[oliverne]

This is a bug fix rather than a feature, so per CONTRIBUTING's "Sending a
change" I sent the PR directly instead of opening an issue first. If you would
rather discuss the design before reviewing, I am happy to move it to an issue.

What changed

A note's filename comes from its first-line H1 through slugify in
src/shared/slug.ts. That function kept only [a-z0-9] and turned everything
else into a separator, so a heading in any non-Latin script slugged to nothing
and the note landed as untitled.md. A Korean heading now names its file
한국어-제목.md, and a Korean workspace display name creates a Korean folder.

Three changes in slugify:

• The allowed set is \p{L}\p{N} instead of [a-z0-9], so letters and numbers
  in any script survive. Dots, separators, colons and control characters are
  still removed, so the "safe by construction" property is unchanged: the
  emitted slug can never contain a path character or start with a dot.
• After the existing NFD + combining-mark fold, the result is recomposed with
  normalize("NFC"). Without this, Korean text that arrives decomposed (NFD,
  which is what macOS input methods and the filesystem produce) would keep its
  jamo and be written as a name no other platform matches.
• Truncation at MAX_SLUG now counts code points rather than UTF-16 units, so a
  long CJK heading is not cut in the middle of a surrogate pair. The comment
  documents the limit as 60 code points, at most 240 UTF-8 bytes, which keeps
  room under the 255-byte filename limit for the enumerated -2 suffix and the
  extension.

createManaged in src/bun/workspaces.ts uses the same slugify, so workspace
folders inherit this with no separate code path. Both the desktop view and the
iOS/Android clients load src/shared/slug.ts, so all three name files the same
way.

Why

A note titled # 한국어 제목 was saved as untitled.md, and every following
Korean-titled note enumerated untitled-2.md, untitled-3.md and so on. The
title was still in the file, but the file on disk and the note browser had lost
the name the user wrote.

How verified

bunx tsc --noEmit    # clean
bun test             # 4504 pass, 0 fail (153 files)

Tests land in this change:

• src/shared/slug.test.ts: Korean and mixed-language headings, NFD input
  producing the same composed slug, unsafe path characters still removed
  (../../한국어/제목\초안:메모\0 becomes 한국어-제목-초안-메모), and a long
  CJK heading truncated without splitting a letter and staying under 255 bytes.
  The old assertion that 日本語 slugs to null is replaced by one that a pure
  emoji heading (🎉, which is neither \p{L} nor \p{N}) does.
• src/bun/notes.fs.test.ts: two Korean headings create 한국어-제목.md and
  한국어-제목-2.md without clobbering, and an untitled note can be retitled to
  a Korean heading that moves the file.
• src/bun/workspaces.fs.test.ts: a Korean display name creates a Korean folder.

No UI behavior changed, so bun run test:e2e does not apply here and there is no
native seam to check live.

Docs rules changed

This is called out per CONTRIBUTING's "Sending a change": two normative pages
stated [a-z0-9-] as the slug invariant, and they now say Unicode letters,
numbers and hyphens.

• docs/contributor/architecture.md — the trust-boundary bullet ("the view never
  names a file") and the filesystem-invariants subsection on the folder name
  being the only validated name.
• docs/contributor/remote.md — the note on why .server is dotted, which
  relied on managed workspace slugs never being able to collide with it. The
  reasoning still holds (a slug cannot start with a dot), but the stated range
  was wrong.

docs/contributor/interactions.md and the user manual needed no edit: the
filename rule is not user-facing text there.

[danhstevens]

Thanks for this! The diagnosis and the shape of the change were right. Reviewing it turned up three problems that Unicode filenames introduce, so I pushed a commit on top rather than sending you another round:

1. Overwrite on macOS. Finder writes Korean names decomposed (NFD), and APFS treats NFD and NFC spellings as the same file. uniqueName compared raw readdir names, so a new # 회의록 note next to a Finder-made 회의록.md replaced it without warning. It now compares names in NFC.
2. Mark stripping. Removing every combining mark was harmless when only ASCII survived, but now it changed the kept letters: ガイド became カイト and Hindi lost its vowel signs. Accents are now stripped only from Latin letters.
3. Length. The longest name built from a slug is the save’s temp file (..md.tmp--). With 60 four-byte letters that is over 255 bytes, which ext4 refuses. Slugs are now capped at 200 UTF-8 bytes too.

[oliverne]

Thanks for the thorough review and the additional fixes!