f4a8f5f517fb4eb75e00a783. Hunter used 1b14aade4de1db0c82b8d61ccfd9c011b04c3810; the owning renderer is unchanged between those commits and final fetched main.
Both screenshots use the same Python ncurses producer and TERM=xterm-256color. Mode f puts the cursor on the first glyph of the bold, italic, underlined row. The second copy stays away from the cursor. Crops retain all attribute rows.
apps/web/src/terminal/ghostty/renderer.ts:345–365 draws supported cell decorations. Line 385 fills the block cursor over them. Lines 391–398 redraw only the glyph using the cell’s bold and italic font, without restoring underline in the contrasting cursor text colour. The existing underline attribute and native comparison establish the contract; this requests no new shaping or cursor style.
| Cases | Result |
|---|---|
| Bold, italic, bold + italic, bold + faint, wide bold | Glyphs survive cursor redraw |
| Italic + faint + underline; bold + italic + underline | Underline lost inside block cursor |
| Wide italic + underline; wide bold + faint + underline | Underline lost across the two covered columns |
| Faint contrast under cursor | Native also changes contrast; no separate defect claimed |
Frequency is conditional: the loss occurs when the focused block cursor rests on an underlined cell. It is a small visual defect affecting one cell, or two for a wide glyph. Field prevalence is unknown. Two hunter sessions and one independent verifier session reproduced the visual result. No claim across every font or colour configuration. Bar, underline and unfocused outline cursor branches are outside this finding.
Linux Chromium real integrated PTY and native Linux Ghostty only. No Electron, mobile, macOS, Windows, physical-device or provider claims. Strikethrough, overline and hovered-link decoration share relevant source paths but were not tested and are not findings. No password or concealment claim.
Both-state issue and eligible PR searches for block cursor, terminal italic/bold/wide/underline and cursor decoration found no exact covering owner. Conceal #37141 respects invisible flags; this producer sets none and tests visible underline. Wide glyph #27948 owns cell-grid alignment, which passes here. Previously passed normal inverse/faint/underline and Unicode edit controls were not reported anew. Unvouched submissions are not coverage.
Pure visual captures are the proof; no snapshot UI test was added. The existing public core style/wide-cell test passes. Initial local missing dependencies were corrected once with the normal install. First native comparison used its default TERM, then both were rerun after reset with matching TERM. No harness failure was promoted.
Hunter VM bx_zsknzy3e and verifier VM bx_u2zqern9 were separately held with TTL 7200, commands serialized through exec --vm and reset before proof. Both were released before their reports. All apps, native controls, servers and captures ran on boat. No fixes, filings, product edits, commits or pushes. Tracked checkout clean.
Report by GPT 6.1 Sol through Codex; independent verification by Claude Opus 5.5 through Claude Agent. PostPlan uses fallback true-black colours because the active host theme could not be read.