Web Visual Verification logo

Web Visual Verification

Community
kevintsai1202
web-visual-verification

Use this skill whenever you need to verify that a web UI actually works the way it's supposed to — clicking through flows, asserting state, catching console errors, taking screenshots across multiple viewports for visual review. Triggers on phrases like "驗證網頁", "verify the site", "Playwright tests", "visual regression", "RWD verification", "screenshot comparison", "responsive check", "console error check", "看看手機版有沒有壞", "視覺驗證", "Playwright 測試", "截圖比對", "RWD 驗證", "驗 sidebar", "看 console 有沒有錯", or any post-change moment where the user wants to know "did I break anything?". This is the runtime-behaviour verifier — use `web-content-audit` instead if the question is about file/data consistency rather than rendered behaviour.

Overview

Publisherkevintsai1202
Repositoryteaching-site-skills
Skill nameweb-visual-verification
Stars
52
Forks
10
Bundled files
Instructions only
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

    Published by kevintsai1202 on GitHub. Read the source before you install it.

Installation

Install the Web Visual Verification AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/kevintsai1202/teaching-site-skills/tree/main/web-visual-verification
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/kevintsai1202/teaching-site-skills.git /tmp/teaching-site-skills
mkdir -p .claude/skills
cp -r /tmp/teaching-site-skills/web-visual-verification .claude/skills/web-visual-verification
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Web Visual Verification in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Web Visual Verification on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Web Visual Verification is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Web Visual Verification

Schema authority: when verifying that data renders correctly, the source-of-truth field names come from _shared/domain-primitives.md. E.g. asserting unit.illustrations.length >= 1 matches §11's Coverage Floor.

Reference implementation: d:/GitHub/ai-workshop/scripts/verify-*.mjs for production verify-script patterns (multi-viewport, console-error capture, screenshot-on-fail).

This skill produces Playwright-based runtime verification scripts that drive the actual rendered site and answer "does it behave correctly?". The skill emerged from a teaching-website project but applies to any static or dynamic web UI.

When to Invoke

  • After any interaction-layer change (Stage 4 interactions, sidebar logic, theme, zoom).
  • After any visual asset change (Stage 5 — illustrations, screenshots).
  • After any layout / CSS change ("the sidebar collapsed weirdly", "the modal lost its backdrop").
  • Before tagging a release or handing off a corporate edition.
  • When the user asks "did that change break anything else?".

Do NOT invoke for file/data consistency checks (use web-content-audit).

The Four Script Roles — Use the Right One

Every Playwright script in this skill has exactly one of these jobs. Mixing roles produces unmaintainable scripts.

RolePrefixHas assertions?OutputsWhen to use
Verifyverify-*.mjsYes (assert.*)Pass/fail + screenshotsDefault — runtime regression check
Capturecapture-*.mjsNoScreenshots onlyNeed visual evidence without judgement (design review, change log)
Diagnosediagnose-*.mjsNo (but logs heavily)Console dump + DOM tree + screenshotsWhen a verify failed and you need to figure out why
Probecheck-*.mjs, find-*.mjsTargeted, often ad-hocConsole outputQuick CLI: "does this string appear?", "where is element X?"

The discipline: a verify script must fail loudly. A capture script must never fail (it's just evidence). A diagnose script is exploratory. A probe is a one-liner replacement for browser devtools.

Skeleton: Multi-Viewport Verify Script

This is the workhorse pattern. Adapt for any UX flow.

js
// scripts/verify-mobile.mjs
import { chromium, devices } from 'playwright';
import { resolve, dirname } from 'node:path';
import { fileURLToPath } from 'node:url';
import { mkdir } from 'node:fs/promises';
import assert from 'node:assert/strict';

const __dirname = dirname(fileURLToPath(import.meta.url));
const ROOT = resolve(__dirname, '..');
const SCREENSHOT_DIR = resolve(ROOT, 'data', 'mobile-verify');
const URL = process.env.URL || 'http://localhost:3000/';

const PROFILES = [
  { id: 'iphone-se',  device: devices['iPhone SE'] },
  { id: 'iphone-13',  device: devices['iPhone 13'] },
  { id: 'ipad-mini',  device: devices['iPad Mini'] },
  { id: 'desktop',    device: { viewport: { width: 1280, height: 900 }, userAgent: 'Mozilla/5.0' } },
];

async function verifyProfile(browser, profile) {
  const ctx = await browser.newContext({ ...profile.device, locale: 'zh-TW' });
  const page = await ctx.newPage();

  // === CRITICAL: collect runtime errors ===
  const errors = [];
  page.on('pageerror', e => errors.push(`[pageerror] ${e.message}`));
  page.on('console', m => { if (m.type() === 'error') errors.push(`[console.error] ${m.text()}`); });

  await page.goto(URL, { waitUntil: 'networkidle', timeout: 30000 });
  await page.waitForSelector('section.chapter', { timeout: 10000 });

  // === Per-profile assertions go here ===
  const isMobile = (profile.device.viewport?.width ?? 1280) <= 768;
  // Example: assert horizontal overflow doesn't exist
  const overflow = await page.evaluate(() => document.documentElement.scrollWidth > window.innerWidth);
  assert.equal(overflow, false, `${profile.id}: 水平溢出`);

  // Screenshot at key states
  await mkdir(SCREENSHOT_DIR, { recursive: true });
  await page.screenshot({ path: resolve(SCREENSHOT_DIR, `${profile.id}-home.png`), fullPage: false });

  // === Fail if any console errors collected ===
  assert.equal(errors.length, 0, `${profile.id} console errors:\n${errors.join('\n')}`);
  await ctx.close();
}

const browser = await chromium.launch({ headless: true });
try {
  for (const p of PROFILES) await verifyProfile(browser, p);
  console.log('✅ All profiles passed');
} finally { await browser.close(); }

Reusable Assertions (Copy When Adding New Verify Scripts)

Horizontal overflow (the most common layout bug)

js
const overflow = await page.evaluate(() => ({
  scrollWidth: document.documentElement.scrollWidth,
  innerWidth: window.innerWidth,
}));
assert.ok(overflow.scrollWidth <= overflow.innerWidth, `水平溢出 ${overflow.scrollWidth} > ${overflow.innerWidth}`);

Window scroll actually works (catches overflow-x: hidden implicit lock)

js
await page.evaluate(() => window.scrollTo(0, 1500));
const y = await page.evaluate(() => window.scrollY);
assert.ok(y > 0, 'window 無法捲動(檢查 ancestor 是否有 overflow:hidden)');

Sidebar overlay behaviour (mobile)

js
if (isMobile) {
  // Default: sidebar hidden
  const before = await page.locator('.sidebar').boundingBox();
  assert.ok(!before || before.x < 0, 'sidebar should be off-screen by default on mobile');

  // Open via hamburger
  await page.locator('.menu-toggle').click();
  await page.waitForTimeout(400);  // animation
  const after = await page.locator('.sidebar').boundingBox();
  assert.ok(after && after.x >= 0, 'sidebar should slide in after toggle');

  // Backdrop click closes
  await page.locator('.sidebar-backdrop').click();
  await page.waitForTimeout(400);
}

localStorage round-trip (progress persistence)

js
await page.locator('input[type=checkbox]').first().check();
await page.reload();
await page.waitForSelector('input[type=checkbox]');
const checked = await page.locator('input[type=checkbox]').first().isChecked();
assert.equal(checked, true, '勾選未持久化(檢查 localStorage key 與 file:// 限制)');

Content zoom isolated to .content (not sidebar)

js
const sidebarRect = await page.locator('.sidebar').boundingBox();
// Trigger zoom in
await page.evaluate(() => document.documentElement.style.setProperty('--content-zoom', '1.35'));
await page.waitForTimeout(100);
const sidebarRectAfter = await page.locator('.sidebar').boundingBox();
assert.equal(sidebarRect.width, sidebarRectAfter.width, 'sidebar should not be affected by zoom');

NetworkIdle vs domcontentloaded — Choose Carefully

waitUntil:When to use
'domcontentloaded'DOM exists but JS may still be running. Fast. Use when verifying static content or when you waitForSelector afterward anyway.
'networkidle'All XHR/fetch quiet for 500ms. Use when site depends on data loaded post-DOMContentLoaded (e.g. course-data.js injected COURSE renders units). Slower but robust.
'load'Only the document's load event. Often misleading — async data may still be pending. Avoid for SPAs.

Default to 'networkidle' for SPA verification; switch to 'domcontentloaded' + explicit waitForSelector only if you hit timeouts.

Screenshot Conventions

data/
├── mobile-verify/
│   ├── iphone-se-home.png
│   ├── iphone-se-sidebar-open.png
│   ├── iphone-13-home.png
│   └── desktop-home.png
├── toolbox-verify/
│   └── ...
└── design-audit/        ← capture-*.mjs goes here (no assertions, just evidence)

One folder per verify domain. Filename pattern: {profile-id}-{state}.png. Don't dump everything into data/.

Diagnose Script Pattern (When Verify Fails)

A diagnose script does NOT assert — it gathers context. Output is a dump of DOM, console, computed styles, screenshots.

js
// scripts/diagnose-day2-visibility.mjs
const html = await page.evaluate(() => {
  const day2 = document.querySelector('#day2');
  if (!day2) return 'NOT FOUND';
  return {
    boundingBox: day2.getBoundingClientRect().toJSON(),
    computedStyle: {
      display: getComputedStyle(day2).display,
      visibility: getComputedStyle(day2).visibility,
      opacity: getComputedStyle(day2).opacity,
      transform: getComputedStyle(day2).transform,
    },
    innerHTML: day2.innerHTML.slice(0, 500),
    childCount: day2.children.length,
  };
});
console.log(JSON.stringify(html, null, 2));

Run when a verify script reports "section.day2 not visible" — the diagnose gives you the actual display: none (or transform: translateY(20px) if fade-in observer didn't fire) within minutes.

Probe Script Pattern (Ad-hoc CLI)

These are 20-line one-shot tools. Don't over-engineer.

js
// scripts/check-text-absence.mjs "招生人數" "100 人"
const phrases = process.argv.slice(2);
const browser = await chromium.launch({ headless: true });
const page = await (await browser.newContext({ locale: 'zh-TW' })).newPage();
await page.goto(process.env.URL || 'http://localhost:3000/');
const body = await page.evaluate(() => document.body.innerText);
const found = phrases.filter(p => body.includes(p));
console.log(found.length ? `❌ Found: ${found.join(', ')}` : '✅ All absent');
await browser.close();
process.exit(found.length ? 1 : 0);

CLI args for the phrase list keeps it reusable. Don't wrap in a verify-* script.

Anti-Patterns

  • Verify scripts without assert.* — just a console.log("ok") is not a verify, it's a capture. Be honest about which role you're writing.
  • One mega-script that does everything — verify-mobile.mjs that also runs audit-design and diagnose-day2. Each script has one job. Compose at the package.json level ("verify:all": "verify-mobile && verify-toolbox && ...").
  • Ignoring pageerror / console.error — silent JS errors will sail through every assertion if you don't listen for them. ALWAYS attach the two listeners and assert empty at end.
  • Hardcoded localhost:3000 — always honour process.env.URL. The same script needs to run against deployed URLs in CI.
  • page.waitForTimeout(N) without comment — fine for animations (note why, e.g. "// 400ms slide-in animation"), bad as a "just wait and hope" tool. Prefer waitForSelector or waitForFunction.
  • Diagnose scripts that also assert — the moment they fail mid-run you lose the context you came for. Diagnose = log everything, exit 0.

CI Integration

json
{
  "scripts": {
    "verify": "node scripts/verify-mobile.mjs && node scripts/verify-toolbox.mjs && node scripts/verify-instructor-card.mjs",
    "capture": "node scripts/capture-overview.mjs && node scripts/capture-all-zones.mjs"
  }
}

Run verify in CI (must pass); run capture manually before design reviews.

Hand-off

When this skill finishes:

  • A set of verify/capture/diagnose/probe scripts is in scripts/.
  • Each has a one-line header comment explaining its role and a usage example.
  • npm run verify runs the full verify suite.
  • Screenshots organised under data/{domain}-verify/.

If anything failed and you don't know why → reach for the diagnose script you wrote alongside the verify. Don't add console.log to the verify itself.

Frequently asked questions

What does the Web Visual Verification AI skill do?

Use this skill whenever you need to verify that a web UI actually works the way it's supposed to — clicking through flows, asserting state, catching console errors, taking screenshots across multiple viewports for visual review. Triggers on phrases like "驗證網頁", "verify the site", "Playwright tests", "visual regression", "RWD verification", "screenshot comparison", "responsive check", "console error check", "看看手機版有沒有壞", "視覺驗證", "Playwright 測試", "截圖比對", "RWD 驗證", "驗 sidebar", "看 console 有沒有錯", or any post-change moment where the user wants to know "did I break anything?". This is the runtime-...

Why use Web Visual Verification on TypingMind?

Because you install it once and use it with any model. Web Visual Verification is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Web Visual Verification in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/kevintsai1202/teaching-site-skills/tree/main/web-visual-verification. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Web Visual Verification?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Web Visual Verification?

As many as you like. As long as a model supports skills, you can use Web Visual Verification with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Web Visual Verification AI skill free?

Yes. It is published on GitHub by kevintsai1202 under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇