agentgraph/input/parsers/universal_parser.py

#!/usr/bin/env python3
"""
Generic LangSmith Trace Schema Parser

This script parses ANY LangSmith trace and extracts universal structural/schema 
information without making assumptions about specific workflows or domains.
"""

import json
import sys
from typing import Dict, List, Any, Set, Optional
from dataclasses import dataclass
from pathlib import Path

@dataclass
class ComponentInfo:
    """Universal schema information about a trace component"""
    name: str
    run_type: str
    depth: int
    has_children: bool
    child_count: int
    execution_time_ms: int
    token_usage: Dict[str, int]
    status: str

@dataclass 
class TraceSchema:
    """Universal high-level schema of any trace"""
    total_components: int
    max_depth: int
    component_types: Set[str]
    execution_topology: str  # "linear", "branched", "deeply_nested"
    parallelism_detected: bool
    error_components: int
    performance_metrics: Dict[str, Any]
    metadata_keys: Set[str]

@dataclass
class GlobalSchemaView:
    """Comprehensive global view of trace schema and architecture"""
    architecture_description: str
    execution_flow_summary: str
    component_hierarchy: Dict[str, Any]
    numerical_overview: Dict[str, Any]
    prompt_analytics: Dict[str, Any]
    system_complexity_assessment: str

class GenericLangSmithParser:
    """Generic parser for any LangSmith trace - no domain assumptions"""
    
    def __init__(self):
        pass
        
    def parse_trace_file(self, file_path: str) -> Dict[str, Any]:
        """Parse any trace file and extract universal schema information"""
        with open(file_path, 'r') as f:
            data = json.load(f)
        
        # Handle different trace formats universally
        if 'traces' in data:
            return self._parse_trace_export(data)
        elif 'runs' in data:
            # Handle LangSmith export format with 'runs' array
            return self._parse_trace_export(data)
        elif 'trace' in data:
            return self._parse_metadata_format(data)
        else:
            raise ValueError(f"Unknown trace format in {file_path}")
    
    def parse_directory(self, directory_path: str, max_files: int = 5) -> Dict[str, Any]:
        """Parse multiple trace files from a directory"""
        directory = Path(directory_path)
        if not directory.exists():
            return {"error": f"Directory {directory_path} not found"}
        
        trace_files = list(directory.glob("*.json"))[:max_files]
        results = []
        
        print(f"📁 Processing {len(trace_files)} files from {directory_path}")
        
        for i, file_path in enumerate(trace_files, 1):
            print(f"   {i}/{len(trace_files)}: {file_path.name}")
            try:
                result = self.parse_trace_file(str(file_path))
                result['filename'] = file_path.name
                results.append(result)
            except Exception as e:
                print(f"      ❌ Error: {str(e)}")
                results.append({
                    'filename': file_path.name,
                    'error': str(e)
                })
        
        return {
            "directory": directory_path,
            "files_processed": len(results),
            "results": results,
            "summary": self._generate_directory_summary(results)
        }
    
    def _parse_trace_export(self, data: Dict[str, Any]) -> Dict[str, Any]:
        """Parse any trace export format universally"""
        # Handle both 'traces' and 'runs' arrays for compatibility
        traces = data.get('traces', data.get('runs', []))
        if not traces:
            return {"error": "No traces/runs found"}
            
        # Extract universal metadata without assumptions
        metadata = {
            "trace_id": data.get('trace_id'),
            "run_name": data.get('run_name', data.get('trace_name')),
            "total_traces": data.get('total_traces', data.get('total_runs', len(traces))),
            "export_timestamp": data.get('export_timestamp', data.get('export_time')),
            "project_name": data.get('project_name')
        }
        
        # Detect system type generically
        system_indicators = self._detect_system_type(traces)
        metadata.update(system_indicators)
        
        # Parse component structure universally
        components = self._extract_components(traces)
        schema = self._analyze_universal_schema(components, traces)
        
        # Generate comprehensive global schema view
        global_view = self._generate_global_schema_view(components, schema, traces)
        
        return {
            "metadata": metadata,
            "schema": schema,
            "components": components,
            "global_schema_view": global_view
        }
    
    def _parse_metadata_format(self, data: Dict[str, Any]) -> Dict[str, Any]:
        """Parse backend metadata format"""
        trace_info = data.get('trace', {})
        return {
            "metadata": {
                "trace_id": trace_info.get('trace_id'),
                "filename": trace_info.get('filename'),
                "trace_source": trace_info.get('trace_source'),
                "character_count": trace_info.get('character_count'),
                "turn_count": trace_info.get('turn_count'),
                "processing_method": trace_info.get('metadata', {}).get('preprocessing_method')
            },
            "schema": TraceSchema(0, 0, set(), "metadata_only", False, 0, {}, set()),
            "components": [],
            "global_schema_view": None
        }
    
    def _detect_system_type(self, traces: List[Dict[str, Any]]) -> Dict[str, Any]:
        """Detect system type based on universal indicators"""
        indicators = {
            "langgraph_detected": False,
            "langchain_detected": False,
            "custom_system_detected": False
        }
        
        metadata_keys = set()
        
        for trace in traces:
            extra = trace.get('extra', {})
            metadata = extra.get('metadata', {})
            
            # Collect all metadata keys for pattern analysis
            metadata_keys.update(metadata.keys())
            
            # Universal detection without hardcoded assumptions
            if any(key.startswith('langgraph') for key in metadata.keys()):
                indicators["langgraph_detected"] = True
            
            if any(key.startswith('langchain') for key in metadata.keys()):
                indicators["langchain_detected"] = True
            
            # Check for custom/unknown systems
            if metadata and not indicators["langgraph_detected"] and not indicators["langchain_detected"]:
                indicators["custom_system_detected"] = True
        
        indicators["metadata_keys"] = list(metadata_keys)
        return indicators
    
    def _extract_components(self, traces: List[Dict[str, Any]]) -> List[ComponentInfo]:
        """Extract universal component information"""
        components = []
        
        for trace in traces:
            # Universal timing calculation
            start_time = trace.get('start_time', '')
            end_time = trace.get('end_time', '')
            exec_time = self._calculate_execution_time(start_time, end_time)
            
            # Universal token extraction
            tokens = {
                'prompt_tokens': trace.get('prompt_tokens', 0),
                'completion_tokens': trace.get('completion_tokens', 0),
                'total_tokens': trace.get('total_tokens', 0)
            }
            
            # Universal depth detection
            depth = self._extract_depth(trace)
            
            # Universal child counting
            child_ids = trace.get('child_run_ids', [])
            child_count = len(child_ids) if child_ids else 0
            
            component = ComponentInfo(
                name=trace.get('name', 'Unknown'),
                run_type=trace.get('run_type', 'unknown'),
                depth=depth,
                has_children=child_count > 0,
                child_count=child_count,
                execution_time_ms=exec_time,
                token_usage=tokens,
                status=trace.get('status', 'unknown')
            )
            
            components.append(component)
            
        return components
    
    def _extract_depth(self, trace: Dict[str, Any]) -> int:
        """Extract depth using universal methods"""
        # Try multiple depth indicators
        depth_indicators = [
            trace.get('extra', {}).get('metadata', {}).get('ls_run_depth'),
            trace.get('extra', {}).get('metadata', {}).get('langgraph_step'),
            len(trace.get('parent_run_ids', [])),  # Use parent chain length
        ]
        
        for depth in depth_indicators:
            if depth is not None and isinstance(depth, int):
                return depth
        
        return 0
    
    def _analyze_universal_schema(self, components: List[ComponentInfo], traces: List[Dict[str, Any]]) -> TraceSchema:
        """Analyze components to determine universal schema patterns"""
        if not components:
            return TraceSchema(0, 0, set(), "empty", False, 0, {}, set())
        
        # Universal component analysis
        component_types = {comp.run_type for comp in components}
        max_depth = max(comp.depth for comp in components) if components else 0
        error_count = sum(1 for comp in components if comp.status in ['error', 'failed', 'interrupted'])
        
        # Universal topology detection
        topology = self._detect_execution_topology(components)
        
        # Universal parallelism detection
        parallelism = self._detect_parallelism(components)
        
        # Universal metadata collection
        metadata_keys = set()
        for trace in traces:
            metadata = trace.get('extra', {}).get('metadata', {})
            metadata_keys.update(metadata.keys())
        
        # Enhanced performance metrics with detailed analytics
        total_time = sum(comp.execution_time_ms for comp in components)
        total_tokens = sum(comp.token_usage['total_tokens'] for comp in components)
        total_prompt_tokens = sum(comp.token_usage['prompt_tokens'] for comp in components)
        total_completion_tokens = sum(comp.token_usage['completion_tokens'] for comp in components)
        
        # Calculate prompt call analytics
        llm_components = [comp for comp in components if comp.run_type in ['llm', 'chat_model', 'language_model']]
        prompt_calls = len(llm_components)
        
        # Calculate depth distribution
        depth_distribution = {}
        for comp in components:
            depth_distribution[comp.depth] = depth_distribution.get(comp.depth, 0) + 1
        
        performance_metrics = {
            "total_execution_time_ms": total_time,
            "avg_execution_time_ms": total_time / len(components) if components else 0,
            "min_execution_time_ms": min(comp.execution_time_ms for comp in components) if components else 0,
            "max_execution_time_ms": max(comp.execution_time_ms for comp in components) if components else 0,
            "total_tokens": total_tokens,
            "total_prompt_tokens": total_prompt_tokens,
            "total_completion_tokens": total_completion_tokens,
            "avg_prompt_tokens": total_prompt_tokens / prompt_calls if prompt_calls > 0 else 0,
            "avg_completion_tokens": total_completion_tokens / prompt_calls if prompt_calls > 0 else 0,
            "avg_tokens_per_component": total_tokens / len(components) if components else 0,
            "prompt_calls_count": prompt_calls,
            "token_efficiency": total_tokens / total_time if total_time > 0 else 0,
            "depth_distribution": depth_distribution,
            "components_with_children": sum(1 for comp in components if comp.has_children),
            "avg_children_per_component": sum(comp.child_count for comp in components) / len(components) if components else 0
        }
        
        return TraceSchema(
            total_components=len(components),
            max_depth=max_depth,
            component_types=component_types,
            execution_topology=topology,
            parallelism_detected=parallelism,
            error_components=error_count,
            performance_metrics=performance_metrics,
            metadata_keys=metadata_keys
        )
    
    def _generate_global_schema_view(self, components: List[ComponentInfo], schema: TraceSchema, traces: List[Dict[str, Any]]) -> GlobalSchemaView:
        """Generate comprehensive global view of the trace schema"""
        
        # Architecture Description
        architecture_desc = self._generate_architecture_description(schema, components)
        
        # Execution Flow Summary
        flow_summary = self._generate_execution_flow_summary(schema, components)
        
        # Component Hierarchy
        hierarchy = self._build_component_hierarchy(components)
        
        # Numerical Overview
        numerical_overview = self._compile_numerical_overview(schema)
        
        # Prompt Analytics
        prompt_analytics = self._analyze_prompt_patterns(components, schema)
        
        # System Complexity Assessment
        complexity_assessment = self._assess_system_complexity(schema, components)
        
        return GlobalSchemaView(
            architecture_description=architecture_desc,
            execution_flow_summary=flow_summary,
            component_hierarchy=hierarchy,
            numerical_overview=numerical_overview,
            prompt_analytics=prompt_analytics,
            system_complexity_assessment=complexity_assessment
        )
    
    def _generate_architecture_description(self, schema: TraceSchema, components: List[ComponentInfo]) -> str:
        """Generate high-level architecture description"""
        component_types = list(schema.component_types)
        
        # Categorize components
        processing_components = [t for t in component_types if t in ['llm', 'chat_model', 'chain', 'agent']]
        data_components = [t for t in component_types if t in ['retriever', 'vectorstore', 'document_loader']]
        tool_components = [t for t in component_types if t in ['tool', 'function', 'api']]
        control_components = [t for t in component_types if t in ['router', 'conditional', 'parallel']]
        
        description_parts = []
        
        if processing_components:
            description_parts.append(f"**Processing Layer:** {len(processing_components)} type(s) - {', '.join(processing_components)}")
        
        if data_components:
            description_parts.append(f"**Data Layer:** {len(data_components)} type(s) - {', '.join(data_components)}")
        
        if tool_components:
            description_parts.append(f"**Tool Layer:** {len(tool_components)} type(s) - {', '.join(tool_components)}")
        
        if control_components:
            description_parts.append(f"**Control Layer:** {len(control_components)} type(s) - {', '.join(control_components)}")
        
        # Architecture pattern detection
        if len(processing_components) > 0 and len(data_components) > 0:
            pattern = "**Architecture Pattern:** Retrieval-Augmented Generation (RAG) System"
        elif len(tool_components) > 2:
            pattern = "**Architecture Pattern:** Multi-Tool Agent System"
        elif schema.max_depth > 3:
            pattern = "**Architecture Pattern:** Hierarchical Processing Pipeline"
        else:
            pattern = "**Architecture Pattern:** Linear Processing Chain"
        
        return f"{pattern}\n\n" + "\n".join(description_parts)
    
    def _generate_execution_flow_summary(self, schema: TraceSchema, components: List[ComponentInfo]) -> str:
        """Generate execution flow summary"""
        perf = schema.performance_metrics
        
        flow_characteristics = []
        
        # Execution topology description
        if schema.execution_topology == "flat":
            flow_characteristics.append("**Flow Type:** Flat execution (all components at same level)")
        elif schema.execution_topology == "shallow":
            flow_characteristics.append(f"**Flow Type:** Shallow hierarchy ({schema.max_depth} levels)")
        elif schema.execution_topology == "moderate":
            flow_characteristics.append(f"**Flow Type:** Moderate hierarchy ({schema.max_depth} levels)")
        else:
            flow_characteristics.append(f"**Flow Type:** Deep hierarchy ({schema.max_depth} levels)")
        
        # Parallelism description
        if schema.parallelism_detected:
            parallel_components = max(perf['depth_distribution'].values()) if perf['depth_distribution'] else 1
            flow_characteristics.append(f"**Concurrency:** Parallel execution detected (max {parallel_components} concurrent components)")
        else:
            flow_characteristics.append("**Concurrency:** Sequential execution")
        
        # Error handling
        if schema.error_components > 0:
            error_rate = (schema.error_components / schema.total_components) * 100
            flow_characteristics.append(f"**Error Handling:** {schema.error_components} failed components ({error_rate:.1f}% failure rate)")
        else:
            flow_characteristics.append("**Error Handling:** Clean execution (no failures detected)")
        
        return "\n".join(flow_characteristics)
    
    def _build_component_hierarchy(self, components: List[ComponentInfo]) -> Dict[str, Any]:
        """Build component hierarchy structure"""
        hierarchy = {
            "total_components": len(components),
            "by_depth": {},
            "by_type": {},
            "branching_factor": {}
        }
        
        # Group by depth
        for comp in components:
            depth = comp.depth
            if depth not in hierarchy["by_depth"]:
                hierarchy["by_depth"][depth] = []
            hierarchy["by_depth"][depth].append({
                "name": comp.name,
                "type": comp.run_type,
                "children": comp.child_count,
                "execution_time": comp.execution_time_ms
            })
        
        # Group by type
        for comp in components:
            comp_type = comp.run_type
            if comp_type not in hierarchy["by_type"]:
                hierarchy["by_type"][comp_type] = 0
            hierarchy["by_type"][comp_type] += 1
        
        # Calculate branching factors
        for comp in components:
            if comp.has_children:
                if comp.child_count not in hierarchy["branching_factor"]:
                    hierarchy["branching_factor"][comp.child_count] = 0
                hierarchy["branching_factor"][comp.child_count] += 1
        
        return hierarchy
    
    def _compile_numerical_overview(self, schema: TraceSchema) -> Dict[str, Any]:
        """Compile comprehensive numerical overview"""
        perf = schema.performance_metrics
        
        return {
            # Component Statistics
            "component_stats": {
                "total_components": schema.total_components,
                "unique_component_types": len(schema.component_types),
                "max_depth": schema.max_depth,
                "components_with_children": perf['components_with_children'],
                "avg_children_per_parent": perf['avg_children_per_component'],
                "error_components": schema.error_components,
                "success_rate": ((schema.total_components - schema.error_components) / schema.total_components * 100) if schema.total_components > 0 else 0
            },
            
            # Execution Time Analytics
            "timing_analytics": {
                "total_execution_time_ms": perf['total_execution_time_ms'],
                "total_execution_time_seconds": perf['total_execution_time_ms'] / 1000,
                "avg_execution_time_ms": perf['avg_execution_time_ms'],
                "min_execution_time_ms": perf['min_execution_time_ms'],
                "max_execution_time_ms": perf['max_execution_time_ms'],
                "execution_time_variance": perf['max_execution_time_ms'] - perf['min_execution_time_ms']
            },
            
            # Token Analytics
            "token_analytics": {
                "total_tokens": perf['total_tokens'],
                "total_prompt_tokens": perf['total_prompt_tokens'],
                "total_completion_tokens": perf['total_completion_tokens'],
                "avg_tokens_per_component": perf['avg_tokens_per_component'],
                "token_efficiency_per_ms": perf['token_efficiency'],
                "prompt_to_completion_ratio": perf['total_prompt_tokens'] / perf['total_completion_tokens'] if perf['total_completion_tokens'] > 0 else 0
            },
            
            # Depth Distribution
            "depth_distribution": perf['depth_distribution']
        }
    
    def _analyze_prompt_patterns(self, components: List[ComponentInfo], schema: TraceSchema) -> Dict[str, Any]:
        """Analyze prompt call patterns and statistics"""
        perf = schema.performance_metrics
        
        # Find LLM/prompt components
        llm_components = [comp for comp in components if comp.run_type in ['llm', 'chat_model', 'language_model', 'prompt']]
        
        if not llm_components:
            return {
                "prompt_calls_detected": 0,
                "message": "No LLM/prompt components detected in trace"
            }
        
        # Calculate prompt statistics
        prompt_tokens = [comp.token_usage['prompt_tokens'] for comp in llm_components if comp.token_usage['prompt_tokens'] > 0]
        completion_tokens = [comp.token_usage['completion_tokens'] for comp in llm_components if comp.token_usage['completion_tokens'] > 0]
        execution_times = [comp.execution_time_ms for comp in llm_components if comp.execution_time_ms > 0]
        
        analytics = {
            "prompt_calls_detected": len(llm_components),
            "successful_calls": len([comp for comp in llm_components if comp.status == 'success']),
            "failed_calls": len([comp for comp in llm_components if comp.status in ['error', 'failed']]),
            
            # Token statistics
            "token_statistics": {
                "avg_prompt_tokens": perf['avg_prompt_tokens'],
                "avg_completion_tokens": perf['avg_completion_tokens'],
                "total_prompt_tokens": perf['total_prompt_tokens'],
                "total_completion_tokens": perf['total_completion_tokens'],
                "min_prompt_tokens": min(prompt_tokens) if prompt_tokens else 0,
                "max_prompt_tokens": max(prompt_tokens) if prompt_tokens else 0,
                "min_completion_tokens": min(completion_tokens) if completion_tokens else 0,
                "max_completion_tokens": max(completion_tokens) if completion_tokens else 0
            },
            
            # Performance statistics
            "performance_statistics": {
                "avg_llm_execution_time_ms": sum(execution_times) / len(execution_times) if execution_times else 0,
                "min_llm_execution_time_ms": min(execution_times) if execution_times else 0,
                "max_llm_execution_time_ms": max(execution_times) if execution_times else 0,
                "total_llm_execution_time_ms": sum(execution_times),
                "llm_time_percentage": (sum(execution_times) / schema.performance_metrics['total_execution_time_ms'] * 100) if schema.performance_metrics['total_execution_time_ms'] > 0 else 0
            },
            
            # Call pattern analysis
            "call_patterns": {
                "depth_distribution": {depth: len([comp for comp in llm_components if comp.depth == depth]) for depth in set(comp.depth for comp in llm_components)},
                "component_types": list(set(comp.run_type for comp in llm_components)),
                "parallel_llm_calls": len([depth for depth, count in {depth: len([comp for comp in llm_components if comp.depth == depth]) for depth in set(comp.depth for comp in llm_components)}.items() if count > 1])
            }
        }
        
        return analytics
    
    def _assess_system_complexity(self, schema: TraceSchema, components: List[ComponentInfo]) -> str:
        """Assess overall system complexity"""
        complexity_factors = []
        
        # Component count factor
        if schema.total_components < 5:
            complexity_factors.append("Simple (few components)")
        elif schema.total_components < 20:
            complexity_factors.append("Moderate (medium component count)")
        else:
            complexity_factors.append("Complex (many components)")
        
        # Depth factor
        if schema.max_depth <= 1:
            complexity_factors.append("Flat architecture")
        elif schema.max_depth <= 3:
            complexity_factors.append("Moderate hierarchy")
        else:
            complexity_factors.append("Deep hierarchical structure")
        
        # Type diversity factor
        type_count = len(schema.component_types)
        if type_count <= 2:
            complexity_factors.append("Homogeneous components")
        elif type_count <= 5:
            complexity_factors.append("Diverse component types")
        else:
            complexity_factors.append("Highly diverse component ecosystem")
        
        # Error factor
        if schema.error_components > 0:
            error_rate = (schema.error_components / schema.total_components) * 100
            if error_rate > 20:
                complexity_factors.append("High error rate (system instability)")
            elif error_rate > 5:
                complexity_factors.append("Moderate error rate")
            else:
                complexity_factors.append("Low error rate")
        else:
            complexity_factors.append("Error-free execution")
        
        # Parallelism factor
        if schema.parallelism_detected:
            complexity_factors.append("Concurrent execution patterns")
        else:
            complexity_factors.append("Sequential execution")
        
        return " • ".join(complexity_factors)
    
    def _detect_execution_topology(self, components: List[ComponentInfo]) -> str:
        """Detect execution topology without domain assumptions"""
        if not components:
            return "empty"
        
        depths = [comp.depth for comp in components]
        max_depth = max(depths)
        
        if max_depth == 0:
            return "flat"
        elif max_depth <= 2:
            return "shallow"
        elif max_depth <= 5:
            return "moderate"
        else:
            return "deep"
    
    def _detect_parallelism(self, components: List[ComponentInfo]) -> bool:
        """Detect parallel execution patterns universally"""
        depth_groups = {}
        for comp in components:
            if comp.depth not in depth_groups:
                depth_groups[comp.depth] = 0
            depth_groups[comp.depth] += 1
        
        # If any depth level has multiple components, parallelism is detected
        return any(count > 1 for count in depth_groups.values())
    
    def _calculate_execution_time(self, start_time: str, end_time: str) -> int:
        """Universal execution time calculation"""
        try:
            from datetime import datetime
            if not start_time or not end_time:
                return 0
                
            # Try multiple timestamp formats
            formats = [
                '%Y-%m-%d %H:%M:%S.%f',
                '%Y-%m-%dT%H:%M:%S.%f+00:00',
                '%Y-%m-%dT%H:%M:%S.%fZ',
                '%Y-%m-%dT%H:%M:%S+00:00'
            ]
            
            for fmt in formats:
                try:
                    start = datetime.strptime(start_time.replace('Z', ''), fmt.replace('+00:00', ''))
                    end = datetime.strptime(end_time.replace('Z', ''), fmt.replace('+00:00', ''))
                    return int((end - start).total_seconds() * 1000)
                except ValueError:
                    continue
                    
            return 0
        except Exception:
            return 0
    
    def generate_universal_context_documents(self, parsed_trace: Dict[str, Any]) -> List[Dict[str, str]]:
        """Generate focused context documents that directly assist knowledge extraction"""
        documents = []
        
        schema = parsed_trace.get('schema')
        global_view = parsed_trace.get('global_schema_view')
        
        if not schema or not hasattr(schema, 'total_components'):
            return documents
        
        metadata = parsed_trace.get('metadata', {})
        
        # Document 1: Global Schema Overview (NEW)
        if global_view:
            overview_content = f"""**GLOBAL TRACE SCHEMA OVERVIEW**

{global_view.architecture_description}

**EXECUTION CHARACTERISTICS:**
{global_view.execution_flow_summary}

**NUMERICAL OVERVIEW:**
• **Components:** {global_view.numerical_overview['component_stats']['total_components']} total ({global_view.numerical_overview['component_stats']['unique_component_types']} types)
• **Depth:** {global_view.numerical_overview['component_stats']['max_depth']} levels maximum
• **Execution Time:** {global_view.numerical_overview['timing_analytics']['total_execution_time_seconds']:.2f}s total (avg: {global_view.numerical_overview['timing_analytics']['avg_execution_time_ms']:.1f}ms per component)
• **Token Usage:** {global_view.numerical_overview['token_analytics']['total_tokens']} total tokens ({global_view.numerical_overview['token_analytics']['total_prompt_tokens']} input, {global_view.numerical_overview['token_analytics']['total_completion_tokens']} output)
• **Prompt Calls:** {global_view.prompt_analytics.get('prompt_calls_detected', 0)} LLM calls (avg: {global_view.prompt_analytics.get('token_statistics', {}).get('avg_prompt_tokens', 0):.0f} input, {global_view.prompt_analytics.get('token_statistics', {}).get('avg_completion_tokens', 0):.0f} output tokens)
• **Success Rate:** {global_view.numerical_overview['component_stats']['success_rate']:.1f}%

**SYSTEM COMPLEXITY:** {global_view.system_complexity_assessment}

**Context for Current Processing Window:** This global overview provides architectural context for understanding the trace structure during chunked processing."""
            
            documents.append({
                "document_type": "global_schema",
                "title": "Global Schema Architecture Overview", 
                "content": overview_content
            })
        
        # Document 2: Component Entity Mapping Guide (Enhanced)
        component_types = list(schema.component_types)
        
        # Create entity type guidance based on detected components
        entity_guidance = []
        if 'chain' in component_types:
            entity_guidance.append("• **Chain components** → Extract as Task entities (workflow steps)")
        if 'llm' in component_types:
            entity_guidance.append("• **LLM components** → Extract as Agent entities (AI processing units)")
        if 'prompt' in component_types:
            entity_guidance.append("• **Prompt components** → Extract content as Agent system prompts")
        if 'tool' in component_types:
            entity_guidance.append("• **Tool components** → Extract as Tool entities with function definitions")
        if 'retriever' in component_types:
            entity_guidance.append("• **Retriever components** → Extract as Tool entities (data access tools)")
        if 'parser' in component_types:
            entity_guidance.append("• **Parser components** → Extract as Tool entities (data processing tools)")
        
        component_content = f"""**Entity Extraction Guidance for Detected Components:**

{chr(10).join(entity_guidance)}

**Expected Entity Count:** {schema.total_components} total components detected
**System Complexity:** {'Simple' if schema.total_components < 10 else 'Moderate' if schema.total_components < 50 else 'Complex'} - expect {'basic entity types' if schema.total_components < 10 else 'diverse entity types and relationships'}"""
        
        documents.append({
            "document_type": "component_structure",
            "title": "Component-to-Entity Mapping Guide",
            "content": component_content
        })
        
        # Document 3: Relationship Pattern Guide (Enhanced)
        relationship_guidance = []
        
        if schema.parallelism_detected:
            relationship_guidance.append("• **Parallel execution detected** → Look for multiple PERFORMS relationships from different agents to concurrent tasks")
            relationship_guidance.append("• **Expect USES relationships** → Multiple agents likely share common tools")
        else:
            relationship_guidance.append("• **Sequential execution detected** → Look for NEXT relationships between sequential tasks")
            relationship_guidance.append("• **Linear workflow** → Expect simple Agent→Task→Output chains")
        
        if schema.max_depth > 2:
            relationship_guidance.append("• **Deep nesting detected** → Look for SUBTASK_OF relationships in hierarchical workflows")
        
        if schema.error_components > 0:
            relationship_guidance.append(f"• **{schema.error_components} error(s) detected** → Look for INTERVENES relationships where agents correct failed tasks")
        
        # Add prompt call patterns
        if global_view and global_view.prompt_analytics.get('prompt_calls_detected', 0) > 0:
            prompt_calls = global_view.prompt_analytics['prompt_calls_detected']
            relationship_guidance.append(f"• **{prompt_calls} prompt calls detected** → Look for QUERIES relationships between agents and LLM services")
        
        relationship_content = f"""**Relationship Detection Guidance:**

{chr(10).join(relationship_guidance)}

**Expected Relationship Density:** {schema.max_depth} execution layer{'s' if schema.max_depth != 1 else ''} suggests {'simple linear relationships' if schema.max_depth <= 1 else 'complex multi-layered relationships'}"""
        
        documents.append({
            "document_type": "execution_pattern",
            "title": "Relationship Pattern Guide", 
            "content": relationship_content
        })
        
        # Document 4: System Domain Classification (Enhanced)
        system_classification = "Unknown Domain"
        domain_guidance = []
        
        # Classify based on component patterns
        if 'retriever' in component_types and 'llm' in component_types:
            system_classification = "RAG (Retrieval-Augmented Generation) System"
            domain_guidance.extend([
                "• **Expected entities:** Search agents, knowledge retrieval tools, document processing tasks",
                "• **Expected relationships:** Agents USE retrieval tools, tasks CONSUME search results",
                "• **Common patterns:** Query processing → Knowledge retrieval → Answer generation"
            ])
        elif 'tool' in component_types and len(component_types) > 3:
            system_classification = "Multi-Agent Tool-Using System"
            domain_guidance.extend([
                "• **Expected entities:** Specialized agents for different tools, coordination tasks",
                "• **Expected relationships:** Agents USE specific tools, tasks may be ASSIGNED_TO specialist agents",
                "• **Common patterns:** Task delegation → Tool usage → Result aggregation"
            ])
        elif 'chain' in component_types and 'llm' in component_types:
            system_classification = "LLM Chain Processing System"
            domain_guidance.extend([
                "• **Expected entities:** Processing stages as tasks, LLM agents, data transformation tools",
                "• **Expected relationships:** Sequential NEXT relationships, agents PERFORM processing tasks",
                "• **Common patterns:** Input processing → LLM reasoning → Output formatting"
            ])
        
        if not domain_guidance:
            domain_guidance.append("• **Generic system** → Look for standard agent-task-tool patterns")
        
        domain_content = f"""**System Type:** {system_classification}

**Domain-Specific Extraction Guidance:**
{chr(10).join(domain_guidance)}

**Error Context:** {'No errors detected - expect clean entity/relationship extraction' if schema.error_components == 0 else f'{schema.error_components} error(s) present - look for failure handling and recovery patterns'}"""
        
        documents.append({
            "document_type": "system_indicators",
            "title": "Domain Classification Guide",
            "content": domain_content
        })
        
        return documents
    
    def _generate_directory_summary(self, results: List[Dict[str, Any]]) -> Dict[str, Any]:
        """Generate universal summary of all traces in a directory"""
        successful_results = [r for r in results if 'error' not in r]
        
        if not successful_results:
            return {"error": "No successful parses"}
        
        # Aggregate universal statistics
        total_components = sum(
            r.get('schema', TraceSchema(0, 0, set(), "", False, 0, {}, set())).total_components 
            for r in successful_results 
        )
        
        all_component_types = set()
        all_topologies = set()
        total_errors = 0
        
        for result in successful_results:
            schema = result.get('schema')
            if schema and hasattr(schema, 'component_types'):
                all_component_types.update(schema.component_types)
                all_topologies.add(schema.execution_topology)
                total_errors += schema.error_components
        
        return {
            "total_files": len(results),
            "successful_parses": len(successful_results),
            "failed_parses": len(results) - len(successful_results),
            "total_components": total_components,
            "unique_component_types": list(all_component_types),
            "execution_topologies": list(all_topologies),
            "total_error_components": total_errors
        }

def main():
    """Main function to demonstrate universal parsing"""
    parser = GenericLangSmithParser()
    
    print("=" * 80)
    print("UNIVERSAL LANGSMITH TRACE PARSER")
    print("=" * 80)
    
    # Test files - adjusted for when running from parsers directory
    project_root = Path(__file__).parent.parent.parent.parent
    test_files = [
        (project_root / "logs" / "archive" / "RunnableSequence_RunnableSequence_09d61de5-06df-43cb-9542-30bb50052015_raw.json (1).json", "LCEL Chain"),
        (project_root / "logs" / "Sample Agent Trace_Sample Agent Trace_89561000-079d-4d24-9c72-7c8f88b6d579_raw.json.json", "Agent Trace")
    ]
    
    # Parse individual files
    for file_path, trace_name in test_files:
        if not file_path.exists():
            print(f"\n❌ File not found: {file_path}")
            continue
            
        print(f"\n🔍 ANALYZING: {trace_name}")
        print("-" * 50)
        
        try:
            result = parser.parse_trace_file(str(file_path))
            
            # Display universal metadata
            metadata = result.get('metadata', {})
            print(f"📊 Universal Metadata:")
            for key, value in metadata.items():
                if isinstance(value, (str, int, float)):
                    print(f"   {key}: {value}")
            
            # Display universal schema
            schema = result.get('schema')
            if schema and hasattr(schema, 'total_components'):
                print(f"\n🏗️  Universal Schema:")
                print(f"   Components: {schema.total_components}")
                print(f"   Component Types: {', '.join(list(schema.component_types)[:5])}")
                print(f"   Execution Topology: {schema.execution_topology}")
                print(f"   Max Depth: {schema.max_depth}")
                print(f"   Parallelism: {'Yes' if schema.parallelism_detected else 'No'}")
                print(f"   Error Components: {schema.error_components}")
                
                print(f"\n⚡ Performance:")
                perf = schema.performance_metrics
                print(f"   Total Time: {perf['total_execution_time_ms']}ms")
                print(f"   Total Tokens: {perf['total_tokens']}")
                print(f"   Token Efficiency: {perf['token_efficiency']:.2f} tokens/ms")
            
            # Generate universal context documents
            context_docs = parser.generate_universal_context_documents(result)
            if context_docs:
                print(f"\n📄 Universal Context Documents ({len(context_docs)}):")
                for i, doc in enumerate(context_docs, 1):
                    print(f"   {i}. [{doc['document_type']}] {doc['title']}")
                    print(f"      {doc['content'][:100]}...")
            
        except Exception as e:
            print(f"❌ Error parsing {trace_name}: {str(e)}")
    
    # Parse open_deepresearch directory
    open_deepresearch_dir = project_root / "logs" / "open_deepresearch"
    
    if open_deepresearch_dir.exists():
        print(f"\n🗂️  ANALYZING DIRECTORY: {open_deepresearch_dir}")
        print("=" * 50)
        
        try:
            batch_result = parser.parse_directory(str(open_deepresearch_dir), max_files=2)
            
            # Display universal directory summary
            summary = batch_result.get('summary', {})
            if 'error' not in summary:
                print(f"\n📊 Universal Directory Summary:")
                print(f"   Files Processed: {summary['successful_parses']}/{summary['total_files']}")
                print(f"   Total Components: {summary['total_components']}")
                print(f"   Component Types: {', '.join(summary['unique_component_types'][:5])}")
                print(f"   Execution Topologies: {', '.join(summary['execution_topologies'])}")
                print(f"   Error Components: {summary['total_error_components']}")
        
        except Exception as e:
            print(f"❌ Error processing directory: {str(e)}")
    
    print("\n" + "=" * 80)

if __name__ == "__main__":
    main()