Migrate to rspress

[kdy1]

This reverts commit e7ea679.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
plugins	Ready	Preview, Comment	Jul 10, 2026 10:44am
website	Ready	Preview, Comment	Jul 10, 2026 10:44am

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	@​swc/​types@​0.1.27
	react@​19.2.7
	typescript@​6.0.3
	@​swc/​core@​1.15.43
	react-dom@​19.2.7

View full report

[claude]

Review: Migrate to rspress

Reviewed the diff (145 files, +4291/-2785) that replaces the Next.js + Nextra 4 docs site with Rspress v2. Overall this is a well-executed migration — the README.md documenting the Nextra → Rspress conversion notes, Twoslash caveats, and theme-alignment decisions is genuinely useful and should save the next person a lot of rediscovery. Component ports (Callout, Benchmark, Customers, TeamMembers, Authors, TypeDeclarations) are clean, drop the Next.js/Tailwind coupling as intended, and pick up small hardening wins along the way (rel="noopener noreferrer" on external links, loading="lazy" on avatar/customer images, consistent use of @rspress/core/runtime's BrowserOnly for the client-only chart). No use of dangerouslySetInnerHTML, eval, or other injection-prone patterns.

A couple of things worth double-checking before merge:

1. pnpm-lock.yaml looks out of sync with apps/website/package.json

In the lockfile diff, the apps/website importer's devDependency entry for @types/react still reads:

"@types/react":
  specifier: 19.0.12
  version: 19.0.12

but apps/website/package.json was changed from "19.0.12" to "^19.0.12" (caret added). Since pnpm validates that the lockfile's specifier matches package.json verbatim, this mismatch will make pnpm install --frozen-lockfile at the repo root reject the lockfile as stale (and a plain pnpm install will silently rewrite it). Worth regenerating the lockfile for this importer before merging, or reverting the caret if it wasn't intentional.

2. Workspace vs. standalone install story is a bit split-brained

apps/website is still matched by the root pnpm-workspace.yaml's apps/* glob and still shows up as an importer in the root pnpm-lock.yaml, but:

• apps/website/package.json now pins "packageManager": "pnpm@10.33.1", while the root package.json pins pnpm@8.6.3.
• The new README.md and vercel.json's installCommand both say to install with pnpm install --ignore-workspace.

So there are now two different install paths for this app (root workspace install under pnpm 8 vs. standalone --ignore-workspace install under pnpm 10) that can resolve dependencies differently and drift out of sync over time. If the intent is for apps/website to be fully standalone going forward, it'd be worth removing it from the root workspace glob (or documenting explicitly why it needs to stay); if not, the packageManager pin mismatch should be reconciled.

3. Redirects are duplicated in two places

The legacy-URL redirect list appears twice — once in vercel.json (redirects, server-side) and once in rspress.config.ts via pluginClientRedirects (client-side, for pnpm preview/non-Vercel hosting). Functionally fine, but any future redirect addition/removal has to be kept in sync manually in both files. A short comment in one file pointing at the other (or a shared source of truth) would reduce the chance of drift.

Minor / non-blocking

• Benchmark.tsx's chart.js callback types (items: any[], item: any, options={... as any}) fall back to any rather than chart.js's typed callback signatures — acceptable for a straight TS port, but leaves a gap in the type-safety win from moving off .jsx.
• The previously-unused lib/gtag.js (Google Analytics via a Universal Analytics ID, which Google sunset in 2023) is removed with no replacement — looks like intentional dead-code cleanup rather than a regression, since it wasn't wired into the old app/layout.jsx either, but flagging in case analytics were expected to carry over via some other mechanism.
• No CI workflow in .github/workflows/ actually runs pnpm build for apps/website, so a broken build would currently only surface at Vercel deploy time. Given Rspress v2's built-in dead-link checker (mentioned in the README) is a nice safety net, it'd be worth wiring a build step into CI to catch regressions pre-merge rather than post-deploy.

Test coverage

No automated tests exist for this app before or after the PR (consistent with it being a docs site), so this isn't a regression, but given the scope of the rewrite it'd be reassuring to confirm pnpm build and a manual click-through of the redirect list, playground page, and benchmark chart were run against the built output.

Nice work overall — the scope here (145 files, full framework swap, content reorganization, theme parity) is substantial and the execution is careful.