К основному содержанию
скилл CX Codex

first-screen-check — скилл для Claude Code

Скилл first-screen-check для Claude Code проверяет мобильную версию первого экрана сайта. Находит пропавшие кнопки звонка, обещания из рекламы и отзывы.

MIT ★ 778 обновлён 2026-11-16

Установить в Claude Code · Через маркетплейс плагинов Все способы и платформы ↓
/plugin marketplace add boringmarketer/boring-marketing-skills
/plugin install boring-marketing@boring-marketing-skills
Открыть на GitHub

Что делает first-screen-check

Анализирует посадочную страницу так, как её видит посетитель со смартфона с экраном 390 на 664 точки. Скилл сверяет текст первого экрана с обещаниями из объявления или профиля компании вроде круглосуточного графика или выезда в тот же день. По итогу выдаёт короткий список потерь и одно главное действие для исправления.

Для снятия скриншотов используется Playwright. Без предварительной команды npx playwright install chromium локальный браузер не запустится и снимок сорвётся. Если локального браузера нет, скилл обращается к сервису Firecrawl при наличии ключа FIRECRAWL_API_KEY. При отсутствии обоих вариантов скрипт на Python 3 анализирует сырой HTML страницы.

Какие задачи решает

  • найти расхождения между обещаниями в рекламе и текстом на первом экране сайта
  • проверить наличие кликабельного номера телефона для звонка в один клик
  • оценить видимость рейтинга компании и отзывов на экране смартфона
  • подготовить аудит мобильной посадочной страницы перед переговорами с клиентом

Примеры запросов агенту

  • Проверь первый экран мобильной версии для <url> с обещанием круглосуточного выезда
  • Моя реклама получает клики, но никто не звонит: проверь страницу <url>
  • Оцени мобильный экран сайта потенциального клиента перед встречей

Как установить first-screen-check

1Через маркетплейс плагинов

/plugin marketplace add boringmarketer/boring-marketing-skills
/plugin install boring-marketing@boring-marketing-skills

Команда ставит плагин boring-marketing целиком, со всеми его скиллами

2Через Skills CLI (Vercel)

npx skills add boringmarketer/boring-marketing-skills --skill first-screen-check -a claude-code

3Вручную: положить файлы

~/.claude/skills/first-screen-check/SKILL.md

Папка скилла целиком; для одного проекта — .claude/skills/first-screen-check/

По шагам, с командами для каждой платформы: Как установить скилл в Claude Code: пять способов

Вопросы о first-screen-check

Как установить first-screen-check в Claude Code?

Сначала добавьте каталог командой /plugin marketplace add boringmarketer/boring-marketing-skills. Затем установите сам плагин через /plugin install boring-marketing@boring-marketing-skills.

Что проверяет first-screen-check?

Скилл проверяет экран смартфона на наличие кнопки быстрого звонка, блока отзывов и текста рекламных обещаний. Результатом становится ранжированный список недочётов с первоочередной правкой.

Что нужно для работы first-screen-check?

Для скриншотов нужен установленный через npx playwright install chromium браузер или ключ FIRECRAWL_API_KEY. Если их нет, потребуется Python 3 для разбора HTML встроенным скриптом.

Исходный текст

SKILL.md · MIT
---
name: first-screen-check
description: Looks at a local business's site, or the page its ad opens, the way a customer sees it on a phone, and reports what the first screen is missing (the number with tap-to-call, the promise the ad made, the star rating and other proof, a way to book) plus the site faults that quietly break trust, as a short ranked list with one thing to fix first. Use when the site or the ads aren't making the phone ring, when someone wants a page checked on a phone, or when sizing up a prospect's site before a pitch.
when_to_use: check my site on a phone, why isn't my website getting calls, first screen check, is my phone number easy to find, check the page my ad goes to, does my page match my ad, my ads get clicks but no calls, what's wrong with this website, check a plumber's site, check a prospect's site, mobile site check
argument-hint: "[site URL or the URL an ad opens] [optional: the ad's text or promise]"
metadata:
  boring:
    version: 4.1.1
    layer: execution
    reads: [positioning#chosen-angle, competitors]
    writes: [assets]
    optional-keys: [FIRECRAWL_API_KEY]
    enhances_with:
      - toolset: web-scrape
        gains: "reads pages that build themselves with JavaScript and can return a phone-size screenshot when no local browser is installed"
---

# First Screen Check

Most local customers meet a business on a phone, from a Google search, a map listing or
an ad, and decide in the first screen whether to call, go back, or call the next
result. This skill looks at that screen the way they do and reports what's missing,
ranked by how many calls it costs, with one fix to make first. Done means a short list
an owner can act on today, each item backed by what the page actually shows. The checks
are a floor. When the business in front of you argues for a different order, change it
and say why.

## Context

Only if `./brand/`, `./brands/` or `MARKETING.md` exists (nothing to load otherwise), run:
`python3 ${CLAUDE_SKILL_DIR}/../start-here/scripts/brand_context.py --for first-screen-check`
(`${CLAUDE_SKILL_DIR}` is this skill's folder; outside Claude Code use the folder this
file was read from, e.g. `.agents/skills/first-screen-check`. If the script or python3 is
unavailable, read the brand folder directly: `./brands/<slug>/` when `./brands/active`
names one, else `./brand/`; then `MARKETING.md`, `preferences.md` and the files in
`reads:`. If none exist, proceed standalone. Ask only for the URL, and for the
ad's text if the page is an ad's landing page.)

Checking a prospect's site rather than the user's own? Don't load or write brand files
for it. The report lands in `./campaigns/` and nothing else changes.

No URL given, or only a placeholder (example.com, yourdomain.com)? Ask for the URL in one
line. Never run the checks or the redirect and structured-data commands against a
placeholder.

## What the data says

From hundreds of local business sites checked on a phone in November 2026.
The sample is self-selected (businesses that people chose to look up, mostly US home
services), so read the shares as "common", not as a census.

- Very few passed every check. Most had 3 or more gaps.
- Most businesses running ads that promise 24/7 or same-day service don't say it on the first screen of the page the ad opens, and many never say it anywhere on that page.
- Most have 4.5 stars or better and 100 or more reviews, and most of those don't show the rating on the first screen.

The common failures aren't about design taste. They are things the business already has (a promise it pays to advertise, a rating it earned) that never
reach the screen where the customer decides. Look for those first.

## Phase 1: Pick the page and the promise

**The page.** Check the page customers actually land on, not the one the owner thinks
of as "the site". In order of how much traffic it usually carries:

1. The page each ad opens (its final URL). Paid clicks are the most expensive visitors.
2. The page the Google Business Profile's website button opens (often the homepage,
   sometimes a location page).
3. The page that ranks for "<main service> <city>".

If several differ, check the ad page and the profile page. Two good checks beat five
thin ones.

**The promise.** Write down, word for word, what the customer was promised before
they tapped: the ad's headline and description, the profile's hours ("Open 24 hours"),
and any offer ("$49 drain cleaning", "free estimates"). Free ways to get the ad text:

- The user pastes it, or a screenshot from their ad account.
- Google's Ads Transparency Center (adstransparency.google.com): search the business
  name or domain, set the region, and read the text ads. It searches by advertiser, not
  by keyword.
- A search for "<service> <city>" on a phone in the service area. Ads change by location
  and hour, so say when and where you looked.

No ad running? The promise is whatever the profile and the search snippet said:
the hours, the services, the rating shown beside the listing.

## Phase 2: See the first screen on a phone

Take the best path the agent has. The report says only whether it rests on a phone
screenshot or on the page HTML.

**A. A browser on the machine (best).** Playwright takes a phone-size screenshot from
the command line. The first time on a machine, download its browser with
`npx playwright install chromium`; without it the screenshot command fails with
"Executable doesn't exist".

```bash
npx playwright install chromium   # once per machine
npx playwright screenshot --viewport-size "390,664" --wait-for-timeout 3000 --timeout 45000 <url> first-screen.png
npx playwright screenshot --viewport-size "390,664" --full-page --timeout 45000 <url> full-page.png
```

390 by 664 is what an iPhone 13 shows inside the browser bars. The 3-second wait lets
late popups, chat bubbles and cookie banners load, because the customer sees those too.
(`--device "iPhone 13"` also works but runs WebKit, which needs
`npx playwright install webkit` first.) Read both images. Count how many phone screens
tall the full page is and where the number, the promise and the rating first appear.

A browser extension or MCP browser works the same way: set the window to 390 wide,
load, wait, capture.

**B. No browser, Firecrawl keyed.** We use Firecrawl for this: its scrape returns a
phone-size screenshot and the rendered HTML with no browser installed, and it reads pages
that build themselves with JavaScript (free tier of 1,000 credits a month, no card). Ask for
the mobile screenshot and the rendered HTML. Same reading as A. Without it, path A or C
does the same check.

**C. Neither (fallback).** Read the HTML in order. If python3 is available, run:

```bash
python3 ${CLAUDE_SKILL_DIR}/scripts/page_facts.py <url> --promise "24/7" --promise "same-day"
```

It prints facts only: redirect hops, the H1 and where it starts, the opening words in
reading order (menus left out), tap-to-call links, phone numbers, where each promise
word and "reviews" first appear (as a word position), forms and their fields, the
structured data, the weight of the first images, and widgets that tend to cover the
screen. Without python3, WebFetch the page and read the same things by hand.

**Estimating the first screen from HTML.** It usually holds the header, the headline and
somewhere between 20 and 70 words of copy, fewer when a big photo, video or offer banner
sits above the headline. Anything the script places past about word 80 is usually below the
first screen. In Sources and assumptions, say once in plain words that the first screen is
estimated from the page HTML and that popups and JavaScript-built parts can't be seen this
way.

Two traps that make a page look better or worse than it is:
- **Call tracking swaps numbers.** The HTML can carry one number and the rendered page
  another. That is normal. What matters is that the number on screen rings the business
  and is the same one the customer can tap.
- **Rating badges load by JavaScript.** "Reviews" missing from the HTML doesn't prove
  the rating is missing from the screen. Only a screenshot settles it.

> Optional: boringfunnels (made by the author of this library, paid Pro with a free search) helps with this step: its prospect tool reads a named business's Google rating, newest reviews and ads, and its teardown opens one real page that sells well, section by section with phone-size pictures, to hold your read against. The free path above gets the same answer by hand.

## Phase 3: The first-screen checks

Each check is a question the customer asks in the first few seconds. Load
`references/checks.md` for how to verify each one, what passes, and the edge cases.

| # | The customer's question | What to look for on the first screen |
|---|---|---|
| 1 | Did I land where the ad said? | The promise from Phase 1 in the same words: "24/7", "same-day", the price, the free estimate |
| 2 | Can I call right now? | The number, visible, as one tap (a `tel:` link that dials the right number); a call bar that stays once they scroll |
| 3 | Is this the right kind of business? | A headline that says the service and the area in plain words |
| 4 | Are they any good? | Star rating and review count (when they're strong), years in business, licensed and insured, a guarantee |
| 5 | What if I can't talk right now? | A way to book or a short form reachable from the first screen |
| 6 | Why call today? | Availability ("techs available today"), a response time, or an offer |
| 7 | Is anything in the way? | A popup, chat bubble, cookie banner or offer bar covering the number or the headline |
| 8 | Does it show up fast? | The page appears quickly on a phone; the main photo isn't megabytes |

Judgment for the close calls:
- **Promise match is literal.** If the ad says "same-day" and the page says "fast
  service", that's a miss. The customer is scanning for the word they tapped on.
- **A rating only counts when it helps.** 4.8 from 300 reviews belongs under the
  headline. 3.9 from 11 doesn't, and the fix is in `/reviews`, not on the page.
- **Booking matters more for routine jobs.** For an emergency plumber the call is the
  whole game. For a dentist, a cleaner or a tune-up, people book from work at 9pm, and a
  missing booking link is a top-three item.
- **Speed is a fix only when it's slow enough to see.** Run PageSpeed Insights in a
  browser (pagespeed.web.dev, mobile tab). The keyless API often returns a quota error;
  a free API key from Google Cloud fixes it. The cheapest tell needs no tool: the
  weight of the main photo. A hero image over 1 MB on a phone page is a fix.

## Phase 4: The faults that break trust

These sit below the first screen, in the page source or the site structure, and cost
calls more slowly: through rankings, AI answers, and customers who feel misled. Check
them after Phase 3, and load `references/site-faults.md` for the commands and the
common versions of each. The short list:

- **Made-up prices.** A price range in the FAQ structured data or text that isn't what
  the business charges, such as an hourly range the owner never quotes.
- **Hardcoded review data.** A rating baked into the page's structured data (a perfect
  5.0 that never changes) that doesn't match the live Google profile. Google ignores
  self-serving review markup for local businesses anyway, so it earns nothing and tells
  on the site.
- **Old links that land on a 404.** Links from the Google profile, ads or other sites
  that redirect into an error page.
- **A multi-megabyte hero image.** A desktop-size photo sent to every phone.
- **Duplicate town pages.** Town pages that are the same page with the city swapped.
- **Numbers that don't match.** A different phone number on the site, in the structured
  data, or in a corner of the Google profile, so some calls reach a line nobody tracks.

## Phase 5: Rank and write

Rank by calls lost, then by effort. The default order, which you can change with a
stated reason:

1. **Anything that stops the call outright.** An ad or the profile opening an error
   page, no number on the first screen, a number that isn't tappable or dials wrong, a
   popup covering the screen.
2. **The paid promise missing.** Every click on that ad is paid for, and the visitor
   came for that word.
3. **Proof they have but hide.** Strong rating not shown, license and insurance buried.
4. **A headline that doesn't say what they do and where.**
5. **No way to act without calling** (move up for routine and booked services).
6. **Slow, heavy pages.**
7. **The trust faults from Phase 4.** Move a fault to the top when it's visible to
   customers (a fake price they'll be quoted against) or when it breaks a paid link.

Weight by traffic. A fault on the page every ad opens outranks the same fault on the
About page. A 10-minute fix that unblocks calls goes ahead of a redesign.

Keep the list short: 5 to 7 items. More than that and the owner fixes none of them.
**Fix first** is one item, chosen because it costs the most calls per minute of work,
with that reason in one sentence.

Every item names what the customer sees (quote the page or point at the screenshot),
why it costs calls in plain words, the fix (the actual words or element to add), and
rough effort. No item rests on a guess: if you couldn't see it, say so in the note.

## Output

Write `./campaigns/<business-slug>-first-screen/first-screen-check.md`, plus
`first-screen.png` and `full-page.png` beside it when a screenshot was taken. Append one
line to `./brand/assets.md` when a brand folder exists (never for a prospect). Then the
chat summary: header, the ranked list, **Files saved**, **What's next**.

Write the report for the owner, in plain words, with the screenshot first. Length: the
plan budget (about 1,200 to 2,500 words) is a ceiling, and five to seven items usually
need far less. **Fix first** carries item 1 in full and the table holds the rest in order,
so nothing is said twice. Sources and assumptions closes the report, 10 lines at most:
what the report rests on (page, date, screenshot or HTML), what was not checked, and what
the owner should confirm. A sample, for the sample business Hartline Plumbing, Dallas TX
(illustrative, not a real site):

```markdown
## FIRST SCREEN CHECK: hartlineplumbing.com (sample)
Nov 13, 2026 · the page the ad opens, /emergency-plumber-dallas, on a 390px phone

![First screen on a phone](first-screen.png)

**Fix first:** put "24/7" and "same-day" in the headline. The ad says "24/7 Emergency
Plumber, Same-Day" and the first screen says "Quality Plumbing Since 1998". A click that
doesn't see what the ad promised goes back to Google, and the click is still paid for.
New headline: "24/7 Emergency Plumber in Dallas. Same-Day Service." Ten minutes of work
on the page every paid click lands on.

**Then, in this order:**

| # | What a customer sees | Why it costs calls | Fix | Effort |
|---|---|---|---|---|
| 2 | The number is in the header as plain text | A caller who can't tap it rings the next result | Make it a tap-to-call link and add a call bar that stays on scroll | 15 min |
| 3 | 4.8 stars from 312 Google reviews, shown only in the footer | Their best proof never reaches the person choosing who to call | One line under the headline: "4.8 stars from 312 Google reviews" | 15 min |
| 4 | A chat bubble covers the call button after 3 seconds | The first thing they see is something to close | Move the bubble left or delay it until scroll | 5 min |
| 5 | The FAQ data says "$89-$149 per hour"; the owner quotes a flat price after seeing the job | A customer quoted differently feels baited, and AI answers can repeat the wrong price | Replace with the real pricing in words | 20 min |

**Sources and assumptions:**
- Based on a phone-size screenshot and the rendered page taken Nov 13, 2026, and the ad
  text the owner pasted.
- Not checked: the booking flow past the first step (needs a test booking).
- Confirm with the owner what they actually charge before changing the pricing text.
```

## When the user edits your work

Apply the edit, then make it stick: save a gold case to `./brand/gold/<this skill>/` and
one to three yes/no checks to `./brand/checks.md`, and confirm in one line. Before handing
over any work, answer the checks the context block loaded, silently, and fix every no.
How: `${CLAUDE_SKILL_DIR}/../start-here/references/taste-loop.md`.

## What's next

- `/landing-page`: build the page each ad should open, with the promise in the headline
  and the call bar built in (~30 min)
- `/reviews`: when the rating or the review count is the weak spot, fix the source
  before showing it off (~20 min)
- `/google-ads-tracking`: once the page gets calls, make sure the ads count them (~30 min)
- `/pitch-the-fix`: checking a prospect? Turn the fix-first item into a short email (~10 min)

© James Dickerson (@boringmarketer) · MIT · boringmarketer/boring-marketing-skills

Похожие скиллы

Все скиллы по теме «Маркетинг и контент» →
aesthetic-analysis qodex-ai Скилл aesthetic-analysis для Claude Code разбирает визуальный референс и переносит его композицию, цвета и типографику в React Native. Задача — редизайн лендинга по сайту или изображению. CX Codex +1 193установок screenshots-marketing nexu-io Скилл screenshots-marketing для Claude Code, Codex и Cursor генерирует маркетинговые скриншоты через Playwright. Запускается по фразам «marketing screenshot», «hero shot» или «app store screenshot». CX Codex +1 3 000установок capacitor-splash-screen Cap-go Скилл capacitor-splash-screen разбирается, почему в приложении Capacitor мелькает белый экран или сплэш не исчезает, и правит конфиг, тему Android 12+ и iOS storyboard. CX Codex +1 1 200установок screen-recording официальный github Скилл screen-recording для GitHub Copilot записывает действия в браузере или на рабочем столе в GIF и добавляет аннотации к кадрам. GIF показывают функции, баги и исправления. CX Codex +1 149установок user_screen wangzan101 Промпт user_screen для Claude Code помогает ответить на вопрос по снимку экрана телефона. Он опирается только на то, что видно на изображении, и не добавляет советов или инструкций. GP ChatGPT ★ 9 488звёзд репо mika_open_screen wangzan101 Промпт mika_open_screen для Claude Code открывает нужный экран Mika: чат, настройки, память, планы, браузер, модели, использование, список экранов или сведения о приложении. Он срабатывает на просьбу показать или изменить что-то в Mika. GP ChatGPT ★ 9 488звёзд репо

Из того же набора35

Весь набор →
speed-to-lead Скилл speed-to-lead для Claude Code анализирует историю звонков и заявок, выявляя медленные ответы. Агент строит отчёт и формирует график работы с лидами. CX Codex +1 ★ 778звёзд репо service-area-pages Скилл service-area-pages для Claude Code готовит страницы городов для местного бизнеса и проверяет их на дубли. Он отбирает города по подтверждённым данным о работе бизнеса и местных условиях. CX Codex +1 ★ 778звёзд репо reviews Скилл reviews для Claude Code находит пробелы в отзывах локального бизнеса в Google и готовит ответы владельца. Сравнивает профиль с топ-3 на карте и создаёт сценарии сбора отзывов после заказов. CX Codex +1 ★ 778звёзд репо find-local-clients Скилл find-local-clients для Claude Code ищет потенциальных клиентов среди компаний выбранной отрасли и города. Он включается по запросу составить список лидов или найти местные компании, которым нужна помощь с маркетингом. CX Codex +1 ★ 778звёзд репо creative Скилл creative для Claude Code создаёт изображения, видео и рекламные материалы для бренда. Он подключается к задачам по созданию или редактированию визуалов и подготовке кампаний. CX Codex +1 ★ 778звёзд репо site-audit Скилл site-audit для Claude Code проводит технический аудит сайта. Он проверяет доступность страниц для поисковиков и ИИ-ботов, анализируя структуру, ссылки и производительность. CX Codex +1 ★ 778звёзд репо audience-research Скилл audience-research для Claude Code ищет, где аудитория обсуждает продукт, и собирает её формулировки. Он включается перед позиционированием, работой над текстами или выходом на новый рынок. CX Codex +1 ★ 778звёзд репо programmatic-seo Скилл programmatic-seo для Claude Code проверяет идею и помогает собрать библиотеку SEO-страниц из шаблона и данных. Запускается, когда один шаблон запросов повторяется с разными модификаторами. CX Codex +1 ★ 778звёзд репо

Спонсоры проекта

Компании, которые помогают держать аналитику открытой