fix(repo): make a clean clone pass its own checks on Windows

[kmuzammil101]

I found two small Windows-related issues after cloning the project today and following the quick start:

bun install fails on Windows.
The prepare script uses two redirects (>/dev/null 2>&1), which Bun's shell can't parse on Windows. The dependencies still install, but the command exits with an error at the end. Because of this, core.hooksPath isn't configured, so the pre-push hook isn't installed. The CONTRIBUTING guide says bun install sets this up, but that's not true on Windows. I had to use --no-verify to push.

bun run lint fails because of Windows line endings.
Git for Windows uses CRLF by default, while Biome expects LF. Since the repo doesn't have a .gitattributes file, Windows checkouts get CRLF and linting fails across @crm/db, @crm/ui, and @crm/env. Adding * text=auto eol=lf to .gitattributes fixes this. git add --renormalize . produced no changes, so this shouldn't affect existing files or macOS users—it only makes Git use LF when checking out on Windows.

I also checked the tests. bun run test fails on an unmodified main as well, so it doesn't seem related to these changes. The bulk.spec.ts test times out after around 5 seconds and then the test process hangs.

Separately, bun run db:test also fails on a clean clone because pg is used in test-db.ts but isn't listed in any package.json. Also, test-db.ts expects TEST_DATABASE_URL from process.env, but when running through the packages/db filter, Bun doesn't load the root .env. The existing require-local-db.ts already handles this correctly.

Happy to send a separate fix for the database test issues.

---

Summary by cubic

Fix Windows setup: make the prepare script work with Bun’s shell and enforce LF checkouts so lint passes on Windows. After this, a clean clone on Windows installs Git hooks and bun run lint succeeds.

• Bug Fixes
  • Update prepare to set core.hooksPath without double redirects (git config core.hooksPath .githooks 2>/dev/null || true) so bun install doesn’t error on Windows.
  • Add .gitattributes with * text=auto eol=lf to ensure LF line endings on Windows and avoid Biome CRLF lint failures.

^{Written for commit 6f200fd. Summary will update on new commits.}

[vercel]

@kmuzammil101 is attempting to deploy a commit to the Comp AI - PoC Team on Vercel.

A member of the Team first needs to authorize it.

[cubic-dev-ai]

No issues found across 2 files

_{Re-trigger cubic}

[ripgrim]

@cubic-dev-ai verify this for me rq

[cubic-dev-ai]

@cubic-dev-ai verify this for me rq

@ripgrim Verified both changes—they're sound and match the PR's claims.

prepare script: Dropping git rev-parse --git-dir >/dev/null 2>&1 && in favor of a bare git config core.hooksPath .githooks 2>/dev/null || true is a reasonable fix for the Bun-shell redirect parsing issue. Keeping 2>/dev/null || true still silences the non-repo case and ensures a non-zero exit never fails the install, so hooks get configured on Windows. The only behavioral difference is that the git-dir existence guard is gone, so it would now also attempt the config outside a repo—harmless here given || true.

.gitattributes: * text=auto eol=lf is the right fix. text=auto lets Git keep detecting binaries, while eol=lf forces LF on checkout so Biome stops failing on CRLF. Since git add --renormalize . showed no diff, existing tracked content is unaffected, which matches your reasoning.

One thing I can't fully confirm from the snippet: the actual failure mode you hit with >/dev/null 2>&1 in Bun's shell on Windows. That's a plausible Bun-shell quirk, but if you want certainty it's worth confirming the CI/Windows repro before merging. The fix itself is low-risk regardless.

The separate bun run test / db:test issues you noted are outside this PR's scope, so leaving them for a follow-up is a reasonable call.

_{Tip: get faster answers by chatting with cubic’s review copilot in the review UI}