@iqai/adk 0.1.21 → 0.2.0

package/CHANGELOG.md

@@ -1,5 +1,57 @@
 # @iqai/adk
 
+## 0.2.0
+
+### Minor Changes
+
+- 17341fc: Refactor agent loading and resolution logic with enhanced flexibility and reliability
+
+  This major enhancement improves the ADK CLI server's agent loading capabilities and adds new features to the core framework:
+
+  **CLI Server Improvements:**
+
+  - **Modular Architecture**: Refactored monolithic server file into organized modules (`server/index.ts`, `server/routes.ts`, `server/services.ts`, `server/types.ts`)
+  - **Enhanced Agent Resolution**: New `resolveAgentExport` method supports multiple export patterns:
+    - Direct agent exports: `export const agent = new LlmAgent(...)`
+    - Function factories: `export function agent() { return new LlmAgent(...) }`
+    - Async factories: `export async function agent() { return new LlmAgent(...) }`
+    - Container objects: `export default { agent: ... }`
+    - Primitive exports with fallback scanning
+  - **Improved TypeScript Import Handling**: Better project root detection and module resolution for TypeScript files
+
+  **Core Framework Enhancements:**
+
+  - **New AgentBuilder Method**: Added `withAgent()` method to directly provide existing agent instances with definition locking to prevent accidental configuration overwrites
+  - **Two-Tier Tool Deduplication**: Implemented robust deduplication logic to prevent duplicate function declarations that cause errors with LLM providers (especially Google)
+  - **Better Type Safety**: Improved type definitions and replaced `any[]` usage with proper typed interfaces
+
+  **Testing & Reliability:**
+
+  - **Comprehensive Test Coverage**: New `agent-resolution.test.ts` with extensive fixtures testing various agent export patterns
+  - **Multiple Test Fixtures**: Added 6 different agent export pattern examples for validation
+  - **Edge Case Handling**: Improved error handling and logging throughout the agent loading pipeline
+
+  These changes provide a more flexible, reliable, and maintainable foundation for agent development and deployment while maintaining backward compatibility.
+
+- 1564b7b: Port Python evaluation framework to TypeScript
+
+  This change introduces a comprehensive evaluation framework for testing AI agent performance. Key features include:
+
+  - **Core evaluation engine** with agent-evaluator and local evaluation service
+  - **Built-in evaluators** for response matching, trajectory analysis, LLM-as-judge, and safety checks
+  - **Metrics system** with ROUGE scoring and tool trajectory analysis
+  - **Vertex AI integration** for cloud-based evaluation
+  - **Pluggable registry system** for custom metric evaluators
+  - **Structured evaluation cases and test sets** for organized testing
+
+  The framework is marked as experimental and provides essential tooling for evaluating agent responses, tool usage, and overall performance across different scenarios.
+
+## 0.1.22
+
+### Patch Changes
+
+- c4e642a: downgraded info level logs to debug, removed legacy starter in create-adk-project and new adk cli initial version!
+
 ## 0.1.21
 
 ### Patch Changes