landing-page-auditor

Files (2)

• SKILL.md
• html-template.md · references/

SKILL.md (first lines)

# Landing Page GEO & LLM Auditor

Audits any landing or service page across 48 checks in 10 categories. Given a URL, fetch all required
data, grade every check, calculate scores per category and overall, then render a standalone HTML report.

---

## Step 0 — Parse and Confirm the URL

If the user provided a URL in the invocation arguments, use it directly. If not, ask:
> "Please provide the URL of the page you'd like to audit."

**Normalisation rules:**
- Ensure the URL starts with `https://`
- Strip trailing slashes
- Keep the full path (e.g. `https://infrasity.com/claude-skills`, not just `infrasity.com`)
- Extract the bare domain for file naming (e.g. `infrasity.com`)
- Extract the domain root for auxiliary files (strip subdomains only if they differ — check both)

Echo back the resolved URL before proceeding:
> "Auditing `{URL}` — fetching data now..."

---

## Step 1 — Fetch All Required Data in Parallel

Run ALL of the following fetches simultaneously. Do not wait for one to complete before starting the next.

### 1a — Page Content Overview (WebFetch pass 1)

Fetch `{URL}` with this prompt:
> "Return the complete text content of the page. Include: all headings (H1, H2, H3, H4) with their exact levels and text, all paragraph text, all list items, the opening sentence and first 150 words of body text (excluding nav), any FAQ section content, testimonial or social proof text, pricing info, stats and numbers, target audience language, comparison language like 'unlike X' or 'instead of', negative scoping like 'not for X', measurable outcome claims like 'reduced X by Y%', all internal links with href and anchor text, all external links with href and anchor text, any visible date or 'last updated' text, any breadcrumb navigation, and any comparison tables."

### 1b — Page Schema and Technical (WebFetch pass 2)

Fetch `{URL}` with this prompt:
> "Return: 1) All JSON-LD schema blocks verbatim including their @type values, 2) All image tags with exact src and alt attributes — list every image, 3) The meta title tag content, 4) The meta description content, 5) The canonical URL from the canonical link tag, 6) The robots meta tag content, 7) Any BreadcrumbList schema, 8) Any Service, ProfessionalService, SoftwareApplication or Product schema, 9) Any FAQPage schema, 10) Whether page content is visible in raw HTML without JavaScript execution."

### 1c — robots.txt (WebFetch)

Fetch `{ROOT_DOMAIN}/robots.txt` with this prompt:
> "Return the complete robots.txt content. List every User-agent entry and its Allow/Disallow rules. Specifically call out rules for: GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, and any wildcard User-agent. Note the Sitemap directive URL if present."

### 1d — sitemap.xml (WebFetch)

Fetch `{ROOT_DOMAIN}/sitemap.xml` with this prompt:
> "Check if {FULL_URL} or any variant appears in this sitemap. Return the exact URL match, its lastmod date, changefreq, and priority. Also return the total number of URLs in the sitemap."

If the first sitemap is an index file referencing sub-sitemaps, fetch the most relevant sub-sitemap and repeat the check.

### 1e — llms.txt (WebFetch)

Fetch `{ROOT_DOMAIN}/llms.txt` with this prompt:
> "Return the full content of this llms.txt file. List every URL or path mentioned. Specifically check whether '{PAGE_PATH}' (the path portion of the target URL) or the full URL '{FULL_URL}' appears anywhere in the file."

### 1f — Server Response Time (Bash / curl)

```bash
RESPONSE_MS=$(curl -o /dev/null -s -w "%{time_starttransfer}" --max-time 15 "{URL}" | awk '{printf "%.0f", $1 * 1000}')
echo "TTFB: ${RESPONSE_MS}ms"
```

If curl is unavailable, mark check 10.3 as ⚠ Warning with note "Response time could not be measured."

---

## Step 2 — Grade All 48 Checks

Use the fetched data to evaluate every check below. Assign one of four verdicts:

| Verdict | Symbol | Score Weight | When to use |
|---|---|---|---|
| Pass | ✓ | 1.0 | Clearly present and meets the criteria |
| Warning | ⚠ | 0.5 | Partially present, improvable, or ambiguous |
| Fail | ✗ | 0.0 | Missing, broken, or clearly absent |
| Manual | ○ | Excluded | Requires human review to verify accurately |

Also write a short **evidence note** for each check (1 sentence max) — what you found or didn't find.

---

### Category 1 — Content Clarity & Directness (5 checks)

**1.1 — Clear opening sentence describing the service**

Look at the first sentence of body text (exclude navigation, hero buttons, GitHub badges).

- **Pass:** First sentence clearly states what the product/service is or does
- **Warn:** Opening exists but is preceded by several non-sentence elements (buttons, badges, CTAs) that push it past the visual fold
- **Fail:** No clear opening sentence; page leads with only a tagline, image, or marketing puff

**1.2 — Service definition in first 100 words ('X is a [type] that…')**

Count the first 100 words of actual body copy (strip nav links, CTA button text, icon labels).

- **Pass:** An explicit definitional sentence ("X is a [type] that..." or "X [verb]s [object] so that...") appears within the first 100 words
- **Warn:** Definition exists but appears after 100 words, or it's implied rather than stated
- **Fail:** No definitional sentence found in first 100 words

**1.3 — No unexplained jargon or acronyms**

Scan for acronyms (all-caps sequences 2-5 chars) and technical terms used without definition.

- **Pass:** All acronyms defined on first use (e.g. "GEO (Generative Engine Optimization)"), or no acronyms present
- **Warn:** 1–2 unexplained acronyms/terms but core product concepts are clear
- **Fail:** 3+ acronyms or key terms used without any explanation on first use

**1.4 — Use case statements present ('designed for X', 'ideal for Y teams')**

Look for explicit target audience language.

- **Pass:** "designed for", "built for", "ideal for", "best for", or a named target customer type (company, role, or industry) appears
- **Warn:** Audience implied by content context but never explicitly named
- **Fail:** No use case or target audience language found anywhere on the page

**1.5 — Negative scoping present ('not for X', 'we don't do Z')**

Look for explicit exclusions.

From Infrasity-Labs/dev-gtm-claude-skills, MIT.

More from Infrasity-Labs/dev-gtm-claude-skills