Use when writing acceptance criteria for a task - express each as an observable Given/When/Then that QA can execute, including negative cases
日本語の概要は準備中です。原文の説明を表示しています。
Use when a task changes any screen, form, dialog, menu or control - Lighthouse/axe scan of the changed screens, a keyboard walk, and the thresholds that fail a task
インストール方法を見るインストールする前に、エージェントに与えられる指示の中身を確認できます。
The same bar the frontend developer builds to (accessibility-basics, ui-quality-floor, responsive-mobile-first) — QA and developer share one bar, not a separate QA opinion:
aria-label on icon-only buttons;<input> font ≥16px on phone;<h1>, lang set on <html>, no user-scalable=no;Escape closes it, focus returns to the trigger;aria-invalid.QA=${TMPDIR:-/tmp}/tt-<task-key>/qa; mkdir -p "$QA"; node -v # lighthouse@13 needs Node >=22, else lighthouse@12
CHROME_PATH="$CHROME_BIN" npx -y lighthouse@13 "http://localhost:5173/settings" \
--only-categories=accessibility,best-practices --form-factor=mobile \
--screenEmulation.mobile --screenEmulation.width=360 --screenEmulation.height=800 --screenEmulation.deviceScaleFactor=2 \
--chrome-flags="--headless=new" --output=json --output-path="$QA/lh-settings-360.json" --quiet
jq -r '.audits|to_entries[]|select(.value.score==0)|"\(.key)\t\(.value.title)"' "$QA/lh-settings-360.json" # no jq: node -e 'for (const [k,v] of Object.entries(require(process.argv[1]).audits)) if (v.score===0) console.log(k+"\t"+v.title)' "$QA/lh-settings-360.json"
Run with run_terminal timeout_seconds: 300. errors-in-console in the output doubles as the console check for page load.
Install playwright-core + @axe-core/playwright into $QA, never the workspace:
cd "$QA" && [ -d node_modules/@axe-core/playwright ] || npm i --no-audit --no-fund --silent playwright-core@1.63 @axe-core/playwright@4.13
cat > axe.cjs <<'EOF'
const { chromium } = require('playwright-core');
const { AxeBuilder } = require('@axe-core/playwright');
(async () => {
const [url, width = '360'] = process.argv.slice(2);
const browser = await chromium.launch({ executablePath: process.env.CHROME_BIN });
const page = await browser.newPage({ viewport: { width: Number(width), height: 800 } });
await page.goto(url);
const { violations } = await new AxeBuilder({ page })
.withTags(['wcag2a', 'wcag2aa', 'wcag21a', 'wcag21aa', 'wcag22aa']).analyze();
for (const v of violations) console.log(`${v.impact}\t${v.id}\t${v.nodes.length}\t${v.nodes[0].target.join(' ')}`);
console.log(`violations=${violations.length}`);
await browser.close();
})();
EOF
node axe.cjs http://localhost:5173/settings 360
For a page behind login, add page.fill/page.click lines before the scan.
cd "$QA"; PW="npx -y @playwright/cli@0.1"; $PW --help | head -5
$PW open --browser=chrome http://localhost:5173/settings
$PW press Tab; $PW --raw eval "(() => { const a = document.activeElement; return a.tagName + '|' + (a.getAttribute('aria-label') || a.textContent.trim().slice(0, 40)) + '|outline=' + getComputedStyle(a).outlineStyle })()"
Repeat Tab through every changed control; order must follow visual order, with no outline=none and no box-shadow silently replacing it. Open the dialog, press Escape, eval activeElement = the trigger.
category: visual and a title prefix a11y: — the category enum has no accessibility value of its own.まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Use when writing acceptance criteria for a task - express each as an observable Given/When/Then that QA can execute, including negative cases
日本語の概要は準備中です。原文の説明を表示しています。
Use when the diff adds or changes an endpoint, resolver, RPC, job or query that takes an object id, a role check, a request binding or a tenant filter - BOLA/IDOR, function-level authorization, mass assignment and tenant scoping
日本語の概要は準備中です。原文の説明を表示しています。
Use on every UI change - semantic HTML, labels for controls, keyboard-navigable dialogs/menus, visible focus, and never color as the only signal
日本語の概要は準備中です。原文の説明を表示しています。
How to work a task returned with review, QA or UAT findings. Use when a task is in need_revision or PR review comments are in your context.
日本語の概要は準備中です。原文の説明を表示しています。
Use when deciding whether a request needs an analiz task before implementation - the conditions that require the architect's analysis versus going straight to implementation
日本語の概要は準備中です。原文の説明を表示しています。
Use when you write or revise the analiz deliverable - the ONE self-contained HTML report (spec and plan as sections) a human reviews passage by passage
日本語の概要は準備中です。原文の説明を表示しています。