# The vendored upstream source is compared byte-for-byte against the installed
# package by `src/lib/tw-merge.test.ts` — that comparison is what proves the
# folder has not silently forked from the library it replaces.
#
# Without this line git rewrites the endings on checkout (CRLF on Windows) while
# the copy in node_modules, unpacked from a tarball, keeps LF. The two then
# differ in every file, which reads as a forked vendor copy and is nothing of
# the sort. The test normalises endings as well, so this is belt and braces:
# this keeps the working tree honest, that keeps an already-cloned tree passing.
src/lib/tw-merge/** text eol=lf

# The git hooks are shell scripts run by `sh`, including on Windows through the
# Git Bash that ships with git itself. A CRLF checkout makes every one of them
# fail on the first line with an unreadable error, so they are pinned to LF
# regardless of the platform or of anyone's core.autocrlf setting.
.husky/** text eol=lf

# Generated, and re-read as text by the showcase's token table and by the
# contrast suite. Keeping them LF means a regenerated file diffs against its
# predecessor on content alone.
src/styles/tokens.css text eol=lf
src/icons/generated/** text eol=lf

# Written by the same script, in the same run, as the two lines above — and
# read back by `gen-tokens-json.mjs --check`, which compares byte for byte.
# Without this, git hands a Windows clone a CRLF copy, the generator emits LF,
# and `pnpm verify` fails on a tree nobody has touched.
docs/TOKENS.md text eol=lf

# The same rule, and the same failure: `gen-consumer-skills.mjs --check`
# compares each generated table against the file byte for byte, so a CRLF
# checkout reports all five as stale at once — which reads as a source change
# that landed without paperwork and is nothing of the sort.
consumer-skills/ui-kit-usage/references/** text eol=lf