Agentic SEO

Repeatable workflow for auditing and implementing SEO on a web page. Standard SEO practice applies; the reference files hold the checks, templates and decisions specific to this skill.

First-Time Setup: Gather Site Configuration

Before running any sub-command for the first time in a session, ask the user for their website details if not already known. Store these as working context:

Setting	Ask For	Example
Site Name	"What is your site/company name?"	Acme Corp
Domain	"What is your website domain?"	https://acme.com
X/Twitter Handle	"What is your X/Twitter handle?" (optional)	@acme_corp
Social Image Path	"Where are your social sharing images stored?"	/assets/social/
Social Links	"Any social profile URLs?" (optional)	Twitter, GitHub, LinkedIn URLs
Framework	"What framework is your site built with?"	Angular, React, Next.js, etc.
Build Output Path	"Where does your built/prerendered HTML output to?"	dist/apps/web/browser/
Sitemap Path	"Where is your sitemap.xml?"	public/sitemap.xml

If the project has a CLAUDE.md with SEO patterns already defined, read it and extract these values automatically instead of asking.

Use these values as {SITE_NAME}, {DOMAIN}, {TWITTER_HANDLE}, {SOCIAL_IMAGE_PATH}, {BUILD_OUTPUT}, {SITEMAP_PATH} throughout all sub-commands.

When to use this skill

Trigger this skill when:

• A new page is created and needs SEO implementation
• An SEO audit is requested on an existing page
• Structured data (JSON-LD) needs to be added or updated
• llms.txt or llms-full.txt needs maintenance
• A quick checklist verification is needed before deploying
• A page is missing meta tags, OG tags, canonical URL, or sitemap entry

What Google says about AI search

Dated 2026-10-10: Google's guide "Optimizing your website for generative AI features on Google Search" (https://developers.google.com/search/docs/fundamentals/ai-optimization-guide, last updated 2026-07-10) says you don't need new machine readable files, AI text files, markup or Markdown to appear in Google Search; structured data isn't required for generative AI search and there is no special schema.org markup to add; Google Search doesn't use llms.txt, and having one neither harms nor helps visibility or rankings there; content does not need to be broken into tiny pieces. Standard SEO applies; re-check the guide before relying on this.

For performance, measure the Core Web Vitals (LCP, INP, CLS; https://web.dev/articles/vitals) rather than asking for "fast load times".

Sub-Commands

Command	Description	Reference
audit [path]	Audit one page's source (component and template): pass/fail table	reference/audit.md
implement [path]	Implement SEO on a new page from scratch (Angular example; adapt to the framework)	reference/implement.md
structured-data [path]	Add or update JSON-LD	reference/structured-data.md
llms-txt	Maintain llms.txt, only if the owner wants it	reference/llms-txt.md
crawler-controls	Decide what Google and AI vendor crawlers may use (neutral guide)	reference/crawler-controls.md
checklist [path]	Quick 15-item check against one page's built HTML	reference/checklist.md
audit-build [dir]	Audit every page of the built HTML of a whole site, optionally diffed against the live build; framework-agnostic	reference/audit-build.md

audit reads the source of one page; audit-build reads the built HTML of every page. With no sub-command, default to audit on the target path. Use audit-build when a change touches many pages (a rebuild, a migration, a redesign) or when you need to know what it does to pages that already rank.

Running the build audit script

scripts/audit_build.py is executed, not read into context. Python 3 standard library only.

python3 scripts/audit_build.py BUILD_DIR --domain {DOMAIN} [--baseline BASE_DIR] [--csv out.csv] [--twitter-handle @handle] [--strict-claims]
python3 scripts/audit_build.py --self-test

Exit codes: 0 no page failed a hard check and no site-wide issue was found; 1 at least one page failed a hard check or a site-wide issue was found; 2 bad arguments (missing BUILD_DIR or --domain). An unreadable file stops the run with a message naming it. Run --self-test after changing the script.

Implementation

Follow reference/implement.md for the per-page steps (meta tags, canonical link, JSON-LD, sitemap, prerender). In any framework, JSON-LD must be rendered during SSR/SSG (no browser-only guard) and removed when the component is destroyed.

Done means validated

Before reporting done, run audit or checklist on the built page, then run Google's Rich Results Test and the Schema Markup Validator on it and fix errors.

Anti-Patterns

• Schema for its own sake: add structured data only when it matches visible content; never add a schema to reach a count.
• Inaccurate lastmod: set it only when the page content really changed.
• Hardcoded dates: use the real modification date for dateModified.