Website SEO

Use this skill when public page content should be discoverable by search engines, or when the user asks to implement or diagnose SEO. Keep private application content within its intended access boundary.

Make the intended content available

Identify which public URLs should be indexed and what distinct content each one represents. Provide meaningful page-body HTML, descriptive titles and descriptions, and usable links between those pages. A client-rendered shell with an empty mount element does not by itself contain the content a crawler is meant to index.

Choose static generation when the framework can produce the content before a visit. Use SSR when initial HTML must be generated for the current request. Browser interactivity can be added to either result. SEO does not imply one universal deployment architecture.

Use the actual public URL when declaring a canonical address. Keep sitemap entries consistent with the intended indexable pages and robots behavior consistent with the site's access policy. Do not expose private user URLs or private data merely to improve an SEO score.

Initial HTML and per-route metadata

Build the following from each public route and its actual content, and include them in the initial HTML rather than waiting for a browser effect:

Item	What to provide
Page body	The meaningful public content for that URL, not only an empty app mount or loading placeholder.
Title and description	A route-specific <title> and <meta name="description">; detail pages use the loaded item's content.
Open Graph	Appropriate og:type, plus og:title, og:description, og:url, og:site_name and og:image for the intended preview.
Twitter Card	The intended twitter:card and matching title, description and image metadata.
Canonical	A <link rel="canonical"> using the intended public URL.
HTTP status	A status that reflects the page result, including a real 404 for an unknown route or missing detail item.

The intended preview must be available without executing application JS. Use publicly reachable absolute URLs for canonical and sharing metadata. Escape dynamic head text and attributes using the framework's metadata facilities or context-appropriate escaping.

Resolve missing content before sending the response. Carry a not-found result from route/data loading to the page handler's status; add appropriate noindex metadata to the error page. Returning a success status with a visible "not found" message creates a soft-404 problem. Robots metadata is not a replacement for the correct status. Keep client-side navigation's title and metadata consistent with the initial response for the same URL.

Only public data may enter public HTML or its embedded hydration state. Keep authenticated rendering behind its access checks and private cache policy, and out of public indexing.

Check the actual public response

When verification is required for the SEO task, request representative published pages without executing JavaScript. Check their status, meaningful body content and page metadata, then inspect the served sitemap and robots documents. Follow redirects and distinguish gateway restrictions from missing application content. A rendered browser page or a Dashboard score alone does not prove that the initial HTML contains the content to index.

Include a content-bearing public page, a dynamic detail page when present, and a missing route or detail identifier. Check raw body text, per-route head tags and the actual HTTP status with the normal visitor and relevant search/social crawler User-Agents. A 200 response is not enough: SSR may have failed and returned a shell that the browser later filled in. Inspect render errors when that happens. After a change to rendering, data loading or routing, repeat the affected response checks rather than an unrelated full validation suite.

Platform HTML injection and CSP compatibility remain under published response handling. The following sections describe the existing platform SEO behavior and implementation workflow.

SEO panel and gateway documents

The Dashboard's Fix issues action sends failed SEO checks as an ordinary user chat message, not a separate typed-action protocol. Its documented scoring thresholds are:

Item	Panel result
Meta keywords	3–8 pass; zero fails; 1–2 or more than 8 warns
Title	Missing fails; outside 30–60 characters warns
Meta description	Missing fails; outside 50–160 characters warns
H1 and H2	Missing or longer than 80 characters fails

Generated fix prompts target about 55 title characters and 150 description characters. The panel scans rendered DOM; that is not a measurement of body content already present in the raw HTML. The platform does not supply a prerendered crawler snapshot.

The gateway handles /sitemap.xml and /robots.txt, preferring a supplied project document and otherwise generating a default. The transparent path prefers the backend's document before object storage. The x-manus-seo-source header identifies cdn, server, or generated. The separate Dashboard scan can warn about a missing file while the gateway serves a default.

A generated sitemap lists known project routes. On the canonical domain, generated robots contains User-Agent: *, Allow: /, Disallow: /api/*, and a Sitemap: line. A non-canonical domain gets User-Agent: * and Disallow: /. Literal route declarations for those paths are restricted by routes.

SEO implementation

When discoverability is in scope, ensure meaningful page-body HTML is available without requiring a crawler to run JavaScript. Prefer complete HTML produced by the existing framework's static-generation support; a plain HTML site does not need a new framework or bundler for this. If choosing server rendering would enable a one-way server capability, explain that change before making it and respect explicitly staged requirements.

An alternative is to prerender in the development environment and commit the generated HTML for static publication. Do not put browser-based prerendering in build.command; the hosted static build does not provide that browser and shares its time budget with clone/upload. Keep the prerendered files and the declared output consistent.

Canonical, og:url, and sitemap absolute addresses must use an explicitly configured real public origin, never an internal request-derived address. Until that origin is configured, omit those absolute tags rather than emitting a guessed/internal URL. The complete Preview origin contract is separate from the SEO panel's thresholds above.