npm - agentic-qe - Versions diffs - 3.7.9 → 3.7.11 - Mend

agentic-qe 3.7.9 → 3.7.11

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (401) hide show

package/assets/skills/accessibility-testing/evals/accessibility-testing.yaml CHANGED Viewed

@@ -1,719 +1,719 @@
-# =============================================================================
-# AQE Accessibility Testing Skill Evaluation Suite v1.0.0
-# WCAG 2.2 compliance testing evaluation with POUR principle coverage
-# =============================================================================
-#
-# This evaluation suite validates accessibility testing skill behavior through:
-# 1. POUR principle coverage (Perceivable, Operable, Understandable, Robust)
-# 2. WCAG 2.2 Level A, AA, AAA conformance testing
-# 3. Multi-model consistency across Claude and GPT models
-# 4. Severity classification validation
-# 5. Remediation quality assessment
-#
-# Schema: .validation/schemas/skill-eval.schema.json
-# Runner: scripts/run-skill-eval.ts
-# =============================================================================
-skill: accessibility-testing
-version: 1.0.0
-description: >
-  Comprehensive evaluation suite for WCAG 2.2 accessibility testing skill.
-  Tests POUR principles (Perceivable, Operable, Understandable, Robust),
-  conformance levels, finding detection, and remediation quality across
-  multiple LLM models.
-# =============================================================================
-# Multi-Model Configuration
-# =============================================================================
-models_to_test:
-  - claude-3.5-sonnet    # Primary model (high accuracy expected)
-  - claude-3-haiku       # Fast model (ensure minimum quality)
-  - gpt-4o               # Cross-vendor validation
-# =============================================================================
-# MCP Integration Configuration
-# =============================================================================
-mcp_integration:
-  enabled: true
-  namespace: skill-validation
-  # Query existing accessibility patterns before running evals
-  query_patterns: true
-  # Track each test outcome for learning feedback
-  track_outcomes: true
-  # Store successful patterns (WCAG violations, remediation approaches)
-  store_patterns: true
-  # Share learning with fleet coordinator agents
-  share_learning: true
-  # Update quality gate with accessibility metrics
-  update_quality_gate: true
-  # Agents to share learning with
-  target_agents:
-    - qe-learning-coordinator
-    - qe-queen-coordinator
-    - qe-accessibility-auditor
-# =============================================================================
-# ReasoningBank Learning Configuration
-# =============================================================================
-learning:
-  store_success_patterns: true
-  store_failure_patterns: true
-  pattern_ttl_days: 90
-  min_confidence_to_store: 0.7
-  cross_model_comparison: true
-# =============================================================================
-# Result Format Configuration
-# =============================================================================
-result_format:
-  json_output: true
-  markdown_report: true
-  include_raw_output: false
-  include_timing: true
-  include_token_usage: true
-# =============================================================================
-# Environment Setup
-# =============================================================================
-setup:
-  required_tools:
-    - jq
-  environment_variables:
-    WCAG_VERSION: "2.2"
-    TARGET_LEVEL: "AA"
-  fixtures:
-    - name: accessible_page
-      content: |
-        <!DOCTYPE html>
-        <html lang="en">
-        <head><title>Accessible Page</title></head>
-        <body>
-          <header><nav aria-label="Main"><a href="/">Home</a></nav></header>
-          <main>
-            <h1>Welcome</h1>
-            <img src="hero.jpg" alt="Person using laptop">
-            <form>
-              <label for="email">Email:</label>
-              <input type="email" id="email" name="email" required>
-              <button type="submit">Subscribe</button>
-            </form>
-          </main>
-        </body>
-        </html>
-    - name: inaccessible_page
-      content: |
-        <!DOCTYPE html>
-        <html>
-        <head><title></title></head>
-        <body>
-          <div onclick="navigate()">Menu</div>
-          <img src="logo.png">
-          <input type="text" placeholder="Enter email">
-          <span style="color: #999; background: #fff;">Light gray text</span>
-        </body>
-        </html>
-# =============================================================================
-# Test Cases - POUR Principles Coverage
-# =============================================================================
-test_cases:
-  # -------------------------------------------------------------------------
-  # PERCEIVABLE (WCAG 1.x) - Content can be perceived
-  # -------------------------------------------------------------------------
-  - id: tc001_perceivable_alt_text
-    description: "Detect missing alt text on images (WCAG 1.1.1)"
-    category: perceivable
-    priority: critical
-    input:
-      code: |
-        <img src="product.jpg">
-        <img src="banner.png" alt="">
-        <img src="hero.jpg" alt="Team collaboration in modern office">
-      context:
-        language: html
-        wcagLevel: AA
-    expected_output:
-      must_contain:
-        - "alt"
-        - "1.1.1"
-        - "perceivable"
-      must_not_contain:
-        - "no issues"
-      finding_count:
-        min: 1
-        max: 2
-      severity_classification: serious
-    validation:
-      schema_check: true
-      keyword_match_threshold: 0.9
-      reasoning_quality_min: 0.7
-  - id: tc002_perceivable_color_contrast
-    description: "Detect insufficient color contrast (WCAG 1.4.3)"
-    category: perceivable
-    priority: critical
-    input:
-      code: |
-        <p style="color: #777; background-color: #fff;">Gray text on white</p>
-        <p style="color: #333; background-color: #fff;">Dark text on white</p>
-        <h1 style="color: #aaa; background-color: #eee;">Low contrast heading</h1>
-      context:
-        language: html
-        wcagLevel: AA
-    expected_output:
-      must_contain:
-        - "contrast"
-        - "1.4.3"
-        - "4.5:1"
-      finding_count:
-        min: 1
-        max: 3
-    validation:
-      schema_check: true
-      keyword_match_threshold: 0.8
-  - id: tc003_perceivable_video_captions
-    description: "Detect videos without captions (WCAG 1.2.2)"
-    category: perceivable
-    priority: critical
-    input:
-      code: |
-        <video src="product-demo.mp4" controls>
-          <source src="product-demo.mp4" type="video/mp4">
-        </video>
-        <video controls>
-          <source src="interview.mp4" type="video/mp4">
-          <track kind="captions" src="captions.vtt" srclang="en">
-        </video>
-      context:
-        language: html
-        wcagLevel: AA
-    expected_output:
-      must_contain:
-        - "caption"
-        - "1.2.2"
-        - "track"
-      finding_count:
-        min: 1
-        max: 2
-    validation:
-      schema_check: true
-      keyword_match_threshold: 0.8
-  # -------------------------------------------------------------------------
-  # OPERABLE (WCAG 2.x) - Interface is operable
-  # -------------------------------------------------------------------------
-  - id: tc004_operable_keyboard_access
-    description: "Detect elements not keyboard accessible (WCAG 2.1.1)"
-    category: operable
-    priority: critical
-    input:
-      code: |
-        <div onclick="handleClick()">Click me</div>
-        <span class="button" onclick="submit()">Submit</span>
-        <button onclick="save()">Save</button>
-        <a href="/next">Next Page</a>
-      context:
-        language: html
-        wcagLevel: A
-    expected_output:
-      must_contain:
-        - "keyboard"
-        - "2.1.1"
-        - "operable"
-        - "button"
-      must_not_contain:
-        - "all elements accessible"
-      finding_count:
-        min: 2
-        max: 3
-      severity_classification: critical
-    validation:
-      schema_check: true
-      keyword_match_threshold: 0.9
-      reasoning_quality_min: 0.8
-  - id: tc005_operable_focus_visible
-    description: "Detect missing focus indicators (WCAG 2.4.7)"
-    category: operable
-    priority: high
-    input:
-      code: |
-        <style>
-          button:focus { outline: none; }
-          a:focus { outline: 0; }
-          input:focus { outline: none; border-color: blue; }
-        </style>
-        <button>Submit</button>
-        <a href="/">Home</a>
-        <input type="text" placeholder="Search">
-      context:
-        language: html
-        wcagLevel: AA
-    expected_output:
-      must_contain:
-        - "focus"
-        - "2.4.7"
-        - "outline"
-      finding_count:
-        min: 2
-        max: 3
-    validation:
-      schema_check: true
-      keyword_match_threshold: 0.8
-  - id: tc006_operable_keyboard_trap
-    description: "Detect keyboard traps (WCAG 2.1.2)"
-    category: operable
-    priority: critical
-    input:
-      code: |
-        <div id="modal" tabindex="0" onkeydown="if(event.key==='Tab'){event.preventDefault();}">
-          <h2>Modal Dialog</h2>
-          <input type="text" placeholder="Enter name">
-          <button>Close</button>
-        </div>
-      context:
-        language: html
-        wcagLevel: A
-        description: "Modal with keyboard trap"
-    expected_output:
-      must_contain:
-        - "keyboard trap"
-        - "2.1.2"
-        - "Tab"
-      severity_classification: critical
-    validation:
-      schema_check: true
-  # -------------------------------------------------------------------------
-  # UNDERSTANDABLE (WCAG 3.x) - Content is understandable
-  # -------------------------------------------------------------------------
-  - id: tc007_understandable_form_labels
-    description: "Detect form inputs without labels (WCAG 3.3.2)"
-    category: understandable
-    priority: high
-    input:
-      code: |
-        <form>
-          <input type="text" placeholder="Name">
-          <input type="email" placeholder="Email">
-          <label for="phone">Phone:</label>
-          <input type="tel" id="phone">
-          <button>Submit</button>
-        </form>
-      context:
-        language: html
-        wcagLevel: A
-    expected_output:
-      must_contain:
-        - "label"
-        - "3.3.2"
-        - "understandable"
-      finding_count:
-        min: 2
-        max: 3
-    validation:
-      schema_check: true
-      keyword_match_threshold: 0.8
-  - id: tc008_understandable_error_messages
-    description: "Detect missing error identification (WCAG 3.3.1)"
-    category: understandable
-    priority: high
-    input:
-      code: |
-        <form>
-          <input type="email" required aria-invalid="true">
-          <span style="color: red;">*</span>
-          <input type="password" required aria-describedby="pwd-error">
-          <span id="pwd-error" role="alert">Password must be 8+ characters</span>
-        </form>
-      context:
-        language: html
-        wcagLevel: A
-    expected_output:
-      must_contain:
-        - "error"
-        - "3.3.1"
-      finding_count:
-        min: 1
-        max: 2
-    validation:
-      schema_check: true
-  - id: tc009_understandable_language
-    description: "Detect missing page language (WCAG 3.1.1)"
-    category: understandable
-    priority: high
-    input:
-      code: |
-        <!DOCTYPE html>
-        <html>
-        <head><title>My Page</title></head>
-        <body><h1>Welcome</h1></body>
-        </html>
-      context:
-        language: html
-        wcagLevel: A
-    expected_output:
-      must_contain:
-        - "lang"
-        - "3.1.1"
-        - "language"
-      finding_count:
-        min: 1
-        max: 1
-      severity_classification: serious
-    validation:
-      schema_check: true
-      keyword_match_threshold: 0.9
-  # -------------------------------------------------------------------------
-  # ROBUST (WCAG 4.x) - Compatible with assistive technologies
-  # -------------------------------------------------------------------------
-  - id: tc010_robust_aria_valid
-    description: "Detect invalid ARIA attributes (WCAG 4.1.2)"
-    category: robust
-    priority: high
-    input:
-      code: |
-        <button aria-label="">Submit</button>
-        <div role="button" aria-pressed="maybe">Toggle</div>
-        <input type="checkbox" aria-checked="true">
-        <nav aria-labelledby="nonexistent">Navigation</nav>
-      context:
-        language: html
-        wcagLevel: A
-    expected_output:
-      must_contain:
-        - "ARIA"
-        - "4.1.2"
-        - "robust"
-      finding_count:
-        min: 2
-        max: 4
-    validation:
-      schema_check: true
-      keyword_match_threshold: 0.8
-  - id: tc011_robust_html_parsing
-    description: "Detect HTML parsing errors (WCAG 4.1.1)"
-    category: robust
-    priority: medium
-    input:
-      code: |
-        <html>
-        <head><title>Page</head>
-        <body>
-          <div id="main">
-            <p>Unclosed paragraph
-            <div id="main">Duplicate ID</div>
-          </div>
-        </body>
-        </html>
-      context:
-        language: html
-        wcagLevel: A
-    expected_output:
-      must_contain:
-        - "4.1.1"
-        - "parsing"
-      finding_count:
-        min: 1
-        max: 4
-    validation:
-      schema_check: true
-  # -------------------------------------------------------------------------
-  # Negative Tests (Should NOT find issues)
-  # -------------------------------------------------------------------------
-  - id: tc012_no_false_positives_accessible
-    description: "Fully accessible page should not flag critical issues"
-    category: negative
-    priority: critical
-    input:
-      code: |
-        <!DOCTYPE html>
-        <html lang="en">
-        <head>
-          <title>Accessible Website</title>
-          <meta charset="UTF-8">
-        </head>
-        <body>
-          <a href="#main" class="skip-link">Skip to main content</a>
-          <header>
-            <nav aria-label="Main navigation">
-              <ul>
-                <li><a href="/">Home</a></li>
-                <li><a href="/about">About</a></li>
-              </ul>
-            </nav>
-          </header>
-          <main id="main">
-            <h1>Welcome to Our Site</h1>
-            <img src="hero.jpg" alt="Happy customers using our product">
-            <form>
-              <label for="email">Email address:</label>
-              <input type="email" id="email" name="email" required aria-describedby="email-hint">
-              <span id="email-hint">We'll never share your email.</span>
-              <button type="submit">Subscribe</button>
-            </form>
-          </main>
-          <footer>
-            <p>&copy; 2026 Accessible Company</p>
-          </footer>
-        </body>
-        </html>
-      context:
-        language: html
-        wcagLevel: AA
-    expected_output:
-      must_contain:
-        - "accessible"
-      must_not_contain:
-        - "critical"
-        - "serious"
-      finding_count:
-        max: 2  # Allow minor/informational findings only
-    validation:
-      schema_check: true
-      keyword_match_threshold: 0.7
-  # -------------------------------------------------------------------------
-  # Edge Cases
-  # -------------------------------------------------------------------------
-  - id: tc013_dynamic_content_aria_live
-    description: "Detect missing aria-live for dynamic content"
-    category: edge_cases
-    priority: medium
-    input:
-      code: |
-        <div id="notifications">
-          <!-- JavaScript updates this -->
-        </div>
-        <div id="status" aria-live="polite" role="status">
-          Loading...
-        </div>
-        <div id="alerts" role="alert">
-          Error occurred!
-        </div>
-      context:
-        language: html
-        wcagLevel: AA
-        description: "Dynamic content regions"
-    expected_output:
-      must_contain:
-        - "aria-live"
-        - "dynamic"
-    validation:
-      schema_check: true
-  - id: tc014_heading_hierarchy
-    description: "Detect broken heading hierarchy (WCAG 1.3.1)"
-    category: edge_cases
-    priority: medium
-    input:
-      code: |
-        <h1>Main Title</h1>
-        <h3>Skipped h2!</h3>
-        <h4>Another section</h4>
-        <h2>Back to h2</h2>
-        <h6>Way out of order</h6>
-      context:
-        language: html
-        wcagLevel: A
-    expected_output:
-      must_contain:
-        - "heading"
-        - "hierarchy"
-        - "1.3.1"
-      finding_count:
-        min: 1
-        max: 3
-    validation:
-      schema_check: true
-  # -------------------------------------------------------------------------
-  # Remediation Quality Tests
-  # -------------------------------------------------------------------------
-  - id: tc015_remediation_code_quality
-    description: "Verify remediation includes actionable code examples"
-    category: remediation
-    priority: high
-    input:
-      code: |
-        <img src="product.jpg">
-        <div onclick="buy()">Buy Now</div>
-      context:
-        language: html
-        wcagLevel: AA
-        options:
-          includeRemediation: true
-    expected_output:
-      must_contain:
-        - "alt="
-        - "button"
-        - "role"
-      must_match_regex:
-        - "<img.*alt=\".*\".*>"
-        - "<button.*>.*</button>"
-    validation:
-      schema_check: true
-      grading_rubric:
-        completeness: 0.3
-        accuracy: 0.4
-        actionability: 0.3
-  # -------------------------------------------------------------------------
-  # Multi-page / Complex Scenarios
-  # -------------------------------------------------------------------------
-  - id: tc016_complex_form_validation
-    description: "Complex form with multiple accessibility requirements"
-    category: integration
-    priority: high
-    input:
-      code: |
-        <form id="checkout">
-          <fieldset>
-            <legend>Shipping Address</legend>
-            <input type="text" name="street" placeholder="Street">
-            <input type="text" name="city" placeholder="City">
-            <select name="state">
-              <option value="">Select State</option>
-            </select>
-          </fieldset>
-          <fieldset>
-            <legend>Payment</legend>
-            <input type="text" name="card" maxlength="16">
-            <input type="text" name="cvv" maxlength="3">
-          </fieldset>
-          <div class="error" style="display:none;"></div>
-          <button type="submit">Place Order</button>
-        </form>
-      context:
-        language: html
-        wcagLevel: AA
-        environment: production
-    expected_output:
-      must_contain:
-        - "label"
-        - "form"
-        - "3.3"
-      finding_count:
-        min: 4
-        max: 10
-    validation:
-      schema_check: true
-      keyword_match_threshold: 0.8
-# =============================================================================
-# Success Criteria
-# =============================================================================
-success_criteria:
-  # Minimum percentage of tests that must pass
-  pass_rate: 0.90
-  # Critical tests (POUR principle detection) must have 100% pass rate
-  critical_pass_rate: 1.0
-  # Average reasoning quality across all tests
-  avg_reasoning_quality: 0.75
-  # Maximum time for entire suite (5 minutes)
-  max_execution_time_ms: 300000
-  # Maximum variance between different models (15%)
-  cross_model_variance: 0.15
-# =============================================================================
-# Metadata
-# =============================================================================
-metadata:
-  author: "@aqe-team"
-  created: "2026-02-02"
-  last_updated: "2026-02-02"
-  coverage_target: "POUR principles, WCAG 2.2 A/AA criteria, remediation quality"
-  wcag_criteria_covered:
-    - "1.1.1 Non-text Content"
-    - "1.2.2 Captions (Prerecorded)"
-    - "1.3.1 Info and Relationships"
-    - "1.4.3 Contrast (Minimum)"
-    - "2.1.1 Keyboard"
-    - "2.1.2 No Keyboard Trap"
-    - "2.4.7 Focus Visible"
-    - "3.1.1 Language of Page"
-    - "3.3.1 Error Identification"
-    - "3.3.2 Labels or Instructions"
-    - "4.1.1 Parsing"
-    - "4.1.2 Name, Role, Value"
+# =============================================================================
+# AQE Accessibility Testing Skill Evaluation Suite v1.0.0
+# WCAG 2.2 compliance testing evaluation with POUR principle coverage
+# =============================================================================
+#
+# This evaluation suite validates accessibility testing skill behavior through:
+# 1. POUR principle coverage (Perceivable, Operable, Understandable, Robust)
+# 2. WCAG 2.2 Level A, AA, AAA conformance testing
+# 3. Multi-model consistency across Claude and GPT models
+# 4. Severity classification validation
+# 5. Remediation quality assessment
+#
+# Schema: .validation/schemas/skill-eval.schema.json
+# Runner: scripts/run-skill-eval.ts
+# =============================================================================
+skill: accessibility-testing
+version: 1.0.0
+description: >
+  Comprehensive evaluation suite for WCAG 2.2 accessibility testing skill.
+  Tests POUR principles (Perceivable, Operable, Understandable, Robust),
+  conformance levels, finding detection, and remediation quality across
+  multiple LLM models.
+# =============================================================================
+# Multi-Model Configuration
+# =============================================================================
+models_to_test:
+  - claude-3.5-sonnet    # Primary model (high accuracy expected)
+  - claude-3-haiku       # Fast model (ensure minimum quality)
+  - gpt-4o               # Cross-vendor validation
+# =============================================================================
+# MCP Integration Configuration
+# =============================================================================
+mcp_integration:
+  enabled: true
+  namespace: skill-validation
+  # Query existing accessibility patterns before running evals
+  query_patterns: true
+  # Track each test outcome for learning feedback
+  track_outcomes: true
+  # Store successful patterns (WCAG violations, remediation approaches)
+  store_patterns: true
+  # Share learning with fleet coordinator agents
+  share_learning: true
+  # Update quality gate with accessibility metrics
+  update_quality_gate: true
+  # Agents to share learning with
+  target_agents:
+    - qe-learning-coordinator
+    - qe-queen-coordinator
+    - qe-accessibility-auditor
+# =============================================================================
+# ReasoningBank Learning Configuration
+# =============================================================================
+learning:
+  store_success_patterns: true
+  store_failure_patterns: true
+  pattern_ttl_days: 90
+  min_confidence_to_store: 0.7
+  cross_model_comparison: true
+# =============================================================================
+# Result Format Configuration
+# =============================================================================
+result_format:
+  json_output: true
+  markdown_report: true
+  include_raw_output: false
+  include_timing: true
+  include_token_usage: true
+# =============================================================================
+# Environment Setup
+# =============================================================================
+setup:
+  required_tools:
+    - jq
+  environment_variables:
+    WCAG_VERSION: "2.2"
+    TARGET_LEVEL: "AA"
+  fixtures:
+    - name: accessible_page
+      content: |
+        <!DOCTYPE html>
+        <html lang="en">
+        <head><title>Accessible Page</title></head>
+        <body>
+          <header><nav aria-label="Main"><a href="/">Home</a></nav></header>
+          <main>
+            <h1>Welcome</h1>
+            <img src="hero.jpg" alt="Person using laptop">
+            <form>
+              <label for="email">Email:</label>
+              <input type="email" id="email" name="email" required>
+              <button type="submit">Subscribe</button>
+            </form>
+          </main>
+        </body>
+        </html>
+    - name: inaccessible_page
+      content: |
+        <!DOCTYPE html>
+        <html>
+        <head><title></title></head>
+        <body>
+          <div onclick="navigate()">Menu</div>
+          <img src="logo.png">
+          <input type="text" placeholder="Enter email">
+          <span style="color: #999; background: #fff;">Light gray text</span>
+        </body>
+        </html>
+# =============================================================================
+# Test Cases - POUR Principles Coverage
+# =============================================================================
+test_cases:
+  # -------------------------------------------------------------------------
+  # PERCEIVABLE (WCAG 1.x) - Content can be perceived
+  # -------------------------------------------------------------------------
+  - id: tc001_perceivable_alt_text
+    description: "Detect missing alt text on images (WCAG 1.1.1)"
+    category: perceivable
+    priority: critical
+    input:
+      code: |
+        <img src="product.jpg">
+        <img src="banner.png" alt="">
+        <img src="hero.jpg" alt="Team collaboration in modern office">
+      context:
+        language: html
+        wcagLevel: AA
+    expected_output:
+      must_contain:
+        - "alt"
+        - "1.1.1"
+        - "perceivable"
+      must_not_contain:
+        - "no issues"
+      finding_count:
+        min: 1
+        max: 2
+      severity_classification: serious
+    validation:
+      schema_check: true
+      keyword_match_threshold: 0.9
+      reasoning_quality_min: 0.7
+  - id: tc002_perceivable_color_contrast
+    description: "Detect insufficient color contrast (WCAG 1.4.3)"
+    category: perceivable
+    priority: critical
+    input:
+      code: |
+        <p style="color: #777; background-color: #fff;">Gray text on white</p>
+        <p style="color: #333; background-color: #fff;">Dark text on white</p>
+        <h1 style="color: #aaa; background-color: #eee;">Low contrast heading</h1>
+      context:
+        language: html
+        wcagLevel: AA
+    expected_output:
+      must_contain:
+        - "contrast"
+        - "1.4.3"
+        - "4.5:1"
+      finding_count:
+        min: 1
+        max: 3
+    validation:
+      schema_check: true
+      keyword_match_threshold: 0.8
+  - id: tc003_perceivable_video_captions
+    description: "Detect videos without captions (WCAG 1.2.2)"
+    category: perceivable
+    priority: critical
+    input:
+      code: |
+        <video src="product-demo.mp4" controls>
+          <source src="product-demo.mp4" type="video/mp4">
+        </video>
+        <video controls>
+          <source src="interview.mp4" type="video/mp4">
+          <track kind="captions" src="captions.vtt" srclang="en">
+        </video>
+      context:
+        language: html
+        wcagLevel: AA
+    expected_output:
+      must_contain:
+        - "caption"
+        - "1.2.2"
+        - "track"
+      finding_count:
+        min: 1
+        max: 2
+    validation:
+      schema_check: true
+      keyword_match_threshold: 0.8
+  # -------------------------------------------------------------------------
+  # OPERABLE (WCAG 2.x) - Interface is operable
+  # -------------------------------------------------------------------------
+  - id: tc004_operable_keyboard_access
+    description: "Detect elements not keyboard accessible (WCAG 2.1.1)"
+    category: operable
+    priority: critical
+    input:
+      code: |
+        <div onclick="handleClick()">Click me</div>
+        <span class="button" onclick="submit()">Submit</span>
+        <button onclick="save()">Save</button>
+        <a href="/next">Next Page</a>
+      context:
+        language: html
+        wcagLevel: A
+    expected_output:
+      must_contain:
+        - "keyboard"
+        - "2.1.1"
+        - "operable"
+        - "button"
+      must_not_contain:
+        - "all elements accessible"
+      finding_count:
+        min: 2
+        max: 3
+      severity_classification: critical
+    validation:
+      schema_check: true
+      keyword_match_threshold: 0.9
+      reasoning_quality_min: 0.8
+  - id: tc005_operable_focus_visible
+    description: "Detect missing focus indicators (WCAG 2.4.7)"
+    category: operable
+    priority: high
+    input:
+      code: |
+        <style>
+          button:focus { outline: none; }
+          a:focus { outline: 0; }
+          input:focus { outline: none; border-color: blue; }
+        </style>
+        <button>Submit</button>
+        <a href="/">Home</a>
+        <input type="text" placeholder="Search">
+      context:
+        language: html
+        wcagLevel: AA
+    expected_output:
+      must_contain:
+        - "focus"
+        - "2.4.7"
+        - "outline"
+      finding_count:
+        min: 2
+        max: 3
+    validation:
+      schema_check: true
+      keyword_match_threshold: 0.8
+  - id: tc006_operable_keyboard_trap
+    description: "Detect keyboard traps (WCAG 2.1.2)"
+    category: operable
+    priority: critical
+    input:
+      code: |
+        <div id="modal" tabindex="0" onkeydown="if(event.key==='Tab'){event.preventDefault();}">
+          <h2>Modal Dialog</h2>
+          <input type="text" placeholder="Enter name">
+          <button>Close</button>
+        </div>
+      context:
+        language: html
+        wcagLevel: A
+        description: "Modal with keyboard trap"
+    expected_output:
+      must_contain:
+        - "keyboard trap"
+        - "2.1.2"
+        - "Tab"
+      severity_classification: critical
+    validation:
+      schema_check: true
+  # -------------------------------------------------------------------------
+  # UNDERSTANDABLE (WCAG 3.x) - Content is understandable
+  # -------------------------------------------------------------------------
+  - id: tc007_understandable_form_labels
+    description: "Detect form inputs without labels (WCAG 3.3.2)"
+    category: understandable
+    priority: high
+    input:
+      code: |
+        <form>
+          <input type="text" placeholder="Name">
+          <input type="email" placeholder="Email">
+          <label for="phone">Phone:</label>
+          <input type="tel" id="phone">
+          <button>Submit</button>
+        </form>
+      context:
+        language: html
+        wcagLevel: A
+    expected_output:
+      must_contain:
+        - "label"
+        - "3.3.2"
+        - "understandable"
+      finding_count:
+        min: 2
+        max: 3
+    validation:
+      schema_check: true
+      keyword_match_threshold: 0.8
+  - id: tc008_understandable_error_messages
+    description: "Detect missing error identification (WCAG 3.3.1)"
+    category: understandable
+    priority: high
+    input:
+      code: |
+        <form>
+          <input type="email" required aria-invalid="true">
+          <span style="color: red;">*</span>
+          <input type="password" required aria-describedby="pwd-error">
+          <span id="pwd-error" role="alert">Password must be 8+ characters</span>
+        </form>
+      context:
+        language: html
+        wcagLevel: A
+    expected_output:
+      must_contain:
+        - "error"
+        - "3.3.1"
+      finding_count:
+        min: 1
+        max: 2
+    validation:
+      schema_check: true
+  - id: tc009_understandable_language
+    description: "Detect missing page language (WCAG 3.1.1)"
+    category: understandable
+    priority: high
+    input:
+      code: |
+        <!DOCTYPE html>
+        <html>
+        <head><title>My Page</title></head>
+        <body><h1>Welcome</h1></body>
+        </html>
+      context:
+        language: html
+        wcagLevel: A
+    expected_output:
+      must_contain:
+        - "lang"
+        - "3.1.1"
+        - "language"
+      finding_count:
+        min: 1
+        max: 1
+      severity_classification: serious
+    validation:
+      schema_check: true
+      keyword_match_threshold: 0.9
+  # -------------------------------------------------------------------------
+  # ROBUST (WCAG 4.x) - Compatible with assistive technologies
+  # -------------------------------------------------------------------------
+  - id: tc010_robust_aria_valid
+    description: "Detect invalid ARIA attributes (WCAG 4.1.2)"
+    category: robust
+    priority: high
+    input:
+      code: |
+        <button aria-label="">Submit</button>
+        <div role="button" aria-pressed="maybe">Toggle</div>
+        <input type="checkbox" aria-checked="true">
+        <nav aria-labelledby="nonexistent">Navigation</nav>
+      context:
+        language: html
+        wcagLevel: A
+    expected_output:
+      must_contain:
+        - "ARIA"
+        - "4.1.2"
+        - "robust"
+      finding_count:
+        min: 2
+        max: 4
+    validation:
+      schema_check: true
+      keyword_match_threshold: 0.8
+  - id: tc011_robust_html_parsing
+    description: "Detect HTML parsing errors (WCAG 4.1.1)"
+    category: robust
+    priority: medium
+    input:
+      code: |
+        <html>
+        <head><title>Page</head>
+        <body>
+          <div id="main">
+            <p>Unclosed paragraph
+            <div id="main">Duplicate ID</div>
+          </div>
+        </body>
+        </html>
+      context:
+        language: html
+        wcagLevel: A
+    expected_output:
+      must_contain:
+        - "4.1.1"
+        - "parsing"
+      finding_count:
+        min: 1
+        max: 4
+    validation:
+      schema_check: true
+  # -------------------------------------------------------------------------
+  # Negative Tests (Should NOT find issues)
+  # -------------------------------------------------------------------------
+  - id: tc012_no_false_positives_accessible
+    description: "Fully accessible page should not flag critical issues"
+    category: negative
+    priority: critical
+    input:
+      code: |
+        <!DOCTYPE html>
+        <html lang="en">
+        <head>
+          <title>Accessible Website</title>
+          <meta charset="UTF-8">
+        </head>
+        <body>
+          <a href="#main" class="skip-link">Skip to main content</a>
+          <header>
+            <nav aria-label="Main navigation">
+              <ul>
+                <li><a href="/">Home</a></li>
+                <li><a href="/about">About</a></li>
+              </ul>
+            </nav>
+          </header>
+          <main id="main">
+            <h1>Welcome to Our Site</h1>
+            <img src="hero.jpg" alt="Happy customers using our product">
+            <form>
+              <label for="email">Email address:</label>
+              <input type="email" id="email" name="email" required aria-describedby="email-hint">
+              <span id="email-hint">We'll never share your email.</span>
+              <button type="submit">Subscribe</button>
+            </form>
+          </main>
+          <footer>
+            <p>&copy; 2026 Accessible Company</p>
+          </footer>
+        </body>
+        </html>
+      context:
+        language: html
+        wcagLevel: AA
+    expected_output:
+      must_contain:
+        - "accessible"
+      must_not_contain:
+        - "critical"
+        - "serious"
+      finding_count:
+        max: 2  # Allow minor/informational findings only
+    validation:
+      schema_check: true
+      keyword_match_threshold: 0.7
+  # -------------------------------------------------------------------------
+  # Edge Cases
+  # -------------------------------------------------------------------------
+  - id: tc013_dynamic_content_aria_live
+    description: "Detect missing aria-live for dynamic content"
+    category: edge_cases
+    priority: medium
+    input:
+      code: |
+        <div id="notifications">
+          <!-- JavaScript updates this -->
+        </div>
+        <div id="status" aria-live="polite" role="status">
+          Loading...
+        </div>
+        <div id="alerts" role="alert">
+          Error occurred!
+        </div>
+      context:
+        language: html
+        wcagLevel: AA
+        description: "Dynamic content regions"
+    expected_output:
+      must_contain:
+        - "aria-live"
+        - "dynamic"
+    validation:
+      schema_check: true
+  - id: tc014_heading_hierarchy
+    description: "Detect broken heading hierarchy (WCAG 1.3.1)"
+    category: edge_cases
+    priority: medium
+    input:
+      code: |
+        <h1>Main Title</h1>
+        <h3>Skipped h2!</h3>
+        <h4>Another section</h4>
+        <h2>Back to h2</h2>
+        <h6>Way out of order</h6>
+      context:
+        language: html
+        wcagLevel: A
+    expected_output:
+      must_contain:
+        - "heading"
+        - "hierarchy"
+        - "1.3.1"
+      finding_count:
+        min: 1
+        max: 3
+    validation:
+      schema_check: true
+  # -------------------------------------------------------------------------
+  # Remediation Quality Tests
+  # -------------------------------------------------------------------------
+  - id: tc015_remediation_code_quality
+    description: "Verify remediation includes actionable code examples"
+    category: remediation
+    priority: high
+    input:
+      code: |
+        <img src="product.jpg">
+        <div onclick="buy()">Buy Now</div>
+      context:
+        language: html
+        wcagLevel: AA
+        options:
+          includeRemediation: true
+    expected_output:
+      must_contain:
+        - "alt="
+        - "button"
+        - "role"
+      must_match_regex:
+        - "<img.*alt=\".*\".*>"
+        - "<button.*>.*</button>"
+    validation:
+      schema_check: true
+      grading_rubric:
+        completeness: 0.3
+        accuracy: 0.4
+        actionability: 0.3
+  # -------------------------------------------------------------------------
+  # Multi-page / Complex Scenarios
+  # -------------------------------------------------------------------------
+  - id: tc016_complex_form_validation
+    description: "Complex form with multiple accessibility requirements"
+    category: integration
+    priority: high
+    input:
+      code: |
+        <form id="checkout">
+          <fieldset>
+            <legend>Shipping Address</legend>
+            <input type="text" name="street" placeholder="Street">
+            <input type="text" name="city" placeholder="City">
+            <select name="state">
+              <option value="">Select State</option>
+            </select>
+          </fieldset>
+          <fieldset>
+            <legend>Payment</legend>
+            <input type="text" name="card" maxlength="16">
+            <input type="text" name="cvv" maxlength="3">
+          </fieldset>
+          <div class="error" style="display:none;"></div>
+          <button type="submit">Place Order</button>
+        </form>
+      context:
+        language: html
+        wcagLevel: AA
+        environment: production
+    expected_output:
+      must_contain:
+        - "label"
+        - "form"
+        - "3.3"
+      finding_count:
+        min: 4
+        max: 10
+    validation:
+      schema_check: true
+      keyword_match_threshold: 0.8
+# =============================================================================
+# Success Criteria
+# =============================================================================
+success_criteria:
+  # Minimum percentage of tests that must pass
+  pass_rate: 0.90
+  # Critical tests (POUR principle detection) must have 100% pass rate
+  critical_pass_rate: 1.0
+  # Average reasoning quality across all tests
+  avg_reasoning_quality: 0.75
+  # Maximum time for entire suite (5 minutes)
+  max_execution_time_ms: 300000
+  # Maximum variance between different models (15%)
+  cross_model_variance: 0.15
+# =============================================================================
+# Metadata
+# =============================================================================
+metadata:
+  author: "@aqe-team"
+  created: "2026-02-02"
+  last_updated: "2026-02-02"
+  coverage_target: "POUR principles, WCAG 2.2 A/AA criteria, remediation quality"
+  wcag_criteria_covered:
+    - "1.1.1 Non-text Content"
+    - "1.2.2 Captions (Prerecorded)"
+    - "1.3.1 Info and Relationships"
+    - "1.4.3 Contrast (Minimum)"
+    - "2.1.1 Keyboard"
+    - "2.1.2 No Keyboard Trap"
+    - "2.4.7 Focus Visible"
+    - "3.1.1 Language of Page"
+    - "3.3.1 Error Identification"
+    - "3.3.2 Labels or Instructions"
+    - "4.1.1 Parsing"
+    - "4.1.2 Name, Role, Value"