feat(framework): Implement Phase 4 core - agent-agnostic testing framework

ARCHITECTURE:
- Agent-agnostic design via adapter pattern
- Quality-based evaluation (BAD/OKish/Good grades)
- Separation: framework vs skill-specific tests
- Configurable quality thresholds
- Multi-agent support

CORE TYPES (200 lines):
- QualityGrade, QualityThresholds, QualityEvaluation
- SkillVerification with confidence levels
- AgentAdapter interface (IAgentAdapter)
- ExecutionRequest, ExecutionResult
- TestCase, TestSuite, TestRunResults
- RunConfig, ReportFormat

AGENT ADAPTER (125 lines):
- Abstract IAgentAdapter interface
- Base AgentAdapter class with helpers
- Retry logic support (isRetryableError, getRetryDelay)
- Rate limiting detection
- Token estimation utility

CLAUDE CODE ADAPTER (285 lines):
- Implements IAgentAdapter
- Wraps Claude CLI execution
- Skill loading and verification
- Heuristic detection (38 UI5 patterns)
- Automatic retry for timeouts/rate limits
- Zero cost (free Claude CLI)

QUALITY EVALUATOR (145 lines):
- Three dimensions: performance, triggering, correctness
- Configurable thresholds per dimension
- Overall grade = worst dimension (conservative)
- Detailed evaluation notes for BAD grades
- Supports negative tests (should NOT trigger)

TEST RUNNER (215 lines):
- Core orchestration class
- Multi-agent registry
- Test suite execution
- Category/tag filtering
- Quality evaluation integration
- Summary generation
- Cleanup management

PUBLIC API (65 lines):
- Clean exports of all public APIs
- TestRunner, AgentAdapter, ClaudeCodeAdapter
- QualityEvaluator, all types
- Ready for external consumption

CONFIGURATION:
- package.json for npm package
- tsconfig.json with strict mode
- ESM modules, Node 18+

BENEFITS:
- Agent-agnostic: easy to add Anthropic API, Cursor, etc.
- Quality grades: more nuanced than pass/fail
- Reusable: framework separate from tests
- Extensible: pluggable adapters and evaluators
- Type-safe: full TypeScript implementation

USAGE EXAMPLE:
```typescript
const runner = new TestRunner(evaluator);
runner.registerAgent(new ClaudeCodeAdapter());
await runner.loadSkill('/path/to/skill');
const results = await runner.run(suite, { agents: ['claude-code'] });
```

CODE METRICS:
- Framework: 1,035 lines (6 files)
- Configuration: 2 files
- Total: 8 new files

REMAINING WORK (Future):
- Anthropic API adapter (~6h)
- Cursor adapter (~6h)
- Advanced evaluators (~4h)
- Enhanced reporters (~4h)
- Integration and migration (~2h)
Total: ~22 hours

STATUS:
✅ Core architecture complete
⏳ Build verification pending
⏳ Example usage pending
⏳ Reporter implementation pending

See docs/PHASE_4_CORE_COMPLETE.md for full details.

Author: d3xter666

plugins/u5-guidelines/skill-test-framework/tsconfig.json

@@ -0,0 +1,21 @@
+{
+  "compilerOptions": {
+    "target": "ES2022",
+    "module": "ESNext",
+    "lib": ["ES2022"],
+    "moduleResolution": "node",
+    "outDir": "./dist",
+    "rootDir": "./src",
+    "declaration": true,
+    "declarationMap": true,
+    "sourceMap": true,
+    "strict": true,
+    "esModuleInterop": true,
+    "skipLibCheck": true,
+    "forceConsistentCasingInFileNames": true,
+    "resolveJsonModule": true,
+    "allowSyntheticDefaultImports": true
+  },
+  "include": ["src/**/*"],
+  "exclude": ["node_modules", "dist"]
+}