[prisma-robot]

SEO audit remediation, work package 4 of 8 (audit 2.3 duplicate titles/descriptions/H1s, 3.3 generic anchor text, 3.4 stale bookmark), all in apps/docs.

Versioned docs pages (orm/v6, orm/v7, cli/v7, guides/v7, /v7) shared their <title>, meta description, and H1 with their twin in every other version tree, including the unversioned Prisma 8 tree. Both v6 and v7 stay indexable; the version is now explicit in the metadata, and a canonical is used only where two versioned URLs are the same document.

What it delivers

A. Version-explicit titles, descriptions, and H1 marker

• src/lib/version-metadata.ts (new): getPageVersion(pageUrl) → { version, label } or null for latest/unversioned, built on the existing pathname helpers in version.ts. Labels: /orm/v6/* → "Prisma ORM v6", /orm/v7/* → "Prisma ORM v7", /cli/v7/* → "Prisma CLI v7", /guides/v7/* and /v7/* → "Prisma 7". Plus withVersionTitle / withVersionDescription, which skip pages whose text already names the version. Named getPageVersion/getPageVersionLabel rather than getVersionLabel because version.ts already exports a getVersionLabel(version).
• generateMetadata folds the label in: title → Supported databases (Prisma ORM v6), description → … Applies to Prisma ORM v6.. OG and Twitter read the same variables, so they stay in sync. The %s | Prisma Documentation template is untouched.
• The H1 area renders an Eclipse Badge with the label next to (never inside) <DocsTitle>. Note: badge: frontmatter is only rendered in the sidebar today (sidebar-badge-provider.tsx), not on the page, so there was no existing page badge to sit beside — the version badge reuses that component and its rounded-full! pattern.
• The .md / llms.mdx rendition's title line carries the same label (get-llm-text.ts), so agents see the version too.
• structured-data.tsx: the TechArticle headline/description use the same labelled values, and mainEntityOfPage.@id now points at the canonical instead of always at the page's own URL.

B. Canonical consolidation

New optional canonical frontmatter field (source.config.ts), resolved by src/lib/canonical.ts (page.data.canonical ?? self, /docs base path applied, absolute URLs passed through). scripts/find-version-twin-duplicates.ts (new, re-runnable, --check mode) compares every v6 page against its v7 twin with frontmatter stripped and /orm/v6/ normalised to /orm/v7/. 12 of 133 pairs are materially identical (5 byte-identical, 7 differing only by version paths); each v6 page now canonicals to its v7 twin. No noindex was added.

v6 page	canonical
orm/v6/more/troubleshooting/bundler-issues	/orm/v7/more/troubleshooting/bundler-issues
orm/v6/more/troubleshooting/check-constraints	/orm/v7/more/troubleshooting/check-constraints
orm/v6/more/troubleshooting/graphql-autocompletion	/orm/v7/more/troubleshooting/graphql-autocompletion
orm/v6/more/troubleshooting/nuxt	/orm/v7/more/troubleshooting/nuxt
orm/v6/more/troubleshooting/raw-sql-comparisons	/orm/v7/more/troubleshooting/raw-sql-comparisons
orm/v6/more/comparisons/prisma-and-mongoose	/orm/v7/more/comparisons/prisma-and-mongoose
orm/v6/more/dev-environment/environment-variables	/orm/v7/more/dev-environment/environment-variables
orm/v6/prisma-client/client-extensions/type-utilities	/orm/v7/prisma-client/client-extensions/type-utilities
orm/v6/prisma-client/debugging-and-troubleshooting/debugging	/orm/v7/prisma-client/debugging-and-troubleshooting/debugging
orm/v6/prisma-client/setup-and-configuration/database-polyfills	/orm/v7/prisma-client/setup-and-configuration/database-polyfills
orm/v6/prisma-schema/data-model/multi-schema	/orm/v7/prisma-schema/data-model/multi-schema
orm/v6/reference/supported-databases	/orm/v7/reference/supported-databases

The two audit "Duplicate content" hits, /orm/v7/more/troubleshooting/graphql-autocompletion and /orm/v7/prisma-schema/data-model/multi-schema, duplicate their v6 twins (not an unversioned page: no v6 or v7 page is identical to its Prisma 8 counterpart — verified by the same comparison run against the unversioned tree). Both are covered by the v6 → v7 canonicals above, so the v7 URL is the one that consolidates.

C. Cross-section duplicates — metadata rewritten from each page's own H1 and intro:

Page	Old	New
guides/integrations/embed-studio	title Prisma MCP Server / desc Manage Prisma Postgres databases using LLMs with the Prisma Model-Context-Protocol (MCP) Server	Embed Prisma Studio in a Next.js app / Embed Prisma Studio in a Next.js application with @prisma/studio-core so you can browse and edit your database from inside your app.
postgres/database/query-insights	Query Insights with Prisma Postgres / Inspect slow queries, connect Prisma calls to SQL, and apply focused fixes with Prisma Postgres.	Query Insights for a Prisma Postgres database / How Query Insights surfaces slow queries for a Prisma Postgres database in the Prisma Console, and where the full documentation lives.
orm/v6/reference/errors/index	Errors / Prisma Client, Migrate, Introspection error message reference	Prisma ORM error troubleshooting index / Where to find help for common Prisma ORM errors: connection pool, migration, shadow database, baselining, and MongoDB replica set issues.
orm/v7/reference/errors/index	same as above	same as above

Decisions: /query-insights (the dedicated section, 183 lines) stays primary and is unchanged; the Prisma Postgres page is a short overview that already links to it, so it was differentiated rather than canonicalised. The four ai/tools/* pages and their orm/v6/more/ai-tools/* copies differ in body (checked), so they got no canonical — Part A's "(Prisma ORM v6)" label differentiates the v6 copies. The error-reference pages keep their metadata; only the errors/ hub pages (the non-primary of each same-tree pair) were rewritten.

D. Stale bookmark + a gap in the link linter

orm/v7/.../opentelemetry-tracing.mdx linked #step-1-install-up-to-date-prisma-orm-dependencies; the heading is now "Step 1. Install Prisma ORM dependencies". Fixed (the v6 copy still has the old heading and was correct already). pnpm lint:links did not catch it: next-validate-link resolves fragments only on links that carry a path, so same-page bookmarks pass silently. scripts/lint-links.ts now checks #fragment-only links and href="#..." against the page's toc ids, headings (including ones nested in MDX components), and id="…" attributes. Across all 688 pages it reported exactly that one link and no false positives. Fence tracking moved to scripts/lib/markdown-regions.ts and now honours CommonMark (a closing fence carries no info string) — the old "count every fence line" approach desynchronised on nested examples and hid the tail of long files such as prisma-client-reference.mdx.

E. Generic anchor text — scripts/rewrite-generic-link-text.ts (new, re-runnable, --check mode) rewrote every anchor whose text was exactly one of the 14 generic phrases: 157 links on 154 lines in 65 files. It names the destination from the target page's frontmatter title, or the target heading when the link has a fragment; external destinations come from a curated name table and are never guessed (a destination it cannot name is reported and left alone — none remain). Hrefs are unchanged; code blocks and headings are skipped. Output was reviewed line by line and four sentences were hand-adjusted afterwards (create-db and prisma platform references in the v6/v7 CLI reference). Re-running the script is now a no-op.

Folder	Links	Lines	Files
orm/v6	87	85	36
orm/v7	56	55	17
guides	9	9	7
accelerate	2	2	2
ai	2	2	2
console	1	1	1
total	157	154	65

F. Housekeeping

• .claude/CLAUDE.md: the "Version Fallback" paragraph (and the content/docs.v6/ structure bullets and the stale src/app/docs/[version]/… route path) now describe reality — versions are content folders, there is no fallback, and the metadata/canonical helpers are named.
• Deleted src/components/version-banner.tsx (nothing imported it).
• src/lib/content-metadata.test.ts (new): fails if two pages in the same version tree share metaTitle + metaDescription, and if any page is missing either. Twins across trees stay allowed because Part A differentiates them at render time.

How it was tested

• pnpm check (includes the new tests): green — 29 tests in apps/docs, 31 in the repo.
• pnpm types:check: docs, blog, eclipse, @prisma/eclipse, @prisma-docs/ui green. site#types:check fails in this sandbox with DOCS_ORIGIN is required in production — reproduced on unmodified main, and apps/site is untouched here. @prisma-docs/ui also needs pnpm --filter @prisma/eclipse build first in a fresh clone.
• pnpm lint:links: 0 errors, now including same-page fragments. Verified the new check actually fails by reintroducing the stale bookmark.
• pnpm --filter docs run lint:spellcheck: 827 files, 0 issues.
• apps/docs unit tests cover the helpers directly: 16 for version-metadata, 4 for canonical, 2 content invariants.
• No next dev / next build for apps/docs or apps/blog (they OOM in the sandbox, per the task), so the badge placement and rendered <head> were verified by reading the source, not in a browser.

Assumptions and choices

• The description suffix is a sentence (Applies to Prisma ORM v6.) rather than a parenthetical, so it reads naturally when Google truncates it.
• "Materially identical" for a canonical means byte-identical after frontmatter removal and /orm/v6/ → /orm/v7/ normalisation. Every one of the 12 pairs was diffed; all differences were version paths only. Nothing that differs in substance was consolidated.
• The docs test script now runs node --import tsx --test instead of bare node --test: without the loader, a test cannot import a module that uses extensionless relative imports (version-metadata.ts → ./version). tsx is already a devDependency and the gate is unchanged in what it asserts.
• The canonical frontmatter value is a docs-relative path; absolute URLs also work.

Follow-ups not done here (out of scope)

• The sitemap still lists v6 URLs that now declare a different canonical (src/app/(docs)/sitemap.ts); it should probably drop or deprioritise pages with a foreign canonical.
• scripts/find-version-twin-duplicates.ts --check is not wired into a gate. It would catch a v6 page that drifts away from its twin while keeping a canonical, but content edits would then fail the build, which deserves its own decision.
• pnpm check (oxfmt) reformats two unrelated apps/blog files on a clean checkout of main; reverted here to keep this diff scoped.
• Several metaTitles are still generic across sections (for example Errors on both error-reference pages, one per tree); they are unique within a tree, so the new test passes, but they are weak titles.

[vercel]

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

Project	Deployment	Actions	Updated
blog	Ready	Preview	Sep 11, 2026 10:50am UTC
docs	Ready	Preview	Sep 11, 2026 10:50am UTC
eclipse	Ready	Preview	Sep 11, 2026 10:50am UTC
site	Ready	Preview	Sep 11, 2026 10:50am UTC

[coderabbitai]

Important

Review skipped

Bot user detected.

To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Advanced

Run ID: 7cc7dc93-5d26-4aa5-9a4d-342695f0dac6

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

🍈 Lychee Link Check Report

2454 links: ✅ 864 OK | 🚫 0 errors | 🔀 91 redirects | 👻 1590 excluded

✅ All links are working!

---

Full Statistics Table

Status	Count
✅ Successful	864
🔀 Redirected	91
👻 Excluded	1590
🚫 Errors	0
⛔ Unsupported	0
⏳ Timeouts	0
❓ Unknown	0

[prisma-robot]

Reviewed f8ba56fc672f0bdd3e9596c58281a42eb15e7a96: full diff, the code around it, and the gates run locally (pnpm --filter docs test 29/29, lint:links 0 errors incl. the new same-page fragment check, types:check for docs, cspell 0 issues, find-version-twin-duplicates --check and rewrite-generic-link-text --check both clean, oxfmt/oxlint clean in apps/docs). The 12 canonicals match what the comparison script reports. The Vercel preview is login-gated, so rendered <head> was checked from source, not in a browser.

Three findings, fixed and pushed in cf6ffa1259b6779869da2a33924f5daace5759de:

1. structured-data.tsx — JSON-LD @id breaks on an absolute canonical. mainEntityOfPage.@id was ${baseUrl}${canonical}, but resolveCanonicalUrl passes absolute canonical: values through by design (documented in this PR and covered by its tests), so such a page would emit https://www.prisma.iohttps://www.prisma.io/docs/.... None of the 12 current canonicals are absolute, so nothing is broken today, but the feature advertises the form. Added absoluteCanonicalUrl(canonical, baseUrl) in canonical.ts (relative → joined, absolute → untouched), used it there, and added two tests (31 now pass).
2. .claude/CLAUDE.md — wrong path. The new "Content Structure" bullet named content/docs/v7/ for the versioned getting-started tree; that folder does not exist, the pages live in content/docs/(index)/v7/ and are served at /v7. Corrected.
3. Anchor rewrite dropped a noun. In orm/v6 and orm/v7 prisma-client/client-extensions/shared-extensions/index.mdx the script turned "visit this page." into "visit the Shared packages & examples." Dropped the article so it reads "visit Shared packages & examples." Re-running the rewrite script is still a no-op.

I will re-review the new head from scratch shortly.

— reviewer

[prisma-robot]

Re-reviewed cf6ffa1259b6779869da2a33924f5daace5759de from scratch (fresh clone, full diff including my round-1 fixes, gates run locally: pnpm --filter docs test 31/31, types:check for docs, lint:links exit 0 with the same-page fragment check, lint:agent-ready 0 failures, cspell 0 issues, find-version-twin-duplicates --check and rewrite-generic-link-text --check both no-ops, oxfmt/oxlint clean in apps/docs; the 12 canonicals match the comparison script; Badge spreads its props so data-markdown-ignore reaches the DOM).

One finding, fixed and pushed in d613787b3b9d89d1946393048a4c0a597a54fbe0:

version-metadata.ts — the anti-stutter guard misses "Prisma 7". mentionsVersion only recognised the full label (Prisma ORM v7) or the bare v7 token, so a title that names the version without the v still got the label appended. In this content that produces What is Prisma 7? (Prisma ORM v7) (orm/v7/index.mdx metaTitle, and its title: Prisma 7 in the .md rendition) and What is Prisma ORM? (Prisma 6 overview) (Prisma ORM v6) (orm/v6/overview/introduction/what-is-prisma.mdx) — exactly the stutter the helper's docstring says it avoids. It now also matches Prisma [ORM|CLI] N with N followed by a non-digit, so Prisma 6.16 counts as naming the 6 line and Prisma 60 does not. Tests added for both directions (32 now pass). Nothing else in the versioned trees is affected; to-v7.mdx's description ("…upgrade to Prisma ORM 7") on the v6 page still gets "Applies to Prisma ORM v6.", which is correct.

Everything else read clean. I will re-review the new head from scratch shortly.

— reviewer

[prisma-robot]

Review clean at d613787

Third cold read (fresh clone, branch 0 behind main, full source and content diff including the two previous fix commits, gates run locally): pnpm --filter docs test 32/32, types:check for docs, lint:links exit 0 with the same-page fragment check, lint:agent-ready 0 failures, cspell 0 issues in 827 files, find-version-twin-duplicates --check and rewrite-generic-link-text --check both no-ops, oxfmt/oxlint clean in apps/docs. No human comments in the thread. Nothing left to fix.

What the PR does (all in apps/docs): versioned pages (orm/v6, orm/v7, cli/v7, guides/v7, /v7) get a version label folded into <title>, meta description, JSON-LD headline, the .md rendition title, and a Badge beside the H1, so twins across version trees stop sharing identical metadata; a new optional canonical frontmatter field points 12 v6 pages that are byte-identical to their v7 twin at the v7 URL (verified by the new comparison script); four cross-section duplicate title/description pairs are rewritten; one stale same-page bookmark is fixed and lint:links now catches that class; 157 generic anchors (here, learn more, …) are rewritten to name their destination with hrefs unchanged; .claude/CLAUDE.md now describes the real content layout; an unused version-banner.tsx is deleted; a content test enforces unique metaTitle + metaDescription within a tree.

Risk: low. Every change is confined to the docs app and its content; the helpers are unit-tested including the edge cases found in rounds 1–2; the anchor rewrites change only link text and every href still resolves; the canonicals are reversible frontmatter on pages whose bodies are provably identical. The one thing not verified in a browser (the Vercel preview is login-gated) is the badge placement next to the H1 — it is a flex-wrap wrapper around DocsTitle and a Badge with the sidebar's existing rounded-full! pattern, so the downside is cosmetic.

Proposed follow-up (awaiting a human's approval, not blocking this PR): a builder task to drop pages with a foreign canonical from src/app/(docs)/sitemap.ts, which the PR body names as out of scope — the 12 v6 URLs are still listed there while their <link rel=canonical> points at v7.

— reviewer

[prisma-robot]

Review clean at 782cf66

Cold read of the new head after the merge of main (fresh clone, branch 0 behind main). The merge commit is a clean merge: the PR's diff against its main parent is byte-for-byte the same 98 files / +1361 −206 as before, and what main brought in (#8237: unversioned ORM/CLI content and cspell.json) touches no file this PR touches. I re-read the full source and content diff, including the fixes from rounds 1–3, then ran the gates on the merged head: pnpm --filter docs test 32/32, types:check for docs (after @prisma/eclipse build), lint:links 0 errors with the same-page fragment check, find-version-twin-duplicates --check (12 identical, 0 missing, 0 stale), rewrite-generic-link-text --check no-op, lint:agent-ready 0 failures, cspell 0 issues in 827 files, oxfmt clean, oxlint only the pre-existing unused-import warnings. I also planted a generic [here](…) link and a dead #fragment on a page and confirmed both new checks fail on it (then restored the file). The new content from main passes the new same-tree unique metaTitle + metaDescription test as well.

What the PR does (all in apps/docs): versioned pages (orm/v6, orm/v7, cli/v7, guides/v7, /v7) get a version label folded into <title>, meta description, JSON-LD headline, the .md rendition title, and a Badge beside the H1; an optional canonical frontmatter field points 12 v6 pages that are identical to their v7 twin at the v7 URL; four cross-section duplicate title/description pairs are rewritten; one stale same-page bookmark is fixed and lint:links now catches that class; 157 generic anchors are rewritten to name their destination with hrefs unchanged; .claude/CLAUDE.md describes the real content layout; the unused version-banner.tsx is deleted; a content test enforces unique metaTitle + metaDescription within a tree.

Risk: low. Everything is confined to the docs app and its content, the helpers are unit-tested including the edge cases found in earlier rounds, every rewritten anchor still resolves, and the canonicals are reversible frontmatter on provably identical pages. Not verified in a browser (the Vercel preview is login-gated): the badge placement beside the H1, which is cosmetic at worst. One cosmetic note, not a fix: two v7 comparison pages now link "[Type safety Overview]" because that is the v7 page's own frontmatter title (its metaTitle is "Type safety"); that title predates this PR.

The sitemap follow-up (drop or deprioritise pages with a foreign canonical in src/app/(docs)/sitemap.ts) was already proposed as a builder task in the previous review round and awaits a human's decision; I have not proposed it again.

— reviewer