SEO: noindex near-duplicate + thin entity pages

src/app/[locale]/agents/[slug]/page.tsx

@@ -18,14 +18,19 @@ export async function generateMetadata({ params }: { params: Promise<{ locale: s
   const t = await getTranslations({ locale, namespace: "AgentDetail" });
   const agent = await prisma.agent.findFirst({
     where: { slug, status: "APPROVED" },
-    select: { name: true, description: true },
+    select: { name: true, description: true, kind: true, _count: { select: { skills: true } } },
   });
   if (!agent) return { title: t("notFound") };
   const description = (agent.description || `${agent.name} on ${SITE_NAME}`).slice(0, 200);
+  // Thin: a scraped PROJECT entry with no declared skills and a barely-there
+  // description adds little unique value over the GitHub repo itself — keep it out
+  // of the index to limit scaled-content exposure. Substantive entries stay indexed.
+  const thin = agent.kind === "PROJECT" && agent._count.skills === 0 && (agent.description ?? "").trim().length < 80;
   return {
     metadataBase: new URL(absoluteUrl("")),
     title: agent.name,
     description,
+    ...(thin ? { robots: { index: false, follow: true } } : {}),
     alternates: localizedAlternates(locale, `/agents/${slug}`),
     openGraph: { title: `${agent.name} — ${SITE_NAME}`, description, url: absoluteUrl(`/agents/${slug}`), type: "website", locale: localeOg(locale), images: [absoluteUrl("/opengraph-image")] },
     twitter: { card: "summary_large_image", title: `${agent.name} — ${SITE_NAME}`, description },

src/proxy.ts

@@ -1,10 +1,34 @@
 import createMiddleware from "next-intl/middleware";
 import { routing } from "@/i18n/routing";
+import type { NextRequest } from "next/server";
 
 // In Next.js 16 the request interceptor file is `proxy.ts` (formerly
 // `middleware.ts`). next-intl's middleware factory returns a plain
-// `(NextRequest) => NextResponse` handler, so it works unchanged here.
-export default createMiddleware(routing);
+// `(NextRequest) => NextResponse` handler, which we wrap to add SEO headers.
+const handle = createMiddleware(routing);
+
+const NON_DEFAULT_LOCALES = new Set<string>(routing.locales.filter((l) => l !== routing.defaultLocale));
+
+// Entity DETAIL pages (/{locale}/agents/{slug}, /{locale}/skills/{slug}) carry
+// English-only bodies (scraped repo descriptions / AgentCards / READMEs); the 14
+// non-default locales are translated UI chrome around identical content. Emit
+// `X-Robots-Tag: noindex` on those so the 15-locale fan-out isn't read as scaled /
+// duplicate content. `follow` keeps link equity flowing, and hreflang still points
+// crawlers at the canonical English page (which stays indexable). Listing,
+// scenario, and marketing pages are genuinely translated, so they keep indexing in
+// every locale.
+function isNonDefaultLocaleEntityDetail(pathname: string): boolean {
+  const seg = pathname.split("/").filter(Boolean);
+  return seg.length >= 3 && NON_DEFAULT_LOCALES.has(seg[0]) && (seg[1] === "agents" || seg[1] === "skills");
+}
+
+export default function proxy(req: NextRequest) {
+  const res = handle(req);
+  if (isNonDefaultLocaleEntityDetail(req.nextUrl.pathname)) {
+    res.headers.set("X-Robots-Tag", "noindex, follow");
+  }
+  return res;
+}
 
 export const config = {
   // Run on page routes only. Exclude: