What Problem This Solves

wrapNoteMessage (packages/terminal-core/src/note.ts) wraps note/diagnostic text to a visible-column budget — every fit decision in wrapLine is made with visibleWidth(). But its splitLongWord fallback (used when a single word is longer than the line budget, e.g. an unbroken CJK run) sliced the word into groups of maxLen code points via Array.from(word). Since maxLen is a visible-column budget and wide characters (CJK / fullwidth / emoji) are 2 columns each, a wide-char run produced lines up to twice the budget — e.g. a 24-character CJK word at maxWidth: 20 emitted a single 40-column line.

Why This Change Was Made

splitLongWord was the lone place that measured in code points while the rest of wrapLine measures visible columns. The fix accumulates grapheme visible width (the same splitGraphemes/visibleWidth decomposition the module already uses) and starts a new segment when the next grapheme would exceed maxLen, so every wrapped segment fits the budget. A single grapheme wider than the budget still emits on its own (preserving loop progress). ASCII wrapping is byte-identical.

User Impact

Note bodies containing long wide-character runs (CJK text, emoji) without separators now wrap within the requested column width instead of overflowing to ~2× and breaking terminal alignment / re-wrapping. No API change.

Evidence

oxfmt, oxlint, tsgo:core pass; the full packages/terminal-core/src/table.test.ts suite passes with a new CJK regression test, and reverting the fix makes that test fail (fail-before).

Real behavior proof

Behavior addressed: wrapNoteMessage emitted wrapped lines up to twice the visible-column budget for long wide-character words, because splitLongWord split by code-point count rather than visible width.

Real environment tested: Ran the exported wrapNoteMessage and visibleWidth from packages/terminal-core/src/{note,ansi}.ts via tsx on Node v22.22.1 (no mocks), before and after the patch, plus the colocated Vitest suite via node scripts/run-vitest.mjs.

Exact steps or command run after this patch: wrapNoteMessage("東京特許許可局長今日休暇許可局長今日休暇東京特許", { maxWidth: 20, columns: 80 }).split("\n"), asserting visibleWidth(line) <= 20 for each line; then node scripts/run-vitest.mjs packages/terminal-core/src/table.test.ts, node scripts/run-oxlint.mjs, pnpm tsgo:core.

Evidence after fix:

BEFORE (origin/main):
  wrapNoteMessage(<24 CJK chars>, {maxWidth:20}) -> line[0] visibleWidth=40, line[1]=8   (40 > 20 budget)
AFTER:
  wrapNoteMessage(<24 CJK chars>, {maxWidth:20}) -> line[0]=20, line[1]=20, line[2]=8     (all <= 20; join("") == input)
  ASCII control ("a".repeat(70), maxWidth 20)    -> unchanged (lines <= 20, as before)
table.test.ts -> exit 0 (all pass, incl. new CJK regression test); fail-before (revert source) -> the new test fails
oxlint -> exit 0 ; tsgo:core -> exit 0

Observed result after fix: every wrapped line of the CJK input is ≤ 20 visible columns (the long word breaks into 3 lines instead of one 40-column line), content is preserved (join("") equals the input), and ASCII wrapping is unchanged.

What was not tested: the actual @clack/prompts terminal render of a note (no interactive terminal in this environment); the fix is a pure wrapping-math change proven at the function and unit-test level using the module's own visibleWidth measurement.