optimise-seo
Implements Next.js crawlability, metadata, canonicals, status codes, JSON-LD, hreflang, and crawler policy with served-page evidence. Use when asked to "fix SEO", "add a sitemap", "fix soft 404s", or "add llms.txt". For demand research, briefs, or Search Console monitoring use seo-program.
mblode · 已有 AI 审核记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add mblode/agent-skills --skill optimise-seo技能说明
name: optimise-seo description: Implements Next.js crawlability, metadata, canonicals, status codes, JSON-LD, hreflang, and crawler policy with served-page evidence. Use when asked to "fix SEO", "add a sitemap", "fix soft 404s", or "add llms.txt". For demand research, briefs, or Search Console monitoring use seo-program.
Optimise SEO
- IS: crawlability, metadata, structured data, canonicals, redirects, status codes, hreflang, AI-crawler policy and extractable page structure, Core Web Vitals, programmatic SEO, security and privacy headers, and error-page behaviour, implemented in a Next.js App Router codebase.
- IS NOT: visual redesign, layout, or page-level UI quality (
ui-designDirection, Build, and Audit modes), demand research, briefs, question maps, or Search Console monitoring (seo-program), tenant routing forrobots.txtandsitemap.xml(multi-tenant-architecture), or writing the article (externalghostwriter).
Allowed surface: metadata and generateMetadata, JSON-LD, semantic HTML, internal links, alt text, app/sitemap.ts, app/robots.ts, app/feed.xml/route.ts, app/llms.txt/route.ts, proxy.ts, next.config.ts redirects and headers, error pages, and performance work. Component styling and layout stay as they are; when a fix needs a layout change, name it as a handoff to ui-design.
Before writing code, read the matching guide in node_modules/next/dist/docs/ (resolved from the app directory; in a monorepo the next package is not visible from the repo root). The bundled docs match the installed version; the notes in references/nextjs-implementation.md were checked against 16.3 and say where behaviour changed.
References
| File | Read when |
|---|---|
| references/indexing-policy.md | Redirects, duplicate URLs, index decisions, or pages generated at scale |
| references/nextjs-implementation.md | Before writing code in steps 2-4: metadata, sitemap and robots, redirects, indexing, streaming 404s, JSON-LD, OG images, CSP |
| references/answer-engines.md | The task touches AI crawlers, robots.ts rules for GPTBot or ClaudeBot, llms.txt, Content-Signal, or "why are we not cited" |
| references/internationalisation.md | The site has more than one locale |
| references/technical-hardening.md | Headers, cookies, consent, GPC, security.txt, or maintenance and error-status tasks |
| references/validation-evidence.md | Step 5: served-page commands and evidence expectations |
| references/seo-checklist.md | Step 5: copy into the report and mark each line with evidence |
Workflow
Copy and track this checklist:
SEO progress:
- [ ] Step 1: Inventory routes and decide index intent per route
- [ ] Step 2: Fix crawl/index foundations (sitemap, robots, canonicals, redirects, status codes)
- [ ] Step 3: Implement metadata and structured data
- [ ] Step 4: Improve semantics, extractable answers, internal links, and Core Web Vitals
- [ ] Step 5: Validate with references/seo-checklist.md and report evidence
Step 1 is a table: route, index or noindex, canonical, reason. Every later step reads from it, and a route with no row gets no work.
Must-have on every site
app/sitemap.tslists every indexable URL withlastModifiedderived from content;app/robots.tsnames it- One canonical per page, one host, one casing, one trailing-slash policy, set through
alternates.canonicalwithmetadataBasein the root layout. Next.js enforces the slash policy itself: by default it 308s/about/to/about, andtrailingSlash: trueinverts that (files with extensions and.well-known/paths are exempt either way) - Preserve the property's existing ownership verification. Use
metadata.verificationonly for a supplied token and the chosen verification method; DNS verification needs no duplicate meta tag - Root layout
robotslifts Google's default preview caps:googleBot: { 'max-snippet': -1, 'max-image-preview': 'large', 'max-video-preview': -1 }. Without them Google truncates the snippet and the image preview, and the snippet cap is what AI surfaces read against when deciding how much of a page they may quote authorsandcreatorset explicitly in the root layout, so person-level attribution does not live only in footer HTML and JSON-LD- Unique title and description per page; title and H1 lead with the non-brand primary keyword and agree in intent (a brand-led title on a category page competes only for navigational queries the site already wins)
- Every page opens with a short, plain-text answer to the question it exists to answer; the h2s are the questions people actually ask
- Open Graph and Twitter Card tags with a 1200x630 image
- JSON-LD: Organization and WebSite once in a root
@graphwith stable@ids, BreadcrumbList on inner pages, then Article, Product, ProfilePage, or LocalBusiness where the content type matches. Define an entity once and reference it by@id - One h1, logical h2-h6, descriptive alt text, internal links between related pages
- Core Web Vitals in target at the p75 of CrUX field data: LCP 2.5 s, INP 200 ms, CLS 0.1 (INP replaced FID in March 2024; a guide still naming FID is stale)
Redirects and indexing policy
Load references/indexing-policy.md when routes move, duplicates need consolidation, or page indexing policy changes. Verify behavior against the installed Next.js version and served responses.
AI crawlers and answer engines
Google needs no special files: AI Overviews and AI Mode run on Googlebot and core ranking, and nosnippet or max-snippet:0 is the only opt-out (it removes the normal snippet too). Everything else is a robots decision per crawler class: search and citation bots (OAI-SearchBot, Claude-SearchBot, `Perpl