# modeReact.py # React mode implementation for workflows import json import logging import re import time from datetime import datetime, timezone from typing import List, Dict, Any from modules.datamodels.datamodelChat import ( TaskStep, TaskContext, TaskResult, ActionItem, TaskStatus, ActionResult ) from modules.datamodels.datamodelChat import ChatWorkflow from modules.datamodels.datamodelAi import AiCallOptions, OperationType, ProcessingMode, Priority from modules.workflows.processing.modes.modeBase import BaseMode from modules.workflows.processing.shared.executionState import TaskExecutionState, shouldContinue from modules.workflows.processing.shared.promptGenerationActionsReact import ( generateReactPlanSelectionPrompt, generateReactParametersPrompt, generateReactRefinementPrompt ) from modules.shared.debugLogger import writeDebugFile from modules.workflows.processing.shared.placeholderFactory import extractReviewContent from modules.workflows.processing.adaptive import IntentAnalyzer, ContentValidator, LearningEngine, ProgressTracker from modules.workflows.processing.adaptive.adaptiveLearningEngine import AdaptiveLearningEngine logger = logging.getLogger(__name__) class ReactMode(BaseMode): """React mode implementation - iterative plan-act-observe-refine loop""" def __init__(self, services, workflow): super().__init__(services, workflow) # Initialize adaptive components self.intentAnalyzer = IntentAnalyzer(services) self.learningEngine = LearningEngine() self.adaptiveLearningEngine = AdaptiveLearningEngine() # New enhanced learning engine self.contentValidator = ContentValidator(services, self.adaptiveLearningEngine) self.progressTracker = ProgressTracker() self.currentIntent = None # Placeholder service no longer used; prompts are generated directly async def generateActionItems(self, taskStep: TaskStep, workflow: ChatWorkflow, previousResults: List = None, enhancedContext: TaskContext = None) -> List[ActionItem]: """React mode doesn't use batch action generation - actions are generated iteratively""" # React mode generates actions one at a time in the execution loop return [] async def executeTask(self, taskStep: TaskStep, workflow: ChatWorkflow, context: TaskContext, taskIndex: int = None, totalTasks: int = None) -> TaskResult: """Execute task using React mode - iterative plan-act-observe-refine loop""" logger.info(f"=== STARTING TASK {taskIndex or '?'}: {taskStep.objective} ===") # NEW: Analyze intents separately for proper validation vs task completion # Workflow-level intent from cleaned original user prompt original_prompt = self.services.currentUserPrompt if self.services and hasattr(self.services, 'currentUserPrompt') else taskStep.objective self.workflowIntent = await self.intentAnalyzer.analyzeUserIntent(original_prompt, context) # Task-level intent from current task objective (used only for task-scoped checks) self.taskIntent = await self.intentAnalyzer.analyzeUserIntent(taskStep.objective, context) logger.info(f"Intent analysis — workflow: {self.workflowIntent}") logger.info(f"Intent analysis — task: {self.taskIntent}") # NEW: Reset progress tracking for new task self.progressTracker.reset() # Update workflow object before executing task if taskIndex is not None: self._updateWorkflowBeforeExecutingTask(taskIndex) # Update workflow context for this task if taskIndex is not None: self.services.workflow.setWorkflowContext(task_number=taskIndex) # Create task start message await self.messageCreator.createTaskStartMessage(taskStep, workflow, taskIndex, totalTasks) state = TaskExecutionState(taskStep) # React mode uses max_steps instead of max_retries state.max_steps = max(1, int(getattr(workflow, 'maxSteps', 5))) logger.info(f"Using React mode execution with max_steps: {state.max_steps}") step = 1 lastReviewDict = None while step <= state.max_steps: self._checkWorkflowStopped(workflow) # Update workflow[currentAction] for UI self._updateWorkflowBeforeExecutingAction(step) self.services.workflow.setWorkflowContext(action_number=step) try: t0 = time.time() selection = await self._planSelect(context) logger.info(f"React step {step}: Selected action: {selection}") # Create user-friendly message BEFORE action execution # Action intention message is now handled by the standard message creator in _actExecute result = await self._actExecute(context, selection, taskStep, workflow, step) observation = self._observeBuild(result) # Attach deterministic label for clarity observation['resultLabel'] = result.resultLabel # NEW: Add content validation (against original cleaned user prompt / workflow intent) if getattr(self, 'workflowIntent', None) and result.documents: validationResult = await self.contentValidator.validateContent(result.documents, self.workflowIntent) observation['contentValidation'] = validationResult quality_score = validationResult.get('qualityScore', 0.0) if quality_score is None: quality_score = 0.0 logger.info(f"Content validation: {validationResult['overallSuccess']} (quality: {quality_score:.2f})") # NEW: Record validation result for adaptive learning actionContext = { 'actionType': selection.get('action', {}).get('action', 'unknown'), 'actionName': selection.get('action', {}).get('action', 'unknown'), 'workflowId': context.workflow_id } self.adaptiveLearningEngine.recordValidationResult( validationResult, actionContext, context.workflow_id, step ) # NEW: Learn from feedback feedback = self._collectFeedback(result, validationResult, self.workflowIntent) self.learningEngine.learnFromFeedback(feedback, context, self.workflowIntent) # NEW: Update progress self.progressTracker.updateProgress(result, validationResult, self.workflowIntent) decision = await self._refineDecide(context, observation) # Store refinement decision in context for next iteration if not hasattr(context, 'previous_review_result') or context.previous_review_result is None: context.previous_review_result = [] if decision: # Only append if decision is not None context.previous_review_result.append(decision) # Update context with learnings from this step if decision and decision.get('reason'): if not hasattr(context, 'improvements'): context.improvements = [] context.improvements.append(f"Step {step}: {decision.get('reason')}") # Telemetry: simple duration per step duration = time.time() - t0 self.services.workflow.storeLog(workflow, { "message": f"react_step_duration_sec={duration:.3f}", "type": "info" }) lastReviewDict = decision # Create user-friendly message AFTER action execution # Action completion message is now handled by the standard message creator in _actExecute except Exception as e: logger.error(f"React step {step} error: {e}") break # NEW: Use adaptive stopping logic progressState = self.progressTracker.getCurrentProgress() continueByProgress = self.progressTracker.shouldContinue(progressState, observation.get('contentValidation', {})) continueByReview = shouldContinue(observation, lastReviewDict, step, state.max_steps) if not continueByProgress or not continueByReview: logger.info(f"Stopping at step {step}: progress={continueByProgress}, review={continueByReview}") break step += 1 # Summarize task result for react mode status = TaskStatus.COMPLETED success = True feedback = lastReviewDict.get('reason') if lastReviewDict and isinstance(lastReviewDict, dict) else 'Completed' if lastReviewDict and isinstance(lastReviewDict, dict) and lastReviewDict.get('decision') == 'stop': success = True # Create task completion message await self.messageCreator.createTaskCompletionMessage(taskStep, workflow, taskIndex, totalTasks, type('ReviewResult', (), {'reason': feedback, 'met_criteria': [], 'quality_score': 8})()) return TaskResult( taskId=taskStep.id, status=status, success=success, feedback=feedback, error=None if success else feedback ) async def _planSelect(self, context: TaskContext) -> Dict[str, Any]: """Plan: select exactly one action. Returns {"action": {method, name}}""" bundle = generateReactPlanSelectionPrompt(self.services, context, self.adaptiveLearningEngine) promptTemplate = bundle.prompt placeholders = bundle.placeholders # Write action selection prompt to debug from modules.shared.debugLogger import writeDebugFile writeDebugFile(promptTemplate, "action_selection_prompt", placeholders) # Centralized AI call for plan selection (use plan generation quality) options = AiCallOptions( operationType=OperationType.GENERATE_PLAN, priority=Priority.QUALITY, compressPrompt=False, compressContext=False, processingMode=ProcessingMode.DETAILED, maxCost=0.10, maxProcessingTime=30 ) response = await self.services.ai.callAi( prompt=promptTemplate, placeholders=placeholders, options=options ) # Write action selection response to debug writeDebugFile(response or '', "action_selection_response") jsonStart = response.find('{') if response else -1 jsonEnd = response.rfind('}') + 1 if response else 0 if jsonStart == -1 or jsonEnd == 0: raise ValueError("No JSON in selection response") selection = json.loads(response[jsonStart:jsonEnd]) if 'action' not in selection or not isinstance(selection['action'], str): raise ValueError("Selection missing 'action' as string") # Validate document references - prevent AI from inventing Message IDs if 'requiredInputDocuments' in selection: self._validateDocumentReferences(selection['requiredInputDocuments'], context) # Enforce spec: Stage 1 must NOT include 'parameters' if 'parameters' in selection: # Remove to avoid accidental carryover try: del selection['parameters'] except Exception: selection['parameters'] = None return selection def _validateDocumentReferences(self, document_refs: List[str], context: TaskContext) -> None: """Validate that document references exist in the current workflow""" if not document_refs: return # Get available documents from the current workflow try: available_docs = self.services.workflow.getAvailableDocuments(self.services.currentWorkflow) if not available_docs or available_docs == "No documents available": logger.warning("No documents available for validation") return # Extract all valid references from available documents valid_refs = [] for line in available_docs.split('\n'): if 'docList:' in line or 'docItem:' in line: # Extract reference from line like " - docList:msg_xxx:label" or " - docItem:xxx:filename with spaces" ref_match = re.search(r'(docList:[^\s]+|docItem:[^\s]+(?:\s+[^\s]+)*)', line) if ref_match: valid_refs.append(ref_match.group(1)) # Prefer non-empty documents: the available_docs index is already filtered to skip empty docs preferred_refs = set(valid_refs) # Check if all provided references are valid and prefer non-empty for ref in document_refs: if ref not in preferred_refs: logger.error(f"Invalid or empty document reference: {ref}") logger.error(f"Available references: {valid_refs}") raise ValueError(f"Document reference '{ref}' not found or refers to empty document. Use only non-empty references from AVAILABLE_DOCUMENTS_INDEX.") except Exception as e: logger.error(f"Error validating document references: {str(e)}") raise ValueError(f"Failed to validate document references: {str(e)}") async def _actExecute(self, context: TaskContext, selection: Dict[str, Any], taskStep: TaskStep, workflow: ChatWorkflow, stepIndex: int) -> ActionResult: """Act: request minimal parameters then execute selected action""" compoundActionName = selection.get('action', '') # Parse compound action name (e.g., "ai.webResearch" -> method="ai", action="webResearch") if '.' not in compoundActionName: raise ValueError(f"Invalid compound action name: {compoundActionName}. Expected format: method.action") methodName, actionName = compoundActionName.split('.', 1) # Always request parameters in Stage 2 (spec: Stage 1 must not provide them) logger.info("Requesting parameters in Stage 2 based on Stage 1 outputs") # Create a permissive Stage 2 context to avoid TaskContext attribute restrictions from types import SimpleNamespace stage2Context = SimpleNamespace() # Copy essential fields from original context for fallbacks (snake_case for placeholderFactory compatibility) stage2Context.task_step = getattr(context, 'task_step', None) stage2Context.workflow_id = getattr(context, 'workflow_id', None) # Set Stage 1 data directly on the permissive context (snake_case for promptGenerationActionsReact compatibility) if isinstance(selection, dict): stage2Context.action_objective = selection.get('actionObjective', '') stage2Context.parameters_context = selection.get('parametersContext', '') stage2Context.learnings = selection.get('learnings', []) else: stage2Context.action_objective = '' stage2Context.parameters_context = '' stage2Context.learnings = [] # Build and send the Stage 2 parameters prompt (always) bundle = generateReactParametersPrompt(self.services, stage2Context, compoundActionName, self.adaptiveLearningEngine) promptTemplate = bundle.prompt placeholders = bundle.placeholders # Write parameters prompt to debug writeDebugFile(promptTemplate, "parameters_prompt", placeholders) # Centralized AI call for parameter suggestion (balanced analysis) options = AiCallOptions( operationType=OperationType.ANALYSE_CONTENT, priority=Priority.BALANCED, compressPrompt=True, compressContext=False, processingMode=ProcessingMode.ADVANCED, maxCost=0.05, maxProcessingTime=30, temperature=0.3, # Slightly higher temperature for better instruction following # maxTokens not set - use model's maximum for big JSON responses resultFormat="json" # Explicitly request JSON format ) paramsResp = await self.services.ai.callAi( prompt=promptTemplate, placeholders=placeholders, options=options ) # Parse JSON response js = paramsResp[paramsResp.find('{'):paramsResp.rfind('}')+1] if paramsResp else '{}' try: paramObj = json.loads(js) parameters = paramObj.get('parameters', {}) if isinstance(paramObj, dict) else {} except Exception as e: logger.error(f"Failed to parse AI parameters response as JSON: {str(e)}") logger.error(f"Response was: {paramsResp}") raise ValueError("AI parameters response invalid JSON") if not isinstance(parameters, dict): raise ValueError("AI parameters response missing 'parameters' object") # Merge Stage 1 resource selections into Stage 2 parameters (only if action expects them) try: requiredDocs = selection.get('requiredInputDocuments') if requiredDocs: # Ensure list if isinstance(requiredDocs, list): # Only attach if target action defines 'documentList' methodName, actionName = compoundActionName.split('.', 1) from modules.workflows.processing.shared.methodDiscovery import getActionParameterList, methods as _methods expectedParams = getActionParameterList(methodName, actionName, _methods) if 'documentList' in expectedParams: parameters['documentList'] = requiredDocs requiredConn = selection.get('requiredConnection') if requiredConn: # Only attach if target action defines 'connectionReference' methodName, actionName = compoundActionName.split('.', 1) from modules.workflows.processing.shared.methodDiscovery import getActionParameterList, methods as _methods expectedParams = getActionParameterList(methodName, actionName, _methods) if 'connectionReference' in expectedParams: parameters['connectionReference'] = requiredConn except Exception: pass # Apply minimal defaults in-code (language) if 'language' not in parameters and hasattr(self.services, 'user') and getattr(self.services.user, 'language', None): parameters['language'] = self.services.user.language # Write parameters response to debug mergedParamObj = { "schema": (paramObj.get('schema') if isinstance(paramObj, dict) else 'parameters_v1'), "parameters": parameters } writeDebugFile(str(mergedParamObj), "parameters_response", mergedParamObj) # Build a synthetic ActionItem for execution routing and labels currentRound = getattr(self.workflow, 'currentRound', 0) currentTask = getattr(self.workflow, 'currentTask', 0) resultLabel = f"round{currentRound}_task{currentTask}_action{stepIndex}_results" taskAction = self._createActionItem({ "execMethod": methodName, "execAction": actionName, "execParameters": parameters, "execResultLabel": resultLabel, "status": TaskStatus.PENDING }) # Execute using existing single action flow (message creation is handled internally) result = await self.actionExecutor.executeSingleAction(taskAction, workflow, taskStep, currentTask, stepIndex, 1) return result def _observeBuild(self, actionResult: ActionResult) -> Dict[str, Any]: """Observe: build compact observation object from ActionResult with full document metadata""" previews = [] notes = [] if actionResult and actionResult.documents: # Process all documents and show full metadata for doc in actionResult.documents: # Extract all available metadata without content docMetadata = { "name": getattr(doc, 'fileName', None) or getattr(doc, 'documentName', 'Unknown'), "mimeType": getattr(doc, 'mimeType', 'Unknown'), "size": getattr(doc, 'size', 'Unknown'), "created": getattr(doc, 'created', 'Unknown'), "modified": getattr(doc, 'modified', 'Unknown'), "typeGroup": getattr(doc, 'typeGroup', 'Unknown'), "documentId": getattr(doc, 'documentId', 'Unknown'), "reference": getattr(doc, 'reference', 'Unknown') } # Remove 'Unknown' values to keep it clean docMetadata = {k: v for k, v in docMetadata.items() if v != 'Unknown'} # Add content size indicator instead of actual content if hasattr(doc, 'documentData') and doc.documentData: if isinstance(doc.documentData, dict) and 'content' in doc.documentData: contentLength = len(str(doc.documentData['content'])) docMetadata['contentSize'] = f"{contentLength} characters" else: contentLength = len(str(doc.documentData)) docMetadata['contentSize'] = f"{contentLength} characters" # Extract comment if available if hasattr(doc, 'documentData') and doc.documentData: data = getattr(doc, 'documentData', None) if isinstance(data, dict): comment = data.get("comment", "") if comment: notes.append(f"Document '{docMetadata.get('name', 'Unknown')}': {comment}") previews.append(docMetadata) observation = { "success": bool(actionResult.success), "resultLabel": actionResult.resultLabel or "", "documentsCount": len(actionResult.documents) if actionResult.documents else 0, "previews": previews, "notes": notes } # NEW: Add content analysis if intent is available if self.currentIntent and actionResult.documents: contentAnalysis = self._analyzeContent(actionResult.documents) observation['contentAnalysis'] = contentAnalysis return observation def _analyzeContent(self, documents: List[Any]) -> Dict[str, Any]: """Analyzes content of documents for adaptive learning""" try: if not documents: return {"contentType": "none", "contentSnippet": "", "intentMatch": False} # Extract content from first document firstDoc = documents[0] content = "" if hasattr(firstDoc, 'documentData'): data = firstDoc.documentData if isinstance(data, dict) and 'content' in data: content = str(data['content']) else: content = str(data) # Classify content type contentType = self._classifyContent(content) # Create content snippet contentSnippet = content[:200] + "..." if len(content) > 200 else content # Assess intent match intentMatch = self._assessIntentMatch(content, self.currentIntent) return { "contentType": contentType, "contentSnippet": contentSnippet, "intentMatch": intentMatch } except Exception as e: logger.error(f"Error analyzing content: {str(e)}") return {"contentType": "error", "contentSnippet": "", "intentMatch": False} def _classifyContent(self, content: str) -> str: """Classifies the type of content""" if not content: return "empty" # Check for code codeIndicators = ['def ', 'function', 'import ', 'class ', 'for ', 'while ', 'if '] if any(indicator in content.lower() for indicator in codeIndicators): return "code" # Check for numbers if re.search(r'\b\d+\b', content): return "numbers" # Check for structured content if any(indicator in content for indicator in ['\n', '\t', '|', '-', '*', '1.', '2.']): return "structured" # Default to text return "text" def _assessIntentMatch(self, content: str, intent: Dict[str, Any]) -> bool: """Assesses if content matches the user intent""" if not intent: return False dataType = intent.get("dataType", "unknown") if dataType == "numbers": # Check if content contains actual numbers, not code hasNumbers = bool(re.search(r'\b\d+\b', content)) isNotCode = not any(keyword in content.lower() for keyword in ['def ', 'function', 'import ']) return hasNumbers and isNotCode elif dataType == "text": # Check if content is readable text words = re.findall(r'\b\w+\b', content) return len(words) > 5 elif dataType == "documents": # Check if content is suitable for document creation hasStructure = any(indicator in content for indicator in ['\n', '\t', '|', '-', '*']) hasContent = len(content.strip()) > 50 return hasStructure and hasContent return True # Default to match for unknown types def _collectFeedback(self, result: Any, validation: Dict[str, Any], intent: Dict[str, Any]) -> Dict[str, Any]: """Collects comprehensive feedback from action execution""" try: # Extract content summary contentDelivered = "" if result.documents: firstDoc = result.documents[0] if hasattr(firstDoc, 'documentData'): data = firstDoc.documentData if isinstance(data, dict) and 'content' in data: content = str(data['content']) contentDelivered = content[:100] + "..." if len(content) > 100 else content else: contentDelivered = str(data)[:100] + "..." if len(str(data)) > 100 else str(data) return { "actionAttempted": result.resultLabel or "unknown", "parametersUsed": {}, # Would be extracted from action context "contentDelivered": contentDelivered, "intentMatchScore": validation.get('qualityScore', 0), "qualityScore": validation.get('qualityScore', 0), "issuesFound": validation.get('improvementSuggestions', []), "learningOpportunities": validation.get('improvementSuggestions', []), "userSatisfaction": None, # Would be collected from user feedback "timestamp": datetime.now(timezone.utc).timestamp() } except Exception as e: logger.error(f"Error collecting feedback: {str(e)}") return { "actionAttempted": "unknown", "parametersUsed": {}, "contentDelivered": "", "intentMatchScore": 0, "qualityScore": 0, "issuesFound": [], "learningOpportunities": [], "userSatisfaction": None, "timestamp": datetime.now(timezone.utc).timestamp() } async def _refineDecide(self, context: TaskContext, observation: Dict[str, Any]) -> Dict[str, Any]: """Refine: decide continue or stop, with reason""" # Create proper ReviewContext for extractReviewContent from modules.datamodels.datamodelChat import ReviewContext reviewContext = ReviewContext( task_step=context.task_step, task_actions=[], action_results=[], # React mode doesn't have action results in this context step_result={'observation': observation}, workflow_id=context.workflow_id, previous_results=[] ) baseReviewContent = extractReviewContent(reviewContext) placeholders = {"REVIEW_CONTENT": baseReviewContent} # NEW: Add content validation to review content enhancedReviewContent = placeholders.get("REVIEW_CONTENT", "") if 'contentValidation' in observation: validation = observation['contentValidation'] enhancedReviewContent += f"\n\nCONTENT VALIDATION:\n" enhancedReviewContent += f"Overall Success: {validation['overallSuccess']}\n" quality_score = validation.get('qualityScore', 0.0) if quality_score is None: quality_score = 0.0 enhancedReviewContent += f"Quality Score: {quality_score:.2f}\n" if validation['improvementSuggestions']: enhancedReviewContent += f"Improvement Suggestions: {', '.join(validation['improvementSuggestions'])}\n" # NEW: Add content analysis to review content if 'contentAnalysis' in observation: analysis = observation['contentAnalysis'] enhancedReviewContent += f"\nCONTENT ANALYSIS:\n" enhancedReviewContent += f"Content Type: {analysis['contentType']}\n" enhancedReviewContent += f"Intent Match: {analysis['intentMatch']}\n" if analysis['contentSnippet']: enhancedReviewContent += f"Content Preview: {analysis['contentSnippet']}\n" # NEW: Add progress state to review content progressState = self.progressTracker.getCurrentProgress() enhancedReviewContent += f"\nPROGRESS STATE:\n" enhancedReviewContent += f"Completed Objectives: {len(progressState['completedObjectives'])}\n" enhancedReviewContent += f"Partial Achievements: {len(progressState['partialAchievements'])}\n" enhancedReviewContent += f"Failed Attempts: {len(progressState['failedAttempts'])}\n" enhancedReviewContent += f"Current Phase: {progressState['currentPhase']}\n" if progressState['nextActionsSuggested']: enhancedReviewContent += f"Next Action Suggestions: {', '.join(progressState['nextActionsSuggested'])}\n" # Update placeholders with enhanced review content placeholders["REVIEW_CONTENT"] = enhancedReviewContent bundle = generateReactRefinementPrompt(self.services, context, enhancedReviewContent) promptTemplate = bundle.prompt placeholders = bundle.placeholders # Write refinement/validation prompt to debug writeDebugFile(promptTemplate, "validation_refinement_prompt", placeholders) # Centralized AI call for refinement decision (balanced analysis) options = AiCallOptions( operationType=OperationType.ANALYSE_CONTENT, priority=Priority.BALANCED, compressPrompt=True, compressContext=False, processingMode=ProcessingMode.ADVANCED, maxCost=0.05, maxProcessingTime=30 ) resp = await self.services.ai.callAi( prompt=promptTemplate, placeholders=placeholders, options=options ) # Write refinement/validation response to debug writeDebugFile(resp or '', "validation_refinement_response") js = resp[resp.find('{'):resp.rfind('}')+1] if resp else '{}' try: decision = json.loads(js) except Exception: decision = {"decision": "continue", "reason": "default"} return decision async def _createReactActionMessage(self, workflow: ChatWorkflow, selection: Dict[str, Any], step: int, maxSteps: int, taskIndex: int, messageType: str, result: ActionResult = None, observation: Dict[str, Any] = None): """Create user-friendly messages for React workflow actions""" try: action = selection.get('action', {}) method = action.get('method', '') actionName = action.get('name', '') # Get user language userLanguage = self.services.user.language if self.services and self.services.user else 'en' if messageType == "before": # Message BEFORE action execution userMessage = await self._generateActionIntentionMessage(method, actionName, userLanguage) messageContent = f"🔄 **Step {step}/{maxSteps}**\n\n{userMessage}" status = "step" actionProgress = "pending" documentsLabel = f"action_{step}_intention" elif messageType == "after": # Message AFTER action execution userMessage = await self._generateActionResultMessage(method, actionName, result, observation, userLanguage) successIcon = "✅" if result and result.success else "❌" messageContent = f"{successIcon} **Step {step}/{maxSteps} Complete**\n\n{userMessage}" status = "step" actionProgress = "success" if result and result.success else "fail" documentsLabel = observation.get('resultLabel') if observation else f"action_{step}_result" else: return # Create workflow message messageData = { "workflowId": workflow.id, "role": "assistant", "message": messageContent, "status": status, "sequenceNr": len(workflow.messages) + 1, "publishedAt": self.services.utils.getUtcTimestamp(), "documentsLabel": documentsLabel, "documents": [], "roundNumber": workflow.currentRound, "taskNumber": taskIndex, "actionNumber": step, "actionProgress": actionProgress } self.services.workflow.storeMessageWithDocuments(workflow, messageData, []) except Exception as e: logger.error(f"Error creating React action message: {str(e)}") async def _generateActionIntentionMessage(self, method: str, actionName: str, userLanguage: str): """Generate user-friendly message explaining what action will do""" try: # Create a simple AI prompt to generate user-friendly action descriptions prompt = f"""Generate a brief, user-friendly message explaining what the {method}.{actionName} action will do. User language: {userLanguage} Return only the user-friendly message, no technical details.""" # Call AI to generate user-friendly message response = await self.services.ai.callAi( prompt=prompt, options=AiCallOptions( operationType=OperationType.GENERATE_CONTENT, priority=Priority.SPEED, compressPrompt=True, maxCost=0.01, maxProcessingTime=5 ) ) return response.strip() if response else f"Executing {method}.{actionName} action..." except Exception as e: logger.error(f"Error generating action intention message: {str(e)}") return f"Executing {method}.{actionName} action..." async def _generateActionResultMessage(self, method: str, actionName: str, result: ActionResult, observation: Dict[str, Any], userLanguage: str): """Generate user-friendly message explaining action results""" try: # Build result context resultContext = "" if result and result.documents: docCount = len(result.documents) resultContext = f"Generated {docCount} document(s)" elif observation and observation.get('documentsCount', 0) > 0: docCount = observation.get('documentsCount', 0) resultContext = f"Generated {docCount} document(s)" # Create AI prompt for result message prompt = f"""Generate a brief, user-friendly message explaining the result of the {method}.{actionName} action. User language: {userLanguage} Success: {result.success if result else 'Unknown'} Result context: {resultContext} Return only the user-friendly message, no technical details.""" # Call AI to generate user-friendly result message response = await self.services.ai.callAi( prompt=prompt, options=AiCallOptions( operationType=OperationType.GENERATE_CONTENT, priority=Priority.SPEED, compressPrompt=True, maxCost=0.01, maxProcessingTime=5 ) ) return response.strip() if response else f"{method}.{actionName} action completed" except Exception as e: logger.error(f"Error generating action result message: {str(e)}") return f"{method}.{actionName} action completed" def _createActionItem(self, actionData: Dict[str, Any]) -> ActionItem: """Creates a new task action for React mode""" try: import uuid # Ensure ID is present if "id" not in actionData or not actionData["id"]: actionData["id"] = f"action_{uuid.uuid4()}" # Ensure required fields if "status" not in actionData: actionData["status"] = TaskStatus.PENDING if "execMethod" not in actionData: logger.error("execMethod is required for task action") return None if "execAction" not in actionData: logger.error("execAction is required for task action") return None if "execParameters" not in actionData: actionData["execParameters"] = {} # Use generic field separation based on ActionItem model simpleFields, objectFields = self.services.interfaceDbChat._separate_object_fields(ActionItem, actionData) # Create action in database createdAction = self.services.interfaceDbChat.db.recordCreate(ActionItem, simpleFields) # Convert to ActionItem model return ActionItem( id=createdAction["id"], execMethod=createdAction["execMethod"], execAction=createdAction["execAction"], execParameters=createdAction.get("execParameters", {}), execResultLabel=createdAction.get("execResultLabel"), expectedDocumentFormats=createdAction.get("expectedDocumentFormats"), status=createdAction.get("status", TaskStatus.PENDING), error=createdAction.get("error"), retryCount=createdAction.get("retryCount", 0), retryMax=createdAction.get("retryMax", 3), processingTime=createdAction.get("processingTime"), timestamp=float(createdAction.get("timestamp", self.services.utils.getUtcTimestamp())), result=createdAction.get("result"), resultDocuments=createdAction.get("resultDocuments", []), userMessage=createdAction.get("userMessage") ) except Exception as e: logger.error(f"Error creating task action: {str(e)}") return None def _updateWorkflowBeforeExecutingTask(self, taskNumber: int): """Update workflow object before executing a task""" try: updateData = { "currentTask": taskNumber, "currentAction": 0, "totalActions": 0 } # Update workflow object self.workflow.currentTask = taskNumber self.workflow.currentAction = 0 self.workflow.totalActions = 0 # Update in database self.services.interfaceDbChat.updateWorkflow(self.workflow.id, updateData) logger.info(f"Updated workflow {self.workflow.id} before executing task {taskNumber}: {updateData}") except Exception as e: logger.error(f"Error updating workflow before executing task: {str(e)}") def _updateWorkflowBeforeExecutingAction(self, actionNumber: int): """Update workflow object before executing an action""" try: updateData = { "currentAction": actionNumber } # Update workflow object self.workflow.currentAction = actionNumber # Update in database self.services.interfaceDbChat.updateWorkflow(self.workflow.id, updateData) logger.info(f"Updated workflow {self.workflow.id} before executing action {actionNumber}: {updateData}") except Exception as e: logger.error(f"Error updating workflow before executing action: {str(e)}") def _createActionItem(self, actionData: Dict[str, Any]) -> ActionItem: """Creates a new task action for React mode""" try: import uuid # Ensure ID is present if "id" not in actionData or not actionData["id"]: actionData["id"] = f"action_{uuid.uuid4()}" # Ensure required fields if "status" not in actionData: actionData["status"] = TaskStatus.PENDING if "execMethod" not in actionData: logger.error("execMethod is required for task action") return None if "execAction" not in actionData: logger.error("execAction is required for task action") return None if "execParameters" not in actionData: actionData["execParameters"] = {} # Use generic field separation based on ActionItem model simpleFields, objectFields = self.services.interfaceDbChat._separate_object_fields(ActionItem, actionData) # Create action in database createdAction = self.services.interfaceDbChat.db.recordCreate(ActionItem, simpleFields) # Convert to ActionItem model return ActionItem( id=createdAction["id"], execMethod=createdAction["execMethod"], execAction=createdAction["execAction"], execParameters=createdAction.get("execParameters", {}), execResultLabel=createdAction.get("execResultLabel"), expectedDocumentFormats=createdAction.get("expectedDocumentFormats"), status=createdAction.get("status", TaskStatus.PENDING), error=createdAction.get("error"), retryCount=createdAction.get("retryCount", 0), retryMax=createdAction.get("retryMax", 3), processingTime=createdAction.get("processingTime"), timestamp=float(createdAction.get("timestamp", self.services.utils.getUtcTimestamp())), result=createdAction.get("result"), resultDocuments=createdAction.get("resultDocuments", []), userMessage=createdAction.get("userMessage") ) except Exception as e: logger.error(f"Error creating task action: {str(e)}") return None def _writeTraceLog(self, contextText: str, data: Any) -> None: """Write trace data and categorized React debug files when in DEBUG level""" try: import os import json from datetime import datetime, UTC # Only write if logger is in debug mode if logger.level > logging.DEBUG: return # Get log directory from configuration logDir = self.services.utils.configGet("APP_LOGGING_LOG_DIR", "./") if not os.path.isabs(logDir): # If relative path, make it relative to the gateway directory gatewayDir = os.path.dirname(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))) logDir = os.path.join(gatewayDir, logDir) # Ensure log directory exists os.makedirs(logDir, exist_ok=True) # Create trace file path (aggregate) traceFile = os.path.join(logDir, "log_trace.log") # Derive a React-category filename based on context def _reactFileForContext(text: str) -> str: t = (text or "").lower() if "task plan" in t: if "prompt" in t: return "react_taskplan_prompt.jsonl" if "response" in t: return "react_taskplan_response.jsonl" if "plan selection" in t: if "prompt" in t: return "react_action_selection_prompt.jsonl" if "response" in t: return "react_action_selection_response.jsonl" if "parameters" in t: if "prompt" in t: return "react_parameter_setting_prompt.jsonl" if "response" in t: return "react_parameter_setting_response.jsonl" if "refinement" in t or "review" in t or "decide" in t: # Treat refinement as validation/next-step decision context if "prompt" in t: return "react_validation_prompt.jsonl" if "response" in t: return "react_next_step_decisions.jsonl" if "extraction" in t: if "prompt" in t: return "react_extraction_prompts.jsonl" if "response" in t: return "react_extraction_responses.jsonl" if "generation" in t: if "prompt" in t: return "react_generation_prompts.jsonl" if "response" in t: return "react_generation_responses.jsonl" if "render" in t: if "prompt" in t: return "react_rendering_prompts.jsonl" if "response" in t: return "react_rendering_responses.jsonl" if "validation" in t: if "prompt" in t: return "react_validation_prompts.jsonl" if "response" in t: return "react_validation_responses.jsonl" if "action result" in t: return "react_action_results.jsonl" # Fallback return "react_misc.jsonl" # Daily suffix for React files dateSuffix = datetime.fromtimestamp(self.services.utils.getUtcTimestamp(), UTC).strftime("%Y%m%d") baseReactFile = _reactFileForContext(contextText) name, ext = os.path.splitext(baseReactFile) reactFile = os.path.join(logDir, f"{name}_{dateSuffix}{ext}") # Format the trace entry with better structure timestamp = datetime.fromtimestamp(self.services.utils.getUtcTimestamp(), UTC).strftime("%Y-%m-%d %H:%M:%S.%f")[:-3] # Create a structured trace entry traceEntry = f"[{timestamp}] {contextText}\n" traceEntry += "=" * 80 + "\n" # Add data if provided with improved formatting if data is not None: try: if isinstance(data, (dict, list)): # Format as pretty JSON with better settings jsonStr = json.dumps(data, indent=2, default=str, ensure_ascii=False, sort_keys=False) traceEntry += f"JSON Data:\n{jsonStr}\n" elif isinstance(data, str): # For string data, try to parse as JSON first, then fall back to plain text try: parsed = json.loads(data) jsonStr = json.dumps(parsed, indent=2, default=str, ensure_ascii=False, sort_keys=False) traceEntry += f"JSON Data (parsed from string):\n{jsonStr}\n" except (json.JSONDecodeError, TypeError): # Not valid JSON, show as plain text with proper line breaks formatted_data = data.replace('\\n', '\n') traceEntry += f"Text Data:\n{formatted_data}\n" else: # For other types, convert to string and try to parse as JSON dataStr = str(data) try: parsed = json.loads(dataStr) jsonStr = json.dumps(parsed, indent=2, default=str, ensure_ascii=False, sort_keys=False) traceEntry += f"JSON Data (parsed from object):\n{jsonStr}\n" except (json.JSONDecodeError, TypeError): # Not valid JSON, show as plain text with proper line breaks formatted_data = dataStr.replace('\\n', '\n') traceEntry += f"Object Data:\n{formatted_data}\n" except Exception as e: # Fallback to simple string representation traceEntry += f"Data (fallback): {str(data)}\n" else: traceEntry += "No data provided\n" traceEntry += "=" * 80 + "\n\n" # Write to trace file with open(traceFile, "a", encoding="utf-8") as f: f.write(traceEntry) # Also write a compact JSONL record to the categorized React file try: # Add workflow and step context if available workflowId = getattr(getattr(self, 'workflow', None), 'id', None) roundNumber = getattr(getattr(self, 'workflow', None), 'currentRound', None) taskNumber = getattr(getattr(self, 'workflow', None), 'currentTask', None) actionNumber = getattr(getattr(self, 'workflow', None), 'currentAction', None) reactRecord = { "timestamp": timestamp, "context": contextText, "workflowId": workflowId, "round": roundNumber, "task": taskNumber, "action": actionNumber, "data": data } with open(reactFile, "a", encoding="utf-8") as rf: rf.write(json.dumps(reactRecord, ensure_ascii=False, default=str) + "\n") except Exception: pass except Exception as e: # Don't log trace errors to avoid recursion pass