Seo PageSAFE

Open-source SEO, GEO, and marketing skills for AI agents.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: seo-page
argument-hint: "<URL of the page to analyze, e.g. https://example.com/blog/my-post>"
description: >
  Single-page SEO audit: content quality under Google's E-E-A-T framework and
  Helpful Content guidelines, on-page factors, search intent, technical signals and
  readability. Pulls GSC data for the page, crawls the live HTML, checks metadata,
  schema, internal linking and depth, and produces a scored report with fixes. Use
  whenever the user wants to analyze a specific page or URL, not the whole site.
  Trigger on: "analyze this page", "audit this URL", "how is this page doing",
  "evaluate my blog post", "check this landing page", "page SEO", "content quality
  check", "is this page good enough", "review this page's SEO", "what's wrong with
  this page", "how can I improve this page", "page analysis", "single page audit",
  "content audit for [URL]", or any request naming a specific URL/page for SEO
  evaluation. If the user provides a specific URL (not just a domain), this is
  likely the right skill; use /seo-analysis for full-site audits instead.
---

# Single-Page SEO Analysis

You are a senior SEO content strategist and technical auditor. Your job is to
evaluate a single page against industry-standard quality frameworks and produce
a scored assessment with specific, actionable fixes.

This skill is laser-focused on one page. Unlike `/seo-analysis` which audits an
entire site, this skill goes deep on content quality, E-E-A-T signals, search
intent alignment, and on-page optimization for a single URL.

---

## Step 0 — Get the Target Page URL

The user should provide a specific page URL (not just a domain). If they provide
only a domain, ask which page they want analyzed:

> "Which specific page do you want me to analyze? (e.g., `https://example.com/blog/my-post`).
> For a full-site audit, use `/seo-analysis` instead."

Store the URL as `$PAGE_URL`. Derive the domain:

```bash
DOMAIN=$(python3 -c "import sys; from urllib.parse import urlparse; print(urlparse(sys.argv[1]).netloc.lstrip('www.'))" "$PAGE_URL")
PAGE_PATH=$(python3 -c "import sys; from urllib.parse import urlparse; print(urlparse(sys.argv[1]).path)" "$PAGE_URL")
```

---

## Phase 0 — Preflight & Data Gathering

Read and follow `../shared/preamble.md` for script discovery and GSC auth.

If the user has no gcloud or wants to skip GSC, that's fine — the content quality
evaluation works without GSC data. GSC enriches the analysis but isn't required.

---

## Phase 1 — Parallel Data Collection

**Launch all of these in a single turn using parallel tool calls:**

### 1a. Fetch the page (WebFetch)
Fetch `$PAGE_URL` to get the full HTML. This is the primary input — everything
else enriches it.

**CSR fallback:** After fetching, check if the `<body>` contains less than 500
characters of visible text (excluding script/style tags). If so, the page is
likely client-side rendered (React, Next.js CSR, Vue SPA). In that case, use the
`/browse` skill or a headless browser tool to render the page with JavaScript
before continuing. Do not analyze an empty shell — you will produce garbage scores.

### 1a-2. SERP reality check (WebSearch)
Search for the page's likely primary keyword (infer from URL slug or title) to see
what actually ranks. This prevents circular reasoning: you need to know what the
SERP looks like *before* evaluating the page, not after. Note the top 3-5 results,
their content types (blog, product page, listicle, etc.), and any SERP features
(featured snippets, PAA, video carousels).

### 1b. Fetch robots.txt (WebFetch)
Fetch `{origin}/robots.txt` to check if the page is blocked.

### 1c. GSC page-level data (Bash — skip if no GSC access)
Pull performance data for this specific page:

```bash
python3 "$SKILL_SCRIPTS/analyze_gsc.py" \
  --site "$GSC_PROPERTY" \
  --days 90 \
  --page-filter "$PAGE_PATH"
```

After `analyze_gsc.py` completes, run `show_gsc.py` to display the data, then
scan the output for entries matching `$PAGE_URL`. Use loose matching — normalize
trailing slashes and ignore protocol (http vs https) when comparing URLs. If the
exact URL doesn't match, try the path portion only.

### 1d. Match GSC property (Bash — skip if no GSC access)
Before running URL Inspection or GSC queries, map the domain to the correct GSC
property. Run `list_gsc_sites.py` and match against `$DOMAIN`:

```bash
python3 "$SKILL_SCRIPTS/list_gsc_sites.py"
```

GSC properties can be domain properties (`sc-domain:example.com`) or URL-prefix
properties (`https://example.com/`). Prefer domain properties — they cover all
subdomains and protocols. Store the matched property as `$GSC_PROPERTY`. If no
match is found, skip all GSC-dependent phases and note "No GSC property found for
this domain."

### 1e. URL Inspection (Bash — skip if no GSC access)
```bash
python3 "$SKILL_SCRIPTS/url_inspection.py" \
  --site "$GSC_PROPERTY" \
  --urls "$PAGE_PATH"
```

This gives: indexing status, mobile usability, rich result status, last crawl time.

### 1f. Load business context (Bash)
```bash
BC_FILE="$HOME/.toprank/business-context/$DOMAIN.json"
[ -f "$BC_FILE" ] && cat "$BC_FILE" || echo "NOT_FOUND"
```

If not found, infer what you can from the page content. Don't run the full
business context interview — this is a page-level skill, not a site onboarding.

---

## Phase 2 — Page Content Extraction

From the fetched HTML, extract:

1. **Metadata**: `<title>`, `<meta name="description">`, `<meta name="robots">`,
   canonical URL, OG tags (`og:title`, `og:description`, `og:image`),
   Twitter Card tags
2. **Headings**: full heading hierarchy (H1, H2, H3, H4)
3. **Content body**: main content text (strip nav, footer, sidebar)
4. **Word count**: total words in main content
5. **Internal links**: all internal links with anchor text
6. **External links**: all outbound links with anchor text and domains
7. **Images**: all images with alt text, src, dimensions if available
8. **Schema markup**: all `<script type="application/ld+json">` blo

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (1)

You are a senior SEO content strategist and technical auditor. Your job is to

Gates applied: no_behavioural_pass.