AI-Augmented Accessibility Framework

An automated testing and auditing engine combining Playwright, Axe-core, and AWS Bedrock AI to find accessibility violations and provide live GOV.UK Design System (GDS) compliant fixes.

---

Key Features

• Intelligent Auto-Fill: Traverses complex GOV.UK forms using domain-aware logic to reach deep-link pages.
• Live GDS Scraper: Dynamically pulls the latest guidance from the GOV.UK Design System to ensure recommendations are always up-to-date.
• AI Resolution Engine: Uses AWS Bedrock (Nova lite) to analyze technical Axe violations and rewrite failing HTML into GDS-compliant code.
• Cumulative Auditing: Scans multiple pages across a session and merges them into a single, high-fidelity HTML report.
• Actionable Fixes: Provides "Download Fix Snippet" buttons for developers to immediately test corrected HTML.

---

Architecture

The framework is divided into four logical layers:

1. Execution Layer (AXEScanner)

   • Leverages Playwright to navigate the service and Axe-core to perform technical audits.
   • State Management: Accumulates violations and screenshots in static concurrent maps.
   • Screenshot Mapping: Captures one full-page screenshot per URL for visual evidence.

2. Context Layer (GovUkScraper)

   • JSoup-based engine scrapes GDS Style and Component pages, providing "Ground Truth" context to prevent AI hallucinations.

3. Intelligence Layer (BedrockAgentAnalyser)

   • Bridges technical errors and human-readable guidance.
   • Processes violations in batches for AI accuracy and robust API communication.

4. Presentation Layer (HtmlReportGenerator)

   • Generates standalone, interactive HTML reports.
   • Fixes sourced via AI → Static Knowledge Base → Hardcoded GDS Fallbacks.
   • Visual Audit: Impact-coded badges and modal screenshot viewers.

---

Setup & Installation

Prerequisites

• Java 21+ (see pom.xml for version)
• Maven
• AWS credentials (configured for Bedrock access)
• Playwright browsers (run: mvn playwright:install)

Dependencies

Ensure the following are in your pom.xml:

• com.deque.html.axe-core:playwright
• org.jsoup:jsoup
• org.json:json
• org.slf4j:slf4j-simple (logging)
• software.amazon.awssdk:bedrockagentruntime (Bedrock integration)
• com.microsoft.playwright:playwright
• org.apache.logging.log4j:log4j-core and log4j-api
• com.fasterxml.jackson.core:jackson-databind and jackson-datatype-jsr310
• org.apache.commons:commons-lang3
• org.junit.jupiter:junit-jupiter-api and junit-jupiter-engine

---

Usage

Running a Scan

Add the scanner to your Playwright test suite:

@Test
void accessibilityAudit() {
   page.navigate("https://your-gov-service.gov.uk");
   AXEScanner.scan(page); // Repeat across multiple pages
}

@AfterAll
static void tearDown() {
   AXEScanner.generateFinalReport(); // Generates the cumulative AI report
}

Configuration

Pass standards via system properties:

-Dstandards.scan=wcag22aa,best-practice

---

Report Output

Reports are generated in:

• target/reports/accessibility-audit-[timestamp].html Screenshots are stored in:
• target/reports/screenshots/

---

Troubleshooting

• Ensure AWS credentials are set up for Bedrock access.
• Run mvn playwright:install if browsers are missing.
• Check target/surefire-reports/ for test logs.

---

License

See LICENSE file for details.