Add streaming channel support to ggraph

Makefile

@@ -1,4 +1,4 @@
-.PHONY: help backend frontend dev install-backend install-frontend install clean delete-db test
+.PHONY: help backend frontend dev install-backend install-frontend install clean delete-db test test-evals
 
 # Default target - show help
 help:
@@ -18,7 +18,8 @@ help:
 	@echo "Utilities:"
 	@echo "  make clean            - Remove build artifacts and data"
 	@echo "  make delete-db        - Delete SQLite database file"
-	@echo "  make test             - Run tests"
+	@echo "  make test             - Run all backend tests"
+	@echo "  make test-evals       - Run evaluation tests (requires server running)"
 	@echo "  make help             - Show this help message"
 
 # Start backend server
@@ -85,3 +86,11 @@ test:
 	cd backend && go test ./...
 	@echo "Tests complete!"
 
+# Run evaluation tests (requires server to be running)
+test-evals:
+	@echo "Running evaluation tests..."
+	@echo "Note: Make sure the backend server is running on http://localhost:8080"
+	@echo ""
+	cd backend && go test -v ./evals -run TestSystemMessages
+	@echo "Evaluation tests complete!"
+

backend/.gitignore

@@ -7,6 +7,10 @@
 *.db-wal
 data/
 
+# Log files
+logs/
+*.log
+
 # Go
 *.exe
 *.exe~

backend/SUMMARY.md

@@ -0,0 +1,178 @@
+# Master Node Refactoring Summary
+
+## Overview
+
+This document summarizes the refactoring of the master node architecture to improve code clarity, eliminate duplication, enhance extensibility, and fix streaming issues.
+
+## Date
+December 2024
+
+## Changes Made
+
+### 1. Architecture Refactoring
+
+#### New Files Created
+
+**`agentFactory.go`**
+- Centralized agent creation logic
+- `AgentFactory` struct with `CreateThinkingAgent()` and `CreateResponseAgent()` methods
+- Moved `ThinkingAgentSystemPrompt` constant
+- Eliminates duplicate agent creation code
+
+**`responseValidator.go`**
+- Pluggable response validation interface
+- `ResponseValidator` interface with `Validate()` and `ToPrompt()` methods
+- `MasterNodeResponseValidator` implementation with robust field validation
+- Validates required fields based on success status
+- `ParseMasterNodeResponse()` as convenience wrapper
+
+**`retryExecutor.go`**
+- Configurable retry logic with exponential backoff
+- `RetryConfig` struct with `MaxAttempts`, `BaseDelay`, `MaxDelay`, `BackoffFactor`
+- `RetryExecutor` with `ExecuteWithRetry()` and `ExecuteWithRetryResult[T]()` methods
+- Reusable across all nodes
+
+**`agentExecutor.go`**
+- Unified agent execution strategy
+- `AgentExecutionStrategy` consolidates thinking and response agent patterns
+- Handles streaming, completion, and validation
+- Supports async execution and retry logic
+- Methods: `ExecuteAsync()`, `ExecuteWithRetry()`, `ExecuteThinkingPlan()`
+
+**`masterProcessor.go`**
+- Main processing orchestration
+- `MasterNodeProcessor` orchestrates the complete workflow
+- Methods: `Process()`, `generateThinkingPlan()`, `executeMainRequest()`, `handleValidatedResponse()`
+- Clear separation of concerns
+
+#### Files Modified
+
+**`masterNode.go`**
+- Simplified main `MasterNode` function (reduced from ~200 lines to ~30 lines)
+- Removed inline agent creation and execution logic
+- Removed `ThinkingAgentSystemPrompt` and `MasterAgentSystemPrompt` constants (moved to `agentFactory.go`)
+- Updated `StreamResponse()` and `StreamThinking()` to remove tag wrapping (fixes fragmentation issue)
+- Now uses `MasterNodeProcessor` for orchestration
+
+**`responseModels.go`**
+- Removed `ParseMasterNodeResponse()` function (moved to `responseValidator.go`)
+- Removed unused imports
+
+**`utils.go`**
+- Updated `attemptRequest()` to use `RetryExecutor` internally
+- Marked as deprecated in favor of `RetryExecutor`
+- Removed unused `log` import
+
+### 2. Streaming Fix
+
+**Problem**: Each streaming chunk was wrapped in `<response>` or `<thinking>` tags, causing fragmentation:
+```
+<response>{</response><response> </response><response>"</response>...
+```
+
+**Solution**: Removed tag wrapping from `StreamResponse()` and `StreamThinking()` methods. The `status` field already indicates content type (`"thinking"` or `"streaming"`), so tags are unnecessary.
+
+**Result**: Clean, unfragmented streaming content.
+
+### 3. Test Script
+
+**`test_stream.go`**
+- Test script for streaming agent chat endpoint
+- Sends POST request to `/api/agent/chat` with "test" message
+- Displays:
+  - Thinking trace in real-time
+  - Tool execution notifications
+  - Streaming response content
+  - Final JSON response object
+- Handles SSE parsing and JSON formatting
+
+## Benefits
+
+### Extensibility
+- Interface-based design allows pluggable validators
+- Configurable retry strategies
+- Factory pattern for agent creation
+- Easy to add new response types or execution strategies
+
+### Clarity
+- Single-responsibility components
+- Clear separation of concerns
+- Well-documented code
+- Reduced complexity in main function
+
+### Robustness
+- Improved error handling
+- Context cancellation support
+- Field validation for responses
+- Configurable retry with exponential backoff
+
+### Readability
+- Eliminated duplicate code patterns
+- Better code organization
+- Consistent patterns across components
+- In-code documentation
+
+## Backward Compatibility
+
+All changes maintain backward compatibility:
+- `MasterNode` function signature unchanged
+- `MasterNodeStreamHandler` API unchanged
+- `ExtractMasterNodeState` function unchanged
+- All existing state keys and routing logic preserved
+
+## Testing
+
+The refactoring was verified with:
+- `test_stream.go` script showing clean output
+- Server health checks
+- Multiple test runs confirming consistent behavior
+- Verification that tags are no longer fragmenting responses
+
+## Files Summary
+
+### Created
+- `backend/internal/workflows/nodes/agentFactory.go`
+- `backend/internal/workflows/nodes/responseValidator.go`
+- `backend/internal/workflows/nodes/retryExecutor.go`
+- `backend/internal/workflows/nodes/agentExecutor.go`
+- `backend/internal/workflows/nodes/masterProcessor.go`
+- `backend/test_stream.go`
+- `backend/SUMMARY.md` (this file)
+
+### Modified
+- `backend/internal/workflows/nodes/masterNode.go`
+- `backend/internal/workflows/nodes/responseModels.go`
+- `backend/internal/workflows/nodes/utils.go`
+
+## Key Improvements
+
+1. **Eliminated Duplication**: Agent creation and execution patterns consolidated
+2. **Fixed Streaming**: Removed tag wrapping that caused fragmentation
+3. **Improved Testability**: Each component can be tested independently
+4. **Better Error Handling**: Configurable retry with proper error propagation
+5. **Enhanced Maintainability**: Clear separation of concerns and single responsibility
+
+## Usage
+
+### Running Tests
+```bash
+cd backend
+go run test_stream.go
+```
+
+### Server Restart
+```bash
+# Stop server
+ps aux | grep "go run cmd/server/main.go" | grep -v grep | awk '{print $2}' | xargs kill
+
+# Start server
+cd backend && nohup go run cmd/server/main.go > /tmp/server.log 2>&1 &
+```
+
+## Notes
+
+- All code follows established patterns (interface-first design, dependency injection, in-code documentation)
+- The refactoring maintains the existing API contracts
+- Streaming now works correctly without tag fragmentation
+- Test script provides easy way to verify streaming functionality
+