agentleFS
qwwiwiqwwiwi/agentos-skills-public/skills/seo-tune/SKILL.md

seo-tune

Skill4 starsChanged 4 months agoMITView on GitHub

Audit and tune ANY existing site (static HTML, Astro, Next.js, WordPress export...) for search + AI discovery: SEO (Yandex & Google), AEO (AI answer engines), GEO (LLM/agent discovery). The agent checks robots.txt, sitemap.xml, canonical, meta/OG tags, JSON-LD, llms.txt and IndexNow, then generates whatever is missing. Use when the user says: optimize my site for search / SEO / чтобы находили в поиске / хочу в Яндекс и Google / AI-ответы. Tunes an existing site — it does not build one.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 2 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before touching anything, get these from the user and keep them in a scratch note (you'll reuse them).

    line 23
  2. Distinctive:

    BUG to flag: every page carrying the SAME canonical (typically → the homepage).

    line 89
  3. Very specific:

    Built-file → public-URL mapping: do not mechanically append the file path to SITE_URL.

    line 93
  4. Very specific:

    The user's own IDs/keys/domain — always: Never invent analytics IDs or reuse someone else's IndexNow key.

    line 162
  5. Very specific:

    Don't break what works: Read a page's existing <head> before editing.

    line 164

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

SEO tune — make an existing site findable by search AND by AI

Three layers, one pass:

  • SEO (classic search — Yandex, Google): robots.txt, sitemap.xml, canonical URLs, a real <title> + meta description, Open Graph / Twitter cards.
  • AEO (AI answer engines — so your page is quotable): structured data (JSON-LD), FAQ blocks with direct question→answer pairs.
  • GEO (LLM / agent discovery): a deliberate AI-crawler robots policy, IndexNow so new pages get picked up fast, and (optional, experimental) llms.txt.

This skill tunes a site that already exists. It doesn't generate a blog from scratch (that's a bigger, separate tool). Point it at a folder of built HTML, or a running site, and it fills the gaps.

Step 0 — ask for the site's basics (never hardcode)

Before touching anything, get these from the user and keep them in a scratch note (you'll reuse them). Nothing here is baked into the skill — it's all theirs:

  • SITE_URL — the canonical origin, e.g. https://example.com
  • BRAND — the site / author name
  • LANG — primary language (e.g. ru, en)
  • Analytics IDs, IF they use them (Yandex Metrika counter, GA4 id) — optional
  • IndexNow key — you'll generate one with them in references/indexnow.md

Step 1 — audit (report, then STOP)

Walk the site (the built HTML files, or fetch the live pages) and check each item. Report a checklist [есть]/[нет] first, then STOP and wait for the owner's explicit approval before you change anything. Do not proceed to Step 2 until they say go.

Presence is not enough — check that each page is effectively indexable and that its canonical is correct, or the audit will "pass" a site that is actually invisible to search or silently collapsing into one URL. A Disallow: /, an all-noindex site, or a site where every page canonicals to the homepage must FAIL this audit, not pass it.

LayerCheckWhere
SEOrobots.txt exists, points to the sitemap, and does NOT block indexing (see effective indexability)/robots.txt
SEOsitemap.xml lists only the desired indexable canonical URLs, absolute <loc>/sitemap.xml
SEOevery page has a unique, real <title> and <meta name="description"><head>
SEOevery page has exactly ONE absolute, correct <link rel="canonical"> (see canonical correctness)<head>
SEOOpen Graph (og:title + og:type + og:image + og:url; og:image:alt when an image is set) + twitter:card<head>
AEOarticle pages carry Article/BlogPosting JSON-LDreferences/jsonld-recipes.md
AEOQ&A / how-to pages carry FAQPage/HowTo JSON-LD (AEO / AI engines only — no Google rich result)same
AEOhome page carries the truthful site-subject type (Organization/Person/LocalBusiness/…) + WebSite JSON-LDsame
GEOllms.txt exists (optional, experimental — see note)references/llms-txt.md
GEOrobots.txt reflects the owner's chosen AI-crawler policy (allow OR intentionally block)references/llms-txt.md
GEOIndexNow key file is published and pages get pinged on publishreferences/indexnow.md

Effective indexability (presence ≠ indexable — do not skip)

A file that merely exists can still be fully blocked. For each page, verify it is actually indexable:

  • robots.txt rules that apply to the REAL crawler. robots.txt is most-specific-group-wins (RFC 9309): work out the group Googlebot / YandexBot actually match, not just the * group. Disallow: / — under * OR under the engine's own user-agent — means the site is blocked. FAIL it.
  • HTML noindex: <meta name="robots" content="noindex"> (or a googlebot / yandex-specific meta) removes the page from the index even if robots.txt allows crawling.
  • HTTP noindex: an X-Robots-Tag: noindex response header does the same, invisibly. Check the response headers, not just the HTML.
  • Final HTTP status: the URL must return a final 200 — not a redirect chain, a 4xx/5xx, or a login / paywall. A canonical or sitemap URL that 301s or 404s is a bug.
  • Rendered content: the main content must actually be present after render. For a JS-heavy site, an empty HTML shell that never hydrates has nothing to index — confirm the real content is in what a crawler receives.

Sources: Google robots meta / noindex, Yandex robots.txt.

Canonical correctness (presence ≠ correct)

<link rel="canonical"> is a strong consolidation signal — a wrong one can collapse a whole multi-page site into a single URL. For each page verify:

  • Exactly one canonical, and it is an absolute URL on your one chosen origin.
  • The canonical URL is itself indexable (returns 200, not noindex, not redirected).
  • It matches this page's content — a page's canonical should point to itself (or a genuine duplicate), never to an unrelated page.
  • BUG to flag: every page carrying the SAME canonical (typically → the homepage). This tells search engines every page IS the homepage and de-indexes the rest. This is exactly the kind of existing tag that must be FIXED, not preserved — "don't break what works" does not apply to a page-collapsing canonical.
  • Built-file → public-URL mapping: do not mechanically append the file path to SITE_URL. /about/index.html on disk is usually the public /about/; handle trailing slashes, query parameters, localized (i18n / hreflang) variants and redirected URLs deliberately, not by string concatenation.

Sources: Google canonicalization, Yandex canonical.

Step 2 — generate what's missing (only after approval)

For each gap, use the recipe file, filling in the user's real SITE_URL/BRAND:

  • robots.txt / sitemap.xml — generate the sitemap from only the desired indexable canonical URLs (see the sitemap rules + template below). <lastmod> is optional and should reflect a real content change; a build mtime that bumps on every deploy is as misleading as now, so omit it rather than fake it.
  • <head> tags — unique real title + a description that summarizes the page (aim ~140–160 chars as an editorial heuristic — there is no fixed Google limit and the snippet is trimmed per device/query). Canonical = the page's real public URL on your one origin (mind the built-file mapping caveat above). OG image per-page if you have one, else a site default.
  • JSON-LD — references/jsonld-recipes.md. Pick the TRUTHFUL subject type for the home page (Organization only if the site is an organization; Person for a personal site; LocalBusiness/OnlineStore where those fit). One <script type="application/ld+json"> per block in <head>.
  • llms.txt — references/llms-txt.md (optional, experimental).
  • IndexNow — references/indexnow.md (generate the key, publish the key file, ping once on every publish — one endpoint fans out to the others).

sitemap.xml — generation rules + template

Include only the URLs you want indexed: the canonical, indexable pages. Exclude redirects, noindex pages, duplicates, 404s, login and purely technical pages.

  • Absolute <loc> URLs on your one chosen origin, canonical-consistent (the same URL you set as <link rel="canonical">).
  • UTF-8 encoding, an XML declaration, the sitemap namespace, and XML-escaped entities (& → &amp;, < → &lt;, …).
  • Max 50,000 URLs / 50 MB per file; beyond that, split into multiple sitemaps and add a sitemap index.
  • <lastmod> is optional; if present, use a W3C-datetime value that reflects a real content change (not now, not a deploy mtime).
<?xml version="1.0" encoding="UTF-8"?>
<urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9">
  <url>
    <loc>https://example.com/</loc>
    <lastmod>2026-07-05</lastmod>
  </url>
  <url>
    <loc>https://example.com/blog/first-post/</loc>
    <lastmod>2026-07-05</lastmod>
  </url>
</urlset>

Sources: Google build a sitemap, Yandex sitemap format.

Rules

  • Truthful metadata only. Description must match the page; JSON-LD fields must match visible content. Mismatched / misleading structured data can lose a page its rich-result eligibility or trigger a structured-data manual action — but Google states this does not change normal web ranking, and Yandex simply ignores markup it can't trust rather than penalizing it. So mark up honestly for eligibility; there is no "ranking penalty" for a bad block, only lost eligibility. Google structured-data policies.
  • The user's own IDs/keys/domain — always. Never invent analytics IDs or reuse someone else's IndexNow key. If they don't use analytics, leave it out.
  • Don't break what works. Read a page's existing <head> before editing; add missing tags, don't duplicate ones already there. Exception: an actively harmful existing tag — a page-collapsing canonical, a stray noindex, a Disallow: / — is a bug to fix, not "what works" to preserve.
  • One canonical origin. Pick https:// + one host (www or bare, not both) and be consistent across canonical, sitemap and OG.

Files

  • references/jsonld-recipes.md — copy-paste JSON-LD for the common page types.
  • references/llms-txt.md — llms.txt + the AI-crawler allowlist (GEO).
  • references/indexnow.md — IndexNow key + ping (fast indexing on Yandex & Bing).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.