К основному содержанию
скилл CX Codex

visant-preflight — скилл для Claude Code: что делает и как установить

Скилл visant-preflight для Claude Code проводит опрос стейкхолдера и проверяет шрифты, графику и брендбук перед стартом крупных клиентских проектов вроде редизайна сайта или запуска кампании.

MIT ★ 91 обновлён 2026-11-11

Установить в Claude Code · Через Skills CLI (Vercel) Все способы и платформы ↓
npx skills add pedrojaques99/visant-skills-public --skill visant-preflight -a claude-code
Открыть на GitHub

Что делает visant-preflight

Скилл запускает подготовку перед версткой сайта или созданием концепта. Процесс разбит на опрос ответственного за проект и проверку исходных данных, чтобы выявить нестыковки до начала работы над макетами и кодом.

Внутрь встроен скрипт scripts/preflight-scan.mjs, работающий через Node.js без внешних библиотек. Он обходит папку ассетов, определяет размеры и форматы файлов PNG, JPEG, WebP, AVIF, GIF, SVG, а также проверяет доступность шрифтов в Google Fonts. Для сетевой проверки шрифтов нужен доступ в интернет: без сети статус отобразится как неопределенный.

Скилл фиксирует ответы в markdown-файле проекта. Если запустить сканер с флагом --strict, процесс завершится с ошибкой при отсутствии изображений шире 2000 пикселей или сбое проверки гарнитур.

Какие задачи решает

  • согласовать ограничения и цели проекта с клиентом в один раунд вопросов
  • проверить разрешение и форматы файлов в каталоге ассетов
  • сверить доступность шрифтовых гарнитур через Google Fonts
  • найти расхождения между рабочими файлами заказчика и брендбуком

Примеры запросов агенту

  • Проведи preflight перед созданием нового сайта для клиента
  • Запусти visant-preflight для проверки шрифтов и графики в папке assets
  • Собери вводные вопросы для брифа по редизайну проекта

Как установить visant-preflight

1Через Skills CLI (Vercel)

npx skills add pedrojaques99/visant-skills-public --skill visant-preflight -a claude-code

2Вручную: положить файлы

~/.claude/skills/visant-preflight/SKILL.md

Папка скилла целиком; для одного проекта — .claude/skills/visant-preflight/

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

Вопросы о visant-preflight

Как установить visant-preflight в Claude Code?

Выполните в терминале команду npx skills add pedrojaques99/visant-skills-public --skill visant-preflight -a claude-code.

Что делает скилл visant-preflight?

Он формирует список обязательных вопросов к владельцу задачи и автоматизирует аудит картинок и шрифтов, чтобы избежать переделок из-за ошибочных исходников.

Что нужно для работы visant-preflight?

Требуется среда Node.js для запуска скрипта preflight-scan.mjs и подключение к интернету для проверки шрифтов через каталог Google Fonts.

Исходный текст

SKILL.md · MIT
---
name: visant-preflight
description: 'Preflight for a large client project: one question round with whoever owns the work, plus five premise checks (brand source of truth, loadable fonts, photo resolution, brand files, sourced numbers). Use BEFORE "new site", "site redesign", "new proposal", "criar site", "redesenhar o site", "nova proposta", "montar campanha", "hidratar a marca".'
metadata:
  family: projeto
  routes: visant-new-site
---

# Visant Preflight

Projects rarely go wrong in execution. They go wrong **in the premise**: the work
comes out well made on top of data that was wrong from the start, and the rework
costs days.

> **Nothing large starts without two things: the question round answered, and
> the five checks green or explicitly waived.**

The whole preflight takes 15 minutes. Each mistake it prevents has already cost
more than a day. The long stories are in [references/cases.md](references/cases.md).

## When to run

Run it when the task is **big enough to hurt**: new site, redesign, proposal with
mockups, campaign, brand hydration, migration. Signal: more than half a day of
work before the first client feedback. Not for a one-off tweak: preflight on a
small task is bureaucracy and you will (rightly) stop running it.

---

## Part 1: the question round

**This part is not optional and cannot be automated.** It is the one most often
missing.

Before designing anything, ask the questions below **to whoever owns the work**
(client, account lead, project owner) and write down the answers. Assume none.
If they answer "whatever", write "whatever": that is an answer too, and it turns
into explicit freedom instead of a silent bet.

Use the harness question tool for closed choices; use conversation for open
ones. **Do it in ONE round**, not drip-fed through the work: a question asked
mid-work always arrives after the rework it would have prevented.

### 1.1 The bet

- **What is the single sentence this work must prove?** If it does not fit in a
  sentence, the scope is not closed yet.
- **Which business variable does it move?** Conversion, retention, ticket, cost
  per meeting. If it moves none, it is decoration, and decoration costs
  attention.
- **Who is on the other side, and at what awareness stage do they arrive?**
  (Schwartz, see `revenue-centric-design`.) The headline speaks to the current
  stage; proof pushes one stage further.
- **What is NOT to be done?** The question that saves the most time and that
  nobody asks.

### 1.2 Reference and taste

- **Is there something the client already approved** (layout, piece, campaign)
  that I should inherit instead of reinventing? Mine before designing.
- **Is there something they hate?** Worth more than ten references they love.
- **What in this brand is untouchable?** A "clean" version that swaps the brand's
  own graphic language for a conventional one gets reverted in one sentence.
- **How much density, and do photos go in this round?** Three full proposals
  rebuilt over "too much text, no images" is the classic cost of skipping this.
- **How many alternatives do they want to see?** One polished or three distinct
  bets. Different jobs; agreeing first avoids delivering the wrong format.

### 1.3 What blocks publishing

- **What still depends on a third party?** Copy, photo, number, legal, domain.
- **Which of those is on the critical path?** It sets the work order: start with
  what depends on no one.
- **Is there a regulatory constraint?** (Professional councils, health agency,
  securities, privacy law such as LGPD.) Finding out after the layout is done
  means redoing the layout.

### 1.4 Definition of done

- **Where will this be seen?** Published link, PDF, Figma, call. Changes
  everything.
- **Is it going live or is it a proposal?** If a proposal, `noindex` and no
  sensitive real data.
- **When is the feedback conversation?** Sets the depth.

> **Required output of Part 1:** the answers written in the project's `.md`,
> before the first line of code or the first frame. An answer that stayed only in
> the conversation becomes an assumption next week.

---

## Part 2: the five checks

The two middle ones (fonts and photos) are automated:

```bash
node <skill-dir>/scripts/preflight-scan.mjs \
  --assets "<client files folder>" \
  --fonts "Display Family,Body Family" \
  --report PREFLIGHT.md
```

The script is plain node, no dependencies, and needs nothing Visant-specific:

| flag | what it does |
|---|---|
| `--assets <dir>` | walks any local folder (synced Drive, Dropbox, a repo's `public/`) and lists every `.png .jpg .jpeg .webp .avif .gif .svg` with width, weight and the role it supports. PNG, JPEG and WebP are measured; SVG is listed as vector; anything whose header it cannot read goes to a "skipped" list with the reason |
| `--fonts "A,B"` | checks each family against Google Fonts (needs network; offline shows as undetermined) |
| `--report <file.md>` | also writes the output to a markdown file |
| `--strict` | exit 1 if any item fails (no image reaches 2000px, or a font is missing from Google Fonts or could not be checked) |
| `--self-test` | validates the script's own parsers and thresholds |

At least one of `--assets` or `--fonts` is required; with neither it prints
usage and exits 2. Output is in English (a markdown report: photography table, skipped list, fonts table).

### 1. Brand source of truth: do the copies agree?

Wherever your brand data is stored for reuse (a brand vault, a tokens file, a
guidelines doc; at Visant, the Visant Labs vault) **is not the truth, it is a
copy**, and copies age or are born wrong. Cross it against the files actually
delivered to the client (the brand folder in Drive or wherever it lives) and the
design file (Figma).

Check in this order: **typography, colors, logo, tagline.** When they disagree,
the folder the client received is stronger evidence than the stored field.

To audit and reconcile Visant's vault, load **[[visant-brand-sync]]** (Visant internal).

> A stored typeface the brand never had shipped as three wrong typefaces over
> months, because nobody opened the client's font folder. See
> [references/cases.md](references/cases.md#1-brand-source-of-truth-client-k).

### 2. Do the fonts load?

Automated in the script. A family not on Google Fonts is not necessarily wrong:
a paid, licensed face is a legitimate decision. What cannot happen is for it to
**vanish silently**: either you have the file and load it via `next/font/local` /
`@font-face`, or you change the family.

Telltale symptom: the layout "is right" but the site looks like another brand.

> If the project consumes Visant's `@visant/brand-tokens`, the token engine
> already checks this at the source: it emits `@import` for what can load and
> **declares** at the top of the CSS what cannot.

### 3. Can the photography carry the layout?

Automated in the script. It lists **every** image with resolution and the role
it supports, and says explicitly what it skipped.

| Role | Minimum file width |
|---|---|
| Full-bleed band | 2000px |
| Hero portrait | 1000px |
| Grid cell | 700px |
| Thumb | 400px |

**Choose the source by resolution BEFORE deciding the framing**, never the other
way round. Upscaling shows, and no layout choice makes up for a small file.

Hence the script's rule: **never filter silently.** A scan that hides is worse
than a scan that does not run, because it produces false confidence, and false
confidence becomes a verdict. The case of the hidden 4562px originals:
[references/cases.md](references/cases.md#2-photo-resolution).

### 4. Do the brand files really exist?

Logo in **vector** (SVG/PDF), not a presentation PNG. Light/dark variants. Square
icon for the favicon. The brand's own graphic elements.

If they do not exist, **do not redraw the brand**: ask for them. A logo redrawn
by approximation is the most expensive failure and the hardest to reverse.

### 5. Does every number have a source?

Every piece of data going live needs its origin noted next to it, in the content
file. A number lifted from a layout **is not a primary source**: it goes up
marked as pending, or it does not go up.

> A number that does not hold up is not proof, it is liability. When the client
> asks where it came from, the answer has to travel with the data, not in the
> memory of whoever assembled it.

---

## Part 3: scars, do not reintroduce the bug you already killed

A visual defect fix **lives in the code as a comment at the fix site**, with the
why. And there is a specific way to lose it: **rewriting the whole file**, which
starts from the new intent and not from the old file.

Two rules, and they are cheap:

1. **Before rewriting a file completely, read it and list the defensive
   comments** (the ones saying "opaque bg is required", "measured in browser",
   "do not swap for", "without this X breaks"). Carry each one to the new
   version, or justify in writing why it no longer applies.
2. **Every fix found in the browser becomes a line in the Scars section** of the
   project's `.md`. It is what survives the rewrite and a change of hands.

```markdown
## Scars
- The bar field belongs to the hero: the proof band needs an opaque `bg`, or the
  decoration cuts through the number. (measured, 08-13)
- Light nav over a light field disappears. Two-tone nav → logo + action only.
- Photo: choose the source by resolution BEFORE framing.
```

**Why it did not become an automatic detector:** a line-based rule cannot tell
the real case from dozens of legitimate edges. A detector that shouts out of
scope is one people turn off; a written scar costs one line and has no false
positives. The story: [references/cases.md](references/cases.md#4-the-scar-that-reopened).

---

## The report

The preflight ends in a block in the project's `.md`:

```markdown
## Preflight: <date>

**The bet:** <the single sentence>
**Variable:** <what it moves>
**Not to be done:** <what was left out>
**Delivery format:** <link / proposal / PDF>, feedback on <when>

| Check | Status |
|---|---|
| Brand source of truth | ✅ / ⚠ <what diverged> |
| Loadable fonts | ✅ / ⚠ <family and plan> |
| Photo resolution | ✅ ceiling <N>px / ⚠ <what cannot be done> |
| Brand files | ✅ / ⚠ <what is missing> |
| Sourced numbers | ✅ / ⚠ <what is pending> |

**Critical path:** <what blocks publishing and who it depends on>
```

A ⚠ does not stop you from starting. It stops you from **pretending it does not
exist**, and sets the order: what does not depend on the blocked item goes first.

## Golden rule

**Every check here was born from a mistake that already happened.** A sixth only
goes in after it costs someone a day. One that catches nothing for a whole year
comes out: a checklist that never flags trains people to sign without reading.

## Neighbors

- **[[visant-design]]**: one screen, no project around it
- **[[visant-brand-sync]]** (Visant internal): reconcile the vault when the preflight finds a divergence
- **[[visant-new-site]]** (Visant internal): the client site flow, which starts with this preflight
- **[[revenue-centric-design]]**: awareness stage, ICP, proof
- **[[visant-frontend]]**: classify the surface after the preflight passed
- **[[visant-killer]]**: the exit gate at the end; this is the entry gate
- **[[visant-demanda]]** (Visant internal): graphic material for a client, not a site or big redesign

© pedrojaques99 · MIT · pedrojaques99/visant-skills-public

Похожие скиллы

Все скиллы по теме «Дизайн» →
brand-kit dotslashsimran Скилл brand-kit для Claude Code считывает оформление компании с её сайта и сохраняет его как бренд-кит. Запускается перед созданием видео, стикеров или графики для бренда. CX Codex +1 ★ 541звёзд репо dicom-series-preflight NVIDIA MedTech Team Скилл dicom-series-preflight для Claude Code проверяет заголовки серии DICOM перед конвертацией или инференсом. Не выполняет деидентификацию или клиническую очистку. CX Codex +1 1 900установок Asset Manager daffy0208 Скилл Asset Manager наводит порядок в папках проекта: сортирует картинки и иконки по правилам именования, сжимает графику и нарезает адаптивные форматы. CX Codex +1 165установок preflight haabe Скилл preflight проверяет готовность задачи до начала реализации: бюджет, контекст, объём, технические условия и критерии приёмки. Запускайте перед новым этапом разработки. CX Codex +1 ★ 54звёзд репо modernize-preflight официальный Anthropic Команда modernize-preflight проверяет, потянет ли окружение анализ и модернизацию системы, и говорит, что починить до запуска остальных команд. CX Codex +1 ★ 47 914звёзд репо spark-preflight Seth Hobson Команда spark-preflight проверяет систему DGX Spark перед запуском обучения или инференса и пишет отчёт env-report.json с вердиктом. CX Codex +1 ★ 42 780звёзд репо

Из того же набора12

Весь набор →
visant-methodology Скилл visant-methodology для Claude Code проводит брендинг через каскад из 9 этапов и 3 фаз. Запускается по запросам о полной стратегии бренда или аудите платформы. CX Codex +1 ★ 91звёзд репо visant-design Скилл visant-design для Claude Code перерабатывает интерфейсы и проверяет результат на экране приложения. Запускается на запросы об улучшении UI/UX, редизайне или удалении шаблонных AI-элементов. CX Codex +1 ★ 91звёзд репо site-reels Скилл site-reels для Claude Code записывает сценарии веб-сайтов в MP4 1080p для Figma Slides и собирает продуктовые шоурилы. CX Codex +1 ★ 91звёзд репо visant-mcp-connect Скилл visant-mcp-connect для Claude Code подключает инструменты Visant Labs через MCP. Он срабатывает на просьбы настроить Visant, вызвать его инструменты или извлечь данные из PDF. CX Codex +1 ★ 91звёзд репо visant-video-editor Скилл visant-video-editor для Claude Code собирает видео из исходных материалов: анализирует клипы, предлагает список склеек и готовит ролик. Включается по запросу смонтировать влог, урок или нарезать тишину. CX Codex +1 ★ 91звёзд репо visant-premiere-editor Скилл visant-premiere-editor для Claude Code собирает черновой монтаж в Adobe Premiere Pro из JSON-плана нарезки. Переносит таймлайн через XML и MCP без потери кадров. CX Codex +1 ★ 91звёзд репо visant-refine Скилл visant-refine для Claude Code помогает оформлять графики, шкалы, спарклайны и панели управления в едином визуальном стиле. Подключается по просьбе добавить график, слайдер симулятора или доработать интерфейс. CX Codex +1 ★ 91звёзд репо visant-copy Скилл visant-copy для Claude Code пишет и правит тексты на pt-BR в голосе Visant, а copy-lint ловит штампы ИИ на pt и en. Запускается на просьбах про заголовок или тон голоса. CX Codex +1 ★ 91звёзд репо

Спонсоры проекта

Компании, которые помогают держать аналитику открытой