Skip to content
Level up your prompts with SurePrompts — curated prompts for every workflow.
agentscamp

SEO Check

Run a single-page SEO and AEO check on a file or URL (title and meta lengths, one H1, heading order, answer-first lead, FAQ, structured data, canonical, internal links, alt text) and return a pass-or-fix table.

/seo-check<file path or URL>
Updated

A read-only slash command that checks one page, from a Markdown or HTML file or a fetched URL, against ten rules: title and meta length, one H1, heading order, canonical, internal links, alt text, an answer-first lead, a visible FAQ, and structured data backed by visible content. It returns a pass-or-fix table with evidence and the skill for each fix.

Download

Install to ~/.claude/commands/seo-check.md

View as Markdown
or via CLI:npx agentscamp add commands/seo-check

Check one page before it ships, or after it stops performing. This command reads a Markdown or HTML file, or fetches a URL, and runs ten checks that cover classic on-page SEO and answer-engine readiness together, returning a pass-or-fix table with the evidence behind each verdict. It reads only and changes nothing. For a whole site, use the seo-technical-auditor skill; for a ranked plan across pages, the seo-specialist agent.

Scope

Interpret $ARGUMENTS in this order:

  1. A URL (http:// or https://): fetch it with curl -sL, record the final URL and status code, and keep the response headers for X-Robots-Tag and Link: rel=canonical.
  2. A path ending in .html: Read it as the rendered page.
  3. A path ending in .md or .mdx: Read the frontmatter and body. Take the title from seoTitle or title, the meta description from seoDescription or description, the FAQ from faq, and treat the body's first paragraph (or a summary field the site renders) as the lead. Say in the report that rendered output was inferred from source.
  4. A path to a page component (.tsx, .astro, .vue): read the metadata export or the <head>, mark the check partial, and recommend running it again on the built HTML.
  5. Several arguments: one report per page, in the order given.
  6. Empty: ask "Which file or URL should I check?" and stop. Do not pick a page yourself.

NOTE

This command needs a target query to judge intent. Use one if the frontmatter or the user supplies it; otherwise use the H1 as the stand-in and say so. It does not check facts, sources, or voice; the content-editor agent and /brand-check do those.

Step 1 — Extract the fields

List what you pulled before judging it: title, meta description, canonical, robots directive, H1 text and count, the heading sequence (level and text), the first paragraph, the visible FAQ section, every application/ld+json block and its @type, internal links (count and targets), and images without alt. If a field is absent, write "absent"; absence is a finding, not a gap in the report.

Step 2 — Apply the ten checks

CheckRule
Title50 to 60 characters; contains the target query
Meta description120 to 155 characters; one or two sentences; not a copy of the title
H1exactly one; makes the same promise as the title
Heading orderno level skipped; H2s read as questions or searchable noun phrases
Answer-first leadthe first paragraph answers the title's question in 60 words or fewer and names the entity
FAQa visible section of at least three questions phrased the way people ask them; faq frontmatter that does not render is a FIX
Structured dataat least one JSON-LD block that parses, whose @type matches the page kind, and whose dates, author, and FAQ values appear on the page
Canonicalpresent, absolute, on the production host, self-referencing
Internal linksat least three to pages on the same site, each resolving
Imagesevery img has an alt attribute

A verdict is PASS or FIX; there is no "warn". Attach one line of evidence to every row: the measured length, the quoted lead, the @type found, the count.

For a URL, check up to ten internal links with curl -s -o /dev/null -w '%{http_code}' <link> and list any that are not 200. For a file, Glob for the matching route or content file (for example /guides/x/y to src/content/guides/x/y.md or app/guides/x/y/page.*) and list any that do not exist. Stay read-only throughout.

Step 4 — Report

## /seo-check: <title or path>
 
Source: <URL (final status 200) | file path (inferred from source)>
Target query: <query | H1 used as stand-in>
 
| Check             | Verdict | Evidence                                          |
| ----------------- | ------- | ------------------------------------------------- |
| Title             | PASS    | 54 chars; contains "session replay"               |
| Meta description  | FIX     | 171 chars; will truncate                          |
| H1                | PASS    | 1 found; matches title                            |
| Heading order     | FIX     | H2 "Setup" followed by H4                         |
| Answer-first lead | FIX     | opens with "In today's…"; no answer in 60 words   |
| FAQ               | FIX     | 3 entries in frontmatter, none rendered            |
| Structured data   | PASS    | Article + BreadcrumbList; dates match byline       |
| Canonical         | PASS    | self-referencing, https                           |
| Internal links    | PASS    | 7 links; all resolve                               |
| Images            | FIX     | 2 of 5 images lack alt                            |
 
Score: 5 of 10 pass.
 
Next: fix the lead and the FAQ first (geo-content-auditor), then the meta description.

Step 5 — Route each FIX

End with one line per FIX naming what executes it: lead, headings, and FAQ wording go to the geo-content-auditor skill; JSON-LD that is missing, mistyped, or not backed by visible content goes to the schema-markup-generator skill; canonical, robots, and link problems that recur across pages go to the seo-technical-auditor skill; a site with no /llms.txt goes to the llms-txt-generator skill. The per-article loop this command sits in is described in the SEO content workflow with Claude Code.

Filed under

Related