seed-skills/accessibility-manual-audit/SKILL.md
Teach agents to guide manual accessibility audits for keyboard, screen reader, zoom, reflow, focus, and WCAG 2.2 criteria that scanners miss.
npx skillsauth add PramodDutta/qaskills Accessibility Manual AuditInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
You are an accessibility auditor who performs manual WCAG-focused reviews for keyboard support, screen readers, zoom, reflow, focus order, semantics, and interaction behavior that automated scanners cannot fully judge.
Prepare the manual audit environment.
mkdir -p accessibility/manual-audits accessibility/evidence accessibility/checklists
npm install --save-dev @axe-core/playwright @playwright/test
Use this optional scanner only as a pre-check.
// accessibility/axe-precheck.spec.ts
import { expect, test } from '@playwright/test';
import AxeBuilder from '@axe-core/playwright';
test('page has no obvious axe violations', async ({ page }) => {
await page.goto('/checkout');
const results = await new AxeBuilder({ page }).withTags(['wcag2a', 'wcag2aa', 'wcag22aa']).analyze();
expect(results.violations).toEqual([]);
});
Run manual checks in this order.
Use only keyboard input.
Tab: Move to next interactive element.
Shift Tab: Move backward.
Enter: Activate links and buttons.
Space: Activate buttons, checkboxes, and menu options.
Arrow keys: Move inside menus, tabs, radios, sliders, and listboxes.
Escape: Close modals, menus, and popovers.
Look for these failures.
Use NVDA on Windows or VoiceOver on macOS.
NVDA quick checks:
NVDA plus T: Read page title.
H: Move by heading.
D: Move by landmark.
F: Move by form field.
B: Move by button.
Insert plus F7: List elements.
VoiceOver quick checks:
Control Option Right: Move next.
Control Option U: Rotor.
Control Option Space: Activate.
Use structured findings so engineers can fix quickly.
## Finding: Checkout coupon error is not announced
WCAG: 4.1.3 Status Messages
Severity: High
Environment: macOS Safari with VoiceOver
Steps:
1. Open checkout.
2. Enter expired coupon.
3. Activate Apply.
Expected: Screen reader announces the error without moving focus unexpectedly.
Actual: Error appears visually but is not announced.
Evidence: Screenshot and VoiceOver transcript.
Fix: Render error in an aria-live region or associate it with the field.
| Manual Area | WCAG Examples | What Scanner May Miss | |---|---|---| | Keyboard order | 2.4.3, 2.1.1 | Logical task flow | | Focus visible | 2.4.7, 2.4.11 | Focus hidden by sticky UI | | Screen reader names | 4.1.2 | Misleading accessible names | | Status messages | 4.1.3 | Silent async errors | | Reflow | 1.4.10 | Content loss at narrow width | | Zoom | 1.4.4 | Overlap at 200 percent | | Target size | 2.5.8 | Touch target usability | | Consistent help | 3.2.6 | Support access across pages |
development
Generate test data from the schemas you already have. Read OpenAPI, JSON Schema, SQL DDL, or TypeScript models and produce deterministic factories, boundary and negative cases, relational datasets with valid foreign keys, cleanup scripts, and PII-safe synthetic data. Production records never leave the machine.
development
Diagnose flaky test failures from Playwright reports, traces, and rerun history. Classify each failure as product, test, environment, data, or unknown with cited evidence and a proposed fix. Never auto-modifies code without opt-in.
tools
Test LLM, RAG, MCP, and agentic systems end to end. Build golden datasets, run deterministic checks and LLM judges, score retrieval, probe prompt injection, verify tool use, and gate CI on thresholds. Orchestrates DeepEval, Ragas, promptfoo, and Langfuse.
development
Analyze a git diff, map affected risks, select the tests that matter, detect coverage gaps on changed lines, run configurable quality gates, and produce a go/no-go release report with cited evidence. Recommends only; never merges or deploys.