kubiya-control-plane-api 0.9.15__py3-none-any.whl

control_plane_api/worker/services/workflow_executor_tools.py

@@ -0,0 +1,1748 @@
+"""
+Workflow Executor Tools for Agent Control Plane Worker
+
+This module provides tools for agents to execute workflows defined via
+JSON or Python DSL. Agents can call these tools to run multi-step workflows
+with parameter injection and streaming execution.
+
+Workflows execute remotely on specified runners using the Kubiya SDK.
+"""
+
+import json
+import structlog
+import asyncio
+import os
+import hashlib
+from typing import Optional, Callable, Dict, Any, List
+from agno.tools import Toolkit
+from control_plane_api.worker.skills.builtin.schema_fix_mixin import SchemaFixMixin
+
+logger = structlog.get_logger(__name__)
+
+
+class WorkflowExecutorTools(SchemaFixMixin, Toolkit):
+    """
+    Workflow Executor toolkit for running workflows from agents.
+
+    Agents can use these tools to:
+    - Execute JSON-defined workflows with parameters
+    - Run Python DSL workflows
+    - Stream workflow execution events
+    - Get workflow execution status
+    """
+
+    def __init__(
+        self,
+        name: Optional[str] = None,
+        workflows: Optional[List[Dict[str, Any]]] = None,
+        validation_enabled: bool = True,
+        default_runner: Optional[str] = None,
+        timeout: int = 3600,
+        default_parameters: Optional[Dict[str, Any]] = None,
+        stream_callback: Optional[Callable[[str], None]] = None,
+        kubiya_api_key: Optional[str] = None,
+        kubiya_api_base: Optional[str] = None,
+        execution_id: Optional[str] = None,  # Add execution_id parameter
+        # Legacy parameters for backwards compatibility
+        workflow_type: Optional[str] = None,
+        workflow_definition: Optional[str] = None,
+        python_dsl_code: Optional[str] = None,
+    ):
+        """
+        Initialize WorkflowExecutorTools.
+
+        Args:
+            name: Skill instance name (defaults to "workflow_executor")
+            workflows: List of workflow definitions. Each workflow becomes a separate tool.
+                      Format: [{"name": "analyze-logs", "type": "json", "definition": {...}}, ...]
+            validation_enabled: Enable pre-execution validation
+            default_runner: Default runner/environment name
+            timeout: Maximum execution timeout in seconds
+            default_parameters: Default parameter values to use for all workflows
+            stream_callback: Optional callback for streaming output
+            kubiya_api_key: Kubiya API key (defaults to KUBIYA_API_KEY env var)
+            kubiya_api_base: Kubiya API base URL (defaults to KUBIYA_API_BASE env var)
+            workflow_type: LEGACY - Type of workflow ("json" or "python_dsl")
+            workflow_definition: LEGACY - JSON workflow definition string
+            python_dsl_code: LEGACY - Python DSL code string
+        """
+        super().__init__(name=name or "workflow_executor")
+
+        self.validation_enabled = validation_enabled
+        self.default_runner = default_runner or "default"
+        self.timeout = timeout
+        self.default_parameters = default_parameters or {}
+        self.stream_callback = stream_callback
+        self.execution_id = execution_id or os.environ.get("EXECUTION_ID")  # Store execution_id
+
+        print(f"\n🔍 WORKFLOW EXECUTOR __init__ DEBUG:")
+        print(f"   Received execution_id param: {execution_id}")
+        print(f"   EXECUTION_ID env var: {os.environ.get('EXECUTION_ID')}")
+        print(f"   Final self.execution_id: {self.execution_id}\n")
+
+        # Get Kubiya API credentials from parameters or environment
+        self.kubiya_api_key = kubiya_api_key or os.environ.get("KUBIYA_API_KEY")
+        self.kubiya_api_base = kubiya_api_base or os.environ.get("KUBIYA_API_BASE", "https://api.kubiya.ai")
+
+        if not self.kubiya_api_key:
+            logger.warning("No KUBIYA_API_KEY provided - workflow execution will fail")
+
+        # Get control plane client for publishing events
+        try:
+            from control_plane_api.worker.control_plane_client import get_control_plane_client
+            self.control_plane = get_control_plane_client()
+        except Exception as e:
+            logger.warning(f"Failed to get control plane client: {e}")
+            self.control_plane = None
+
+        # Initialize Kubiya SDK client for remote execution
+        self.kubiya_client = None
+        if self.kubiya_api_key:
+            try:
+                from kubiya import KubiyaClient
+
+                self.kubiya_client = KubiyaClient(
+                    api_key=self.kubiya_api_key,
+                    base_url=self.kubiya_api_base,
+                    runner=self.default_runner,
+                    timeout=self.timeout
+                )
+                logger.info(f"Initialized Kubiya SDK client for remote workflow execution (runner: {self.default_runner})")
+            except ImportError as e:
+                logger.error(f"Failed to import Kubiya SDK: {e}. Install with: pip install git+https://github.com/kubiyabot/sdk-py.git@main")
+                self.kubiya_client = None
+
+        # Handle legacy single workflow format
+        if workflow_definition or python_dsl_code:
+            logger.info("Using legacy single-workflow format")
+
+            legacy_workflow = {
+                "name": "default",
+                "type": workflow_type or "json",
+            }
+            if workflow_type == "json" and workflow_definition:
+                legacy_workflow["definition"] = workflow_definition
+            elif workflow_type == "python_dsl" and python_dsl_code:
+                legacy_workflow["code"] = python_dsl_code
+
+            workflows = [legacy_workflow]
+
+            # Store legacy attributes for backward compatibility
+            self.workflow_type = workflow_type
+            self.workflow_definition = workflow_definition
+            self.python_dsl_code = python_dsl_code
+
+            # Parse workflow data for legacy JSON workflows
+            if workflow_type == "json" and workflow_definition:
+                try:
+                    self.workflow_data = json.loads(workflow_definition) if isinstance(workflow_definition, str) else workflow_definition
+                except Exception as e:
+                    logger.error(f"Failed to parse legacy workflow definition: {e}")
+                    self.workflow_data = None
+            else:
+                self.workflow_data = None
+        else:
+            # Not using legacy format - no legacy attributes
+            self.workflow_type = None
+            self.workflow_definition = None
+            self.python_dsl_code = None
+            self.workflow_data = None
+
+        # Store workflows collection
+        self.workflows = workflows or []
+
+        # Dynamically register a tool for each workflow
+        for workflow in self.workflows:
+            self._register_workflow_tool(workflow)
+
+        # If no workflows registered (empty or legacy format), register default execution tool
+        if not self.workflows or len(self.workflows) == 0:
+            logger.warning("No workflows configured in WorkflowExecutorTools")
+
+        # Register helper tools
+        self.register(self.list_all_workflows)
+        self.register(self.get_workflow_info)
+
+        # Fix: Rebuild function schemas with proper parameters
+        self._rebuild_function_schemas()
+
+    def _register_workflow_tool(self, workflow: Dict[str, Any]):
+        """
+        Dynamically register a tool method for a specific workflow.
+
+        Creates a method named after the workflow that executes it on the configured runner.
+
+        Args:
+            workflow: Workflow definition dict with name, type, and definition/code
+        """
+        workflow_name = workflow.get("name", "unknown")
+        workflow_type = workflow.get("type", "json")
+
+        # Use clean workflow name as method name (replace hyphens/spaces with underscores)
+        # For "analyze-logs" workflow → method name "analyze_logs"
+        # For "default" workflow (legacy) → use the toolkit name
+        safe_name = workflow_name.replace("-", "_").replace(" ", "_").lower()
+
+        # If this is the default workflow, use the skill name
+        if workflow_name == "default" and self.name != "workflow_executor":
+            method_name = self.name
+        else:
+            method_name = safe_name
+
+        # Create a closure that captures the workflow definition
+        def workflow_executor(parameters: Optional[Dict[str, Any]] = None) -> str:
+            f"""
+            Execute the '{workflow_name}' workflow on the configured runner.
+
+            This workflow executes on the runner specified in the workflow definition
+            using the Kubiya SDK. All steps are executed in dependency order.
+
+            Args:
+                parameters: Dictionary of parameters to inject into the workflow.
+                           Parameters can be referenced in workflow steps using {{{{param_name}}}} syntax.
+
+            Returns:
+                str: Formatted workflow execution results including step outputs and status.
+
+            Examples:
+                # Execute workflow with parameters
+                {method_name}(parameters={{"environment": "production", "version": "v1.2.3"}})
+            """
+            return self._execute_specific_workflow(workflow, parameters)
+
+        # Set proper docstring
+        workflow_executor.__doc__ = f"""
+Execute the '{workflow_name}' workflow on the configured runner.
+
+Type: {workflow_type}
+Runner: Specified in workflow definition or default_runner config
+
+Args:
+    parameters: Optional dictionary of parameters to inject into workflow steps.
+                Reference parameters in steps using {{{{param_name}}}} syntax.
+
+Returns:
+    str: Workflow execution results including all step outputs.
+"""
+
+        # Set method name for proper tool registration
+        workflow_executor.__name__ = method_name
+
+        # Register as a tool
+        self.register(workflow_executor)
+
+        # Also set as attribute on self for direct access
+        setattr(self, method_name, workflow_executor)
+
+        logger.info(f"Registered workflow tool: {method_name} for workflow '{workflow_name}'")
+
+    def _execute_specific_workflow(
+        self,
+        workflow: Dict[str, Any],
+        parameters: Optional[Dict[str, Any]] = None
+    ) -> str:
+        """
+        Execute a specific workflow from the collection.
+
+        Args:
+            workflow: Workflow definition
+            parameters: Execution parameters
+
+        Returns:
+            str: Formatted execution result
+        """
+        try:
+            workflow_name = workflow.get("name", "unknown")
+            workflow_type = workflow.get("type", "json")
+
+            # Merge default parameters with runtime parameters
+            # Runtime parameters override defaults
+            params = {**self.default_parameters, **(parameters or {})}
+
+            # Determine runner
+            effective_runner = self.default_runner
+            if workflow_type == "json":
+                workflow_def = workflow.get("definition")
+                if isinstance(workflow_def, str):
+                    workflow_data = json.loads(workflow_def)
+                else:
+                    workflow_data = workflow_def
+
+                effective_runner = workflow_data.get("runner") or self.default_runner
+            else:
+                effective_runner = self.default_runner
+
+            # Stream start message
+            if self.stream_callback:
+                self.stream_callback(
+                    f"🚀 Starting workflow: {workflow_name}\n"
+                    f"   Type: {workflow_type}\n"
+                    f"   Parameters: {json.dumps(params, indent=2)}\n"
+                    f"   Runner: {effective_runner}\n\n"
+                )
+
+            # Execute based on workflow type
+            if workflow_type == "json":
+                result = self._execute_json_workflow_specific(workflow, params, effective_runner)
+            elif workflow_type == "python_dsl":
+                result = self._execute_python_dsl_workflow_specific(workflow, params, effective_runner)
+            else:
+                raise ValueError(f"Unsupported workflow type: {workflow_type}")
+
+            # Stream completion message
+            if self.stream_callback:
+                self.stream_callback(f"\n✅ Workflow '{workflow_name}' completed successfully\n")
+
+            return result
+
+        except Exception as e:
+            error_msg = f"❌ Workflow '{workflow.get('name', 'unknown')}' execution failed: {str(e)}"
+            logger.error(error_msg, exc_info=True)
+
+            if self.stream_callback:
+                self.stream_callback(f"\n{error_msg}\n")
+
+            return error_msg
+
+    def _execute_json_workflow_specific(
+        self,
+        workflow: Dict[str, Any],
+        parameters: Dict[str, Any],
+        runner: str
+    ) -> str:
+        """Execute a JSON workflow."""
+        workflow_def = workflow.get("definition")
+        if isinstance(workflow_def, str):
+            workflow_data = json.loads(workflow_def)
+        else:
+            workflow_data = workflow_def
+
+        if not workflow_data:
+            raise ValueError("No workflow definition available")
+
+        if not self.kubiya_client:
+            raise RuntimeError("Kubiya SDK client not initialized")
+
+        # Ensure runner is set
+        workflow_data["runner"] = runner
+
+        # Remove 'triggers' key if it exists - not needed for direct execution
+        # The DAG builder rejects this key when executing workflows directly
+        if "triggers" in workflow_data:
+            logger.debug(f"Removing 'triggers' key from workflow definition (not needed for execution)")
+            workflow_data.pop("triggers")
+
+        # Execute remotely
+        from datetime import datetime
+        import time
+        start_time = datetime.utcnow()
+
+        # Generate unique message_id for workflow streaming
+        workflow_message_id = f"{self.execution_id}_{int(time.time() * 1000000)}" if self.execution_id else None
+
+        if self.stream_callback:
+            self.stream_callback(f"▶️  Submitting to runner '{runner}'...\n\n")
+
+        # Publish workflow start to control plane
+        print(f"\n{'='*80}")
+        print(f"📡 WORKFLOW STREAMING DEBUG")
+        print(f"{'='*80}")
+        print(f"control_plane exists: {self.control_plane is not None}")
+        print(f"execution_id: {self.execution_id}")
+        print(f"workflow_message_id: {workflow_message_id}")
+        print(f"{'='*80}\n")
+
+        if self.control_plane and self.execution_id and workflow_message_id:
+            try:
+                print(f"📡 Publishing workflow start to control plane...")
+                self.control_plane.publish_event(
+                    execution_id=self.execution_id,
+                    event_type="message_chunk",
+                    data={
+                        "role": "assistant",
+                        "content": f"🚀 Starting workflow: {workflow_data.get('name', 'unknown')}\n▶️  Submitting to runner '{runner}'...\n\n",
+                        "is_chunk": True,
+                        "message_id": workflow_message_id,
+                        "source": "workflow",
+                    }
+                )
+                print(f"✅ Successfully published workflow start to control plane\n")
+            except Exception as e:
+                print(f"❌ Failed to publish workflow start: {e}\n")
+                logger.error(f"❌ Failed to publish workflow start: {e}", exc_info=True)
+        else:
+            print(f"⚠️  Skipping control plane publish (one or more required fields is None)\n")
+
+        # ✅ Enable streaming to capture real-time workflow output
+        response = self.kubiya_client.execute_workflow(
+            workflow_definition=workflow_data,
+            parameters=parameters,
+            stream=True
+        )
+
+        # Accumulate streaming results
+        accumulated_output = []
+        event_count = 0
+        step_outputs = {}
+        current_step = None
+        seen_events = set()  # Track event hashes to prevent duplicates
+
+        # Register workflow for cancellation tracking
+        from control_plane_api.app.services.workflow_cancellation_manager import workflow_cancellation_manager
+        cancellation_event = workflow_cancellation_manager.register_workflow(self.execution_id, workflow_message_id)
+
+        # Iterate over streaming results (SDK yields JSON strings)
+        for event in response:
+            event_count += 1
+
+            # Check for cancellation FIRST (immediate response)
+            if cancellation_event.is_set():
+                logger.warning("⚠️  Workflow execution cancelled by user")
+                workflow_cancellation_manager.clear_cancellation(self.execution_id, workflow_message_id)
+                return f"❌ Workflow execution cancelled by user\n\nWorkflow: {workflow_data.get('name', 'unknown')}\nCancelled at: {datetime.utcnow().isoformat()}"
+
+            # Skip None/empty events
+            if event is None:
+                logger.debug(f"⏭️  Skipping None event #{event_count}")
+                continue
+
+            # 🔍 DEBUG: Print raw event to understand SDK response format
+            print(f"\n{'='*80}")
+            print(f"🔍 RAW SDK EVENT #{event_count}")
+            print(f"{'='*80}")
+            print(f"Type: {type(event).__name__}")
+            print(f"Length: {len(str(event)) if event else 0}")
+            print(f"Repr: {repr(event)[:500]}")
+            if isinstance(event, (str, bytes)):
+                print(f"First 200 chars: {str(event)[:200]}")
+            print(f"{'='*80}\n")
+
+            # Debug: Log raw event with actual content
+            event_repr = repr(event)[:500]  # Use repr to see exact content
+            logger.info(f"📦 Received event #{event_count} (type={type(event).__name__}, length={len(str(event)) if event else 0})")
+            logger.debug(f"   Raw content: {event_repr}")
+
+            # Parse the event (SDK yields JSON strings or bytes)
+            try:
+                if isinstance(event, bytes):
+                    # Skip empty bytes
+                    if not event:
+                        logger.debug(f"   ⏭️  Skipping empty bytes")
+                        continue
+
+                    # Decode bytes to string first
+                    logger.debug(f"   🔄 Decoding bytes to string...")
+                    event = event.decode('utf-8')
+                    logger.debug(f"   ✅ Decoded to string (length={len(event)})")
+
+                if isinstance(event, str):
+                    # Skip empty strings
+                    if not event.strip():
+                        logger.debug(f"   ⏭️  Skipping empty string event")
+                        continue
+
+                    # Handle SSE (Server-Sent Events) format: "data: 2:{json}"
+                    # The SDK sometimes returns events with this prefix
+                    if event.startswith("data: "):
+                        logger.debug(f"   🔄 Stripping SSE 'data: ' prefix...")
+                        event = event[6:]  # Remove "data: " prefix (6 chars)
+
+                        # Also strip the message ID prefix like "2:"
+                        if ":" in event and event.split(":", 1)[0].isdigit():
+                            logger.debug(f"   🔄 Stripping message ID prefix...")
+                            event = event.split(":", 1)[1]  # Remove "2:" or similar prefix
+
+                        logger.debug(f"   ✅ Cleaned SSE event (length={len(event)})")
+
+                    # Try to parse as JSON
+                    logger.debug(f"   🔄 Parsing JSON string...")
+                    event_data = json.loads(event)
+                    logger.debug(f"   ✅ Parsed JSON: type={event_data.get('type', 'unknown')}")
+                elif isinstance(event, dict):
+                    # Already a dict
+                    logger.debug(f"   ✅ Already a dict: type={event.get('type', 'unknown')}")
+                    event_data = event
+                else:
+                    # Unknown type, treat as plain text
+                    logger.warning(f"   ⚠️  Unknown event type: {type(event).__name__}, treating as plain text")
+                    event_str = str(event)
+                    if event_str.strip():  # Only add non-empty text
+                        accumulated_output.append(event_str)
+                        if self.stream_callback:
+                            self.stream_callback(f"{event_str}\n")
+                    continue
+            except (json.JSONDecodeError, UnicodeDecodeError) as e:
+                # If not valid JSON or can't decode, treat as plain text
+                logger.warning(f"   ⚠️  Failed to parse event: {e}, treating as plain text")
+                event_str = str(event)
+                if event_str.strip():  # Only add non-empty text
+                    accumulated_output.append(event_str)
+                    if self.stream_callback:
+                        self.stream_callback(f"{event_str}\n")
+                continue
+
+            # Extract meaningful content based on event type
+            event_type = event_data.get("type", "unknown")
+            logger.info(f"   🎯 Event type: {event_type}")
+
+            # Handle actual Kubiya workflow event types
+            if event_type == "step_output":
+                # step_output contains the actual workflow output in step.output
+                step = event_data.get("step", {})
+                step_name = step.get("name", "unknown")
+                output = step.get("output", "")
+
+                if output.strip():
+                    # Deduplicate events - SDK sends same event twice (plain JSON + SSE format)
+                    event_hash = hashlib.md5(f"{step_name}:{output}".encode()).hexdigest()
+                    if event_hash in seen_events:
+                        print(f"      ⏭️  Skipping duplicate event: {step_name} - {output[:50]}")
+                        logger.info(f"      ⏭️  Skipping duplicate event: {step_name} - {output[:50]}")
+                        continue
+                    seen_events.add(event_hash)
+
+                    logger.info(f"      📝 Step output: {step_name} - {output[:100]}")
+
+                    # Format for display
+                    formatted_output = f"```\n{output}\n```\n"
+
+                    # Stream to callback if provided
+                    if self.stream_callback:
+                        self.stream_callback(formatted_output)
+
+                    # Publish to control plane as message chunk
+                    if self.control_plane and self.execution_id and workflow_message_id:
+                        try:
+                            print(f"📡 Publishing step output to control plane (len={len(formatted_output)})")
+                            result = self.control_plane.publish_event(
+                                execution_id=self.execution_id,
+                                event_type="message_chunk",
+                                data={
+                                    "role": "assistant",
+                                    "content": formatted_output,
+                                    "is_chunk": True,
+                                    "message_id": workflow_message_id,
+                                    "source": "workflow",
+                                    "metadata": {
+                                        "step_name": step_name,
+                                        "event_type": "step_output",
+                                    }
+                                }
+                            )
+                            print(f"✅ Published step output: {result}")
+                        except Exception as e:
+                            print(f"❌ Failed to publish workflow output: {e}")
+                            logger.error(f"Failed to publish workflow output to control plane: {e}", exc_info=True)
+
+                    accumulated_output.append(output)
+
+                    # Track by step
+                    if step_name not in step_outputs:
+                        step_outputs[step_name] = []
+                    step_outputs[step_name].append(output)
+
+            elif event_type == "step_running":
+                # Step is starting
+                step = event_data.get("step", {})
+                step_name = step.get("name", "unknown")
+                current_step = step_name
+                formatted = f"\n▶️  Step: {step_name}"
+                logger.info(f"      ▶️  Starting step: {step_name}")
+                accumulated_output.append(formatted)
+                if self.stream_callback:
+                    self.stream_callback(f"{formatted}\n")
+
+                # Publish to control plane as message chunk
+                if self.control_plane and self.execution_id and workflow_message_id:
+                    try:
+                        self.control_plane.publish_event(
+                            execution_id=self.execution_id,
+                            event_type="message_chunk",
+                            data={
+                                "role": "assistant",
+                                "content": formatted,
+                                "is_chunk": True,
+                                "message_id": workflow_message_id,
+                                "source": "workflow",
+                                "metadata": {
+                                    "step_name": step_name,
+                                    "event_type": "step_start",
+                                }
+                            }
+                        )
+                    except Exception as e:
+                        logger.debug(f"Failed to publish step_running to control plane: {e}")
+
+            elif event_type == "step_complete":
+                # Step finished
+                step = event_data.get("step", {})
+                step_name = step.get("name", "unknown")
+                status = step.get("status", "unknown")
+                icon = "✅" if status == "finished" else "❌"
+                formatted = f"{icon} Step '{step_name}' {status}"
+                logger.info(f"      {icon} Step completed: {step_name} ({status})")
+                accumulated_output.append(formatted)
+                current_step = None
+                if self.stream_callback:
+                    self.stream_callback(f"{formatted}\n")
+
+                # Publish to control plane as message chunk
+                if self.control_plane and self.execution_id and workflow_message_id:
+                    try:
+                        self.control_plane.publish_event(
+                            execution_id=self.execution_id,
+                            event_type="message_chunk",
+                            data={
+                                "role": "assistant",
+                                "content": formatted,
+                                "is_chunk": True,
+                                "message_id": workflow_message_id,
+                                "source": "workflow",
+                                "metadata": {
+                                    "step_name": step_name,
+                                    "status": status,
+                                    "event_type": "step_complete",
+                                }
+                            }
+                        )
+                    except Exception as e:
+                        logger.debug(f"Failed to publish step_complete to control plane: {e}")
+
+            elif event_type == "workflow_complete":
+                # Workflow finished
+                dag_name = event_data.get("dagName", "unknown")
+                status = event_data.get("status", "unknown")
+                success = event_data.get("success", False)
+                icon = "✅" if success else "❌"
+                formatted = f"{icon} Workflow '{dag_name}' {status}"
+                logger.info(f"      {icon} Workflow completed: {dag_name} ({status}, success={success})")
+                accumulated_output.append(formatted)
+                if self.stream_callback:
+                    self.stream_callback(f"{formatted}\n")
+
+            elif event_type == "log":
+                # Filter out noisy internal workflow runner logs
+                message = event_data.get("message", "")
+                level = event_data.get("level", "info")
+
+                # Skip internal/noisy log messages
+                noisy_patterns = [
+                    "[SSE]",
+                    "Published workflow stream event",
+                    "Stored workflow data",
+                    "Emitting log event",
+                    "msg=status requestId",
+                ]
+
+                # Check if message contains any noisy pattern
+                if any(pattern in message for pattern in noisy_patterns):
+                    logger.debug(f"      🔇 Skipping noisy log: {message[:50]}")
+                    continue
+
+                # Only show meaningful log messages
+                formatted = f"[{level.upper()}] {message}"
+                logger.info(f"      💬 Log message: {message[:100]}")
+                accumulated_output.append(formatted)
+                if self.stream_callback:
+                    self.stream_callback(f"{formatted}\n")
+
+            elif event_type == "error":
+                error_msg = event_data.get("message", str(event_data))
+                formatted = f"❌ Error: {error_msg}"
+                logger.error(f"      ❌ Workflow error: {error_msg}")
+                accumulated_output.append(formatted)
+                if self.stream_callback:
+                    self.stream_callback(f"{formatted}\n")
+
+            elif event_type == "heartbeat":
+                # Skip heartbeat events in output
+                logger.debug(f"      💓 Heartbeat (skipping)")
+                continue
+
+            else:
+                # For unknown event types, log but don't show to user
+                logger.info(f"      ❓ Unknown event type: {event_type}")
+                logger.debug(f"      Raw data: {json.dumps(event_data)[:200]}")
+                # Skip unknown events instead of showing raw JSON
+                continue
+
+        end_time = datetime.utcnow()
+        duration = (end_time - start_time).total_seconds()
+
+        # Format complete results for Claude to see
+        result_text = f"\n{'='*60}\n"
+        result_text += f"Workflow Execution: {workflow_data.get('name', 'unknown')}\n"
+        result_text += f"{'='*60}\n\n"
+        result_text += f"Status: ✅ Completed\n"
+        result_text += f"Duration: {duration:.2f}s\n"
+        result_text += f"Runner: {runner}\n"
+        result_text += f"Parameters: {json.dumps(parameters, indent=2)}\n"
+        result_text += f"\nTotal Events: {event_count}\n"
+
+        # Include all captured output in the result
+        if accumulated_output:
+            result_text += f"\n{'='*60}\n"
+            result_text += f"Workflow Output:\n"
+            result_text += f"{'='*60}\n\n"
+            result_text += "\n".join(accumulated_output)
+            logger.info(f"✅ Workflow execution complete: {event_count} events processed, {len(accumulated_output)} output lines accumulated")
+        else:
+            logger.warning(f"⚠️  No workflow output accumulated (received {event_count} events but none produced output)")
+
+        logger.debug(f"Final result preview: {result_text[:500]}")
+
+        # Close the workflow streaming message (empty content marks end of stream)
+        # The agent will process the workflow result and generate its OWN response
+        # with a different message_id - that's the next message the user sees
+        if self.control_plane and self.execution_id and workflow_message_id:
+            try:
+                print(f"\n📡 Closing workflow streaming message...")
+                self.control_plane.publish_event(
+                    execution_id=self.execution_id,
+                    event_type="message_chunk",
+                    data={
+                        "role": "assistant",
+                        "content": "",  # Empty - just marks end of workflow stream
+                        "is_chunk": False,  # Final chunk - closes the streaming message
+                        "message_id": workflow_message_id,
+                        "source": "workflow",
+                        "metadata": {
+                            "event_type": "workflow_stream_end",
+                            "duration": duration,
+                            "total_events": event_count,
+                        }
+                    }
+                )
+                print(f"✅ Workflow stream closed\n")
+            except Exception as e:
+                logger.debug(f"Failed to close workflow stream: {e}")
+
+        # Return result to agent - agent will process and respond with its OWN message_id
+        return result_text
+
+    def _execute_python_dsl_workflow_specific(
+        self,
+        workflow: Dict[str, Any],
+        parameters: Dict[str, Any],
+        runner: str
+    ) -> str:
+        """Execute a Python DSL workflow."""
+        python_code = workflow.get("code")
+        if not python_code:
+            raise ValueError("No Python DSL code available")
+
+        if not self.kubiya_client:
+            raise RuntimeError("Kubiya SDK client not initialized")
+
+        workflow_name = workflow.get("name", "python-dsl-workflow")
+
+        # Create workflow definition for remote execution
+        workflow_definition = {
+            "name": workflow_name,
+            "description": f"Python DSL workflow: {workflow_name}",
+            "runner": runner,
+            "steps": [
+                {
+                    "name": "execute_python_dsl",
+                    "description": "Execute Python DSL workflow code",
+                    "executor": {
+                        "type": "python_dsl",
+                        "config": {"code": python_code}
+                    }
+                }
+            ]
+        }
+
+        from datetime import datetime
+        start_time = datetime.utcnow()
+
+        if self.stream_callback:
+            self.stream_callback(f"▶️  Submitting to runner '{runner}'...\n\n")
+
+        # ✅ Enable streaming to capture real-time workflow output
+        response = self.kubiya_client.execute_workflow(
+            workflow_definition=workflow_definition,
+            parameters=parameters,
+            stream=True
+        )
+
+        # Accumulate streaming results
+        accumulated_output = []
+        event_count = 0
+        step_outputs = {}
+        current_step = None
+        seen_events = set()  # Track event hashes to prevent duplicates
+
+        # Register workflow for cancellation tracking
+        from control_plane_api.app.services.workflow_cancellation_manager import workflow_cancellation_manager
+        cancellation_event = workflow_cancellation_manager.register_workflow(self.execution_id, workflow_message_id)
+
+        # Iterate over streaming results (SDK yields JSON strings)
+        for event in response:
+            event_count += 1
+
+            # Check for cancellation FIRST (immediate response)
+            if cancellation_event.is_set():
+                logger.warning("⚠️  Workflow execution cancelled by user")
+                workflow_cancellation_manager.clear_cancellation(self.execution_id, workflow_message_id)
+                return f"❌ Workflow execution cancelled by user\n\nWorkflow: {workflow_data.get('name', 'unknown')}\nCancelled at: {datetime.utcnow().isoformat()}"
+
+            # Skip None/empty events
+            if event is None:
+                logger.debug(f"⏭️  Skipping None event #{event_count}")
+                continue
+
+            # 🔍 DEBUG: Print raw event to understand SDK response format
+            print(f"\n{'='*80}")
+            print(f"🔍 RAW SDK EVENT #{event_count}")
+            print(f"{'='*80}")
+            print(f"Type: {type(event).__name__}")
+            print(f"Length: {len(str(event)) if event else 0}")
+            print(f"Repr: {repr(event)[:500]}")
+            if isinstance(event, (str, bytes)):
+                print(f"First 200 chars: {str(event)[:200]}")
+            print(f"{'='*80}\n")
+
+            # Debug: Log raw event with actual content
+            event_repr = repr(event)[:500]  # Use repr to see exact content
+            logger.info(f"📦 Received event #{event_count} (type={type(event).__name__}, length={len(str(event)) if event else 0})")
+            logger.debug(f"   Raw content: {event_repr}")
+
+            # Parse the event (SDK yields JSON strings or bytes)
+            try:
+                if isinstance(event, bytes):
+                    # Skip empty bytes
+                    if not event:
+                        logger.debug(f"   ⏭️  Skipping empty bytes")
+                        continue
+
+                    # Decode bytes to string first
+                    logger.debug(f"   🔄 Decoding bytes to string...")
+                    event = event.decode('utf-8')
+                    logger.debug(f"   ✅ Decoded to string (length={len(event)})")
+
+                if isinstance(event, str):
+                    # Skip empty strings
+                    if not event.strip():
+                        logger.debug(f"   ⏭️  Skipping empty string event")
+                        continue
+
+                    # Handle SSE (Server-Sent Events) format: "data: 2:{json}"
+                    # The SDK sometimes returns events with this prefix
+                    if event.startswith("data: "):
+                        logger.debug(f"   🔄 Stripping SSE 'data: ' prefix...")
+                        event = event[6:]  # Remove "data: " prefix (6 chars)
+
+                        # Also strip the message ID prefix like "2:"
+                        if ":" in event and event.split(":", 1)[0].isdigit():
+                            logger.debug(f"   🔄 Stripping message ID prefix...")
+                            event = event.split(":", 1)[1]  # Remove "2:" or similar prefix
+
+                        logger.debug(f"   ✅ Cleaned SSE event (length={len(event)})")
+
+                    # Try to parse as JSON
+                    logger.debug(f"   🔄 Parsing JSON string...")
+                    event_data = json.loads(event)
+                    logger.debug(f"   ✅ Parsed JSON: type={event_data.get('type', 'unknown')}")
+                elif isinstance(event, dict):
+                    # Already a dict
+                    logger.debug(f"   ✅ Already a dict: type={event.get('type', 'unknown')}")
+                    event_data = event
+                else:
+                    # Unknown type, treat as plain text
+                    logger.warning(f"   ⚠️  Unknown event type: {type(event).__name__}, treating as plain text")
+                    event_str = str(event)
+                    if event_str.strip():  # Only add non-empty text
+                        accumulated_output.append(event_str)
+                        if self.stream_callback:
+                            self.stream_callback(f"{event_str}\n")
+                    continue
+            except (json.JSONDecodeError, UnicodeDecodeError) as e:
+                # If not valid JSON or can't decode, treat as plain text
+                logger.warning(f"   ⚠️  Failed to parse event: {e}, treating as plain text")
+                event_str = str(event)
+                if event_str.strip():  # Only add non-empty text
+                    accumulated_output.append(event_str)
+                    if self.stream_callback:
+                        self.stream_callback(f"{event_str}\n")
+                continue
+
+            # Extract meaningful content based on event type
+            event_type = event_data.get("type", "unknown")
+            logger.info(f"   🎯 Event type: {event_type}")
+
+            # Handle actual Kubiya workflow event types
+            if event_type == "step_output":
+                # step_output contains the actual workflow output in step.output
+                step = event_data.get("step", {})
+                step_name = step.get("name", "unknown")
+                output = step.get("output", "")
+
+                if output.strip():
+                    # Deduplicate events - SDK sends same event twice (plain JSON + SSE format)
+                    event_hash = hashlib.md5(f"{step_name}:{output}".encode()).hexdigest()
+                    if event_hash in seen_events:
+                        print(f"      ⏭️  Skipping duplicate event: {step_name} - {output[:50]}")
+                        logger.info(f"      ⏭️  Skipping duplicate event: {step_name} - {output[:50]}")
+                        continue
+                    seen_events.add(event_hash)
+
+                    logger.info(f"      📝 Step output: {step_name} - {output[:100]}")
+
+                    # Format for display
+                    formatted_output = f"```\n{output}\n```\n"
+
+                    # Stream to callback if provided
+                    if self.stream_callback:
+                        self.stream_callback(formatted_output)
+
+                    # Publish to control plane as message chunk
+                    if self.control_plane and self.execution_id and workflow_message_id:
+                        try:
+                            print(f"📡 Publishing step output to control plane (len={len(formatted_output)})")
+                            result = self.control_plane.publish_event(
+                                execution_id=self.execution_id,
+                                event_type="message_chunk",
+                                data={
+                                    "role": "assistant",
+                                    "content": formatted_output,
+                                    "is_chunk": True,
+                                    "message_id": workflow_message_id,
+                                    "source": "workflow",
+                                    "metadata": {
+                                        "step_name": step_name,
+                                        "event_type": "step_output",
+                                    }
+                                }
+                            )
+                            print(f"✅ Published step output: {result}")
+                        except Exception as e:
+                            print(f"❌ Failed to publish workflow output: {e}")
+                            logger.error(f"Failed to publish workflow output to control plane: {e}", exc_info=True)
+
+                    accumulated_output.append(output)
+
+                    # Track by step
+                    if step_name not in step_outputs:
+                        step_outputs[step_name] = []
+                    step_outputs[step_name].append(output)
+
+            elif event_type == "step_running":
+                # Step is starting
+                step = event_data.get("step", {})
+                step_name = step.get("name", "unknown")
+                current_step = step_name
+                formatted = f"\n▶️  Step: {step_name}"
+                logger.info(f"      ▶️  Starting step: {step_name}")
+                accumulated_output.append(formatted)
+                if self.stream_callback:
+                    self.stream_callback(f"{formatted}\n")
+
+                # Publish to control plane as message chunk
+                if self.control_plane and self.execution_id and workflow_message_id:
+                    try:
+                        self.control_plane.publish_event(
+                            execution_id=self.execution_id,
+                            event_type="message_chunk",
+                            data={
+                                "role": "assistant",
+                                "content": formatted,
+                                "is_chunk": True,
+                                "message_id": workflow_message_id,
+                                "source": "workflow",
+                                "metadata": {
+                                    "step_name": step_name,
+                                    "event_type": "step_start",
+                                }
+                            }
+                        )
+                    except Exception as e:
+                        logger.debug(f"Failed to publish step_running to control plane: {e}")
+
+            elif event_type == "step_complete":
+                # Step finished
+                step = event_data.get("step", {})
+                step_name = step.get("name", "unknown")
+                status = step.get("status", "unknown")
+                icon = "✅" if status == "finished" else "❌"
+                formatted = f"{icon} Step '{step_name}' {status}"
+                logger.info(f"      {icon} Step completed: {step_name} ({status})")
+                accumulated_output.append(formatted)
+                current_step = None
+                if self.stream_callback:
+                    self.stream_callback(f"{formatted}\n")
+
+                # Publish to control plane as message chunk
+                if self.control_plane and self.execution_id and workflow_message_id:
+                    try:
+                        self.control_plane.publish_event(
+                            execution_id=self.execution_id,
+                            event_type="message_chunk",
+                            data={
+                                "role": "assistant",
+                                "content": formatted,
+                                "is_chunk": True,
+                                "message_id": workflow_message_id,
+                                "source": "workflow",
+                                "metadata": {
+                                    "step_name": step_name,
+                                    "status": status,
+                                    "event_type": "step_complete",
+                                }
+                            }
+                        )
+                    except Exception as e:
+                        logger.debug(f"Failed to publish step_complete to control plane: {e}")
+
+            elif event_type == "workflow_complete":
+                # Workflow finished
+                dag_name = event_data.get("dagName", "unknown")
+                status = event_data.get("status", "unknown")
+                success = event_data.get("success", False)
+                icon = "✅" if success else "❌"
+                formatted = f"{icon} Workflow '{dag_name}' {status}"
+                logger.info(f"      {icon} Workflow completed: {dag_name} ({status}, success={success})")
+                accumulated_output.append(formatted)
+                if self.stream_callback:
+                    self.stream_callback(f"{formatted}\n")
+
+            elif event_type == "log":
+                # Filter out noisy internal workflow runner logs
+                message = event_data.get("message", "")
+                level = event_data.get("level", "info")
+
+                # Skip internal/noisy log messages
+                noisy_patterns = [
+                    "[SSE]",
+                    "Published workflow stream event",
+                    "Stored workflow data",
+                    "Emitting log event",
+                    "msg=status requestId",
+                ]
+
+                # Check if message contains any noisy pattern
+                if any(pattern in message for pattern in noisy_patterns):
+                    logger.debug(f"      🔇 Skipping noisy log: {message[:50]}")
+                    continue
+
+                # Only show meaningful log messages
+                formatted = f"[{level.upper()}] {message}"
+                logger.info(f"      💬 Log message: {message[:100]}")
+                accumulated_output.append(formatted)
+                if self.stream_callback:
+                    self.stream_callback(f"{formatted}\n")
+
+            elif event_type == "error":
+                error_msg = event_data.get("message", str(event_data))
+                formatted = f"❌ Error: {error_msg}"
+                logger.error(f"      ❌ Workflow error: {error_msg}")
+                accumulated_output.append(formatted)
+                if self.stream_callback:
+                    self.stream_callback(f"{formatted}\n")
+
+            elif event_type == "heartbeat":
+                # Skip heartbeat events in output
+                logger.debug(f"      💓 Heartbeat (skipping)")
+                continue
+
+            else:
+                # For unknown event types, log but don't show to user
+                logger.info(f"      ❓ Unknown event type: {event_type}")
+                logger.debug(f"      Raw data: {json.dumps(event_data)[:200]}")
+                # Skip unknown events instead of showing raw JSON
+                continue
+
+        end_time = datetime.utcnow()
+        duration = (end_time - start_time).total_seconds()
+
+        result_text = f"\n{'='*60}\n"
+        result_text += f"Python DSL Workflow: {workflow_name}\n"
+        result_text += f"{'='*60}\n\n"
+        result_text += f"Status: ✅ Completed\n"
+        result_text += f"Duration: {duration:.2f}s\n"
+        result_text += f"Runner: {runner}\n"
+        result_text += f"\nTotal Events: {event_count}\n"
+
+        # Include all captured output in the result
+        if accumulated_output:
+            result_text += f"\n{'='*60}\n"
+            result_text += f"Workflow Output:\n"
+            result_text += f"{'='*60}\n\n"
+            result_text += "\n".join(accumulated_output)
+            logger.info(f"✅ Workflow execution complete: {event_count} events processed, {len(accumulated_output)} output lines accumulated")
+        else:
+            logger.warning(f"⚠️  No workflow output accumulated (received {event_count} events but none produced output)")
+
+        logger.debug(f"Final result preview: {result_text[:500]}")
+
+        # Close the workflow streaming message (empty content marks end of stream)
+        # The agent will process the workflow result and generate its OWN response
+        # with a different message_id - that's the next message the user sees
+        if self.control_plane and self.execution_id and workflow_message_id:
+            try:
+                print(f"\n📡 Closing workflow streaming message...")
+                self.control_plane.publish_event(
+                    execution_id=self.execution_id,
+                    event_type="message_chunk",
+                    data={
+                        "role": "assistant",
+                        "content": "",  # Empty - just marks end of workflow stream
+                        "is_chunk": False,  # Final chunk - closes the streaming message
+                        "message_id": workflow_message_id,
+                        "source": "workflow",
+                        "metadata": {
+                            "event_type": "workflow_stream_end",
+                            "duration": duration,
+                            "total_events": event_count,
+                        }
+                    }
+                )
+                print(f"✅ Workflow stream closed\n")
+            except Exception as e:
+                logger.debug(f"Failed to close workflow stream: {e}")
+
+        # Return result to agent - agent will process and respond with its OWN message_id
+        return result_text
+
+    def list_all_workflows(self) -> str:
+        """
+        List all available workflows in this skill instance.
+
+        Returns:
+            str: Formatted list of all workflows with their names, types, and descriptions.
+
+        Examples:
+            # List all workflows
+            list_all_workflows()
+        """
+        if not self.workflows:
+            return "No workflows defined in this skill instance."
+
+        result = f"\n📋 Available Workflows ({len(self.workflows)}):\n"
+        result += "=" * 60 + "\n\n"
+
+        for idx, workflow in enumerate(self.workflows, 1):
+            name = workflow.get("name", "unknown")
+            wf_type = workflow.get("type", "unknown")
+            safe_name = name.replace("-", "_").replace(" ", "_").lower()
+
+            result += f"{idx}. {name} ({wf_type})\n"
+            result += f"   Tool: execute_workflow_{safe_name}()\n"
+
+            # Get description from workflow definition
+            if wf_type == "json":
+                wf_def = workflow.get("definition")
+                if isinstance(wf_def, str):
+                    try:
+                        wf_data = json.loads(wf_def)
+                        desc = wf_data.get("description", "No description")
+                        steps = len(wf_data.get("steps", []))
+                        result += f"   Description: {desc}\n"
+                        result += f"   Steps: {steps}\n"
+                    except:
+                        pass
+                elif isinstance(wf_def, dict):
+                    desc = wf_def.get("description", "No description")
+                    steps = len(wf_def.get("steps", []))
+                    result += f"   Description: {desc}\n"
+                    result += f"   Steps: {steps}\n"
+
+            result += "\n"
+
+        return result
+
+    def execute_workflow(
+        self,
+        parameters: Optional[Dict[str, Any]] = None,
+        override_timeout: Optional[int] = None,
+    ) -> str:
+        """
+        Execute the first configured workflow with the provided parameters.
+
+        LEGACY METHOD: For backward compatibility with single-workflow format.
+        For multi-workflow skills, use execute_workflow_<name>() methods instead.
+
+        This tool allows agents to run multi-step workflows by providing
+        parameter values. The workflow will execute all steps in dependency
+        order and return the results.
+
+        The runner/environment is determined from the workflow definition itself,
+        not passed as a parameter. This ensures workflows execute in their
+        intended environments.
+
+        Args:
+            parameters: Dictionary of parameters to inject into the workflow.
+                       These can be referenced in workflow steps using {{param_name}} syntax.
+            override_timeout: Optional timeout override in seconds.
+                            If not provided, uses the timeout from configuration.
+
+        Returns:
+            str: A formatted string containing the workflow execution results,
+                 including step outputs and any errors encountered.
+
+        Examples:
+            # Execute a deployment workflow with environment parameter
+            execute_workflow(parameters={"environment": "production", "version": "v1.2.3"})
+
+            # Execute with timeout override
+            execute_workflow(
+                parameters={"data_source": "s3://bucket/data"},
+                override_timeout=7200
+            )
+        """
+        try:
+            # For multi-workflow format, execute the first workflow
+            if self.workflows:
+                if len(self.workflows) > 1:
+                    logger.warning(
+                        "Multiple workflows defined but execute_workflow() called. "
+                        "Executing first workflow. Use execute_workflow_<name>() for specific workflows."
+                    )
+                return self._execute_specific_workflow(self.workflows[0], parameters)
+
+            # Legacy single-workflow format
+            # Use provided parameters or empty dict
+            params = parameters or {}
+
+            # Determine runner from workflow definition or use default_runner/default_runner from config
+            effective_runner = None
+            if hasattr(self, 'workflow_type') and self.workflow_type == "json" and hasattr(self, 'workflow_data') and self.workflow_data:
+                # Get runner from workflow definition first, then step-level, then default
+                effective_runner = self.workflow_data.get("runner") or self.default_runner
+            else:
+                effective_runner = self.default_runner
+
+            # Determine timeout
+            effective_timeout = override_timeout or self.timeout
+
+            # Stream start message
+            if self.stream_callback:
+                self.stream_callback(
+                    f"🚀 Starting workflow execution...\n"
+                    f"   Workflow Type: {getattr(self, 'workflow_type', 'unknown')}\n"
+                    f"   Parameters: {json.dumps(params, indent=2)}\n"
+                    f"   Runner: {effective_runner or 'default'}\n"
+                    f"   Timeout: {effective_timeout}s\n\n"
+                )
+
+            # Execute based on workflow type
+            if hasattr(self, 'workflow_type'):
+                if self.workflow_type == "json":
+                    result = self._execute_json_workflow(params, effective_runner, effective_timeout)
+                elif self.workflow_type == "python_dsl":
+                    result = self._execute_python_dsl_workflow(params, effective_runner, effective_timeout)
+                else:
+                    raise ValueError(f"Unsupported workflow type: {self.workflow_type}")
+            else:
+                raise ValueError("No workflow configured")
+
+            # Stream completion message
+            if self.stream_callback:
+                self.stream_callback(f"\n✅ Workflow execution completed successfully\n")
+
+            return result
+
+        except Exception as e:
+            error_msg = f"❌ Workflow execution failed: {str(e)}"
+            logger.error(error_msg, exc_info=True)
+
+            if self.stream_callback:
+                self.stream_callback(f"\n{error_msg}\n")
+
+            return error_msg
+
+    def _execute_json_workflow(
+        self,
+        parameters: Dict[str, Any],
+        runner: Optional[str],
+        timeout: int
+    ) -> str:
+        """Execute a JSON workflow using kubiya SDK (remote execution)."""
+        if not self.workflow_data:
+            raise ValueError("No workflow definition available")
+
+        if not self.kubiya_client:
+            raise RuntimeError("Kubiya SDK client not initialized - cannot execute workflow remotely")
+
+        workflow_name = self.workflow_data.get("name", "unknown")
+        steps = self.workflow_data.get("steps", [])
+
+        if self.stream_callback:
+            self.stream_callback(f"📋 Workflow: {workflow_name}\n")
+            self.stream_callback(f"   Steps: {len(steps)}\n")
+            self.stream_callback(f"   Runner: {runner or self.default_runner}\n\n")
+
+        try:
+            # Execute workflow remotely using Kubiya SDK
+            from datetime import datetime
+            start_time = datetime.utcnow()
+
+            if self.stream_callback:
+                self.stream_callback(f"▶️  Submitting to runner '{runner or self.default_runner}'...\n\n")
+
+            # Submit workflow definition to remote runner
+            # The workflow_data already contains the complete workflow definition
+            workflow_def = dict(self.workflow_data)
+
+            # Ensure runner is set correctly
+            workflow_def["runner"] = runner or self.default_runner
+
+            # Remove 'triggers' key if it exists - not needed for direct execution
+            # The DAG builder rejects this key when executing workflows directly
+            if "triggers" in workflow_def:
+                logger.debug(f"Removing 'triggers' key from workflow definition (not needed for execution)")
+                workflow_def.pop("triggers")
+
+            # ✅ Enable streaming to capture real-time workflow output
+            response = self.kubiya_client.execute_workflow(
+                workflow_definition=workflow_def,
+                parameters=parameters,
+                stream=True
+            )
+
+            # Accumulate streaming results
+            accumulated_output = []
+            event_count = 0
+
+            # Iterate over streaming results
+            for event in response:
+                event_count += 1
+
+                # Stream to user in real-time
+                if self.stream_callback:
+                    if isinstance(event, str):
+                        self.stream_callback(f"{event}\n")
+                        accumulated_output.append(event)
+                    elif isinstance(event, dict):
+                        event_type = event.get("type", "event")
+                        event_data = event.get("data", event)
+                        formatted_event = f"[{event_type}] {json.dumps(event_data, indent=2)}"
+                        self.stream_callback(f"{formatted_event}\n")
+                        accumulated_output.append(formatted_event)
+                    else:
+                        formatted_event = str(event)
+                        self.stream_callback(f"{formatted_event}\n")
+                        accumulated_output.append(formatted_event)
+
+            end_time = datetime.utcnow()
+            duration = (end_time - start_time).total_seconds()
+
+            # Format results
+            result_text = f"\n{'='*60}\n"
+            result_text += f"Workflow Execution Summary\n"
+            result_text += f"{'='*60}\n\n"
+            result_text += f"Workflow: {workflow_name}\n"
+            result_text += f"Runner: {runner or self.default_runner}\n"
+            result_text += f"Status: ✅ Completed\n"
+            result_text += f"Duration: {duration:.2f}s\n"
+            result_text += f"Steps: {len(steps)}\n"
+            result_text += f"Parameters: {json.dumps(parameters, indent=2)}\n"
+            result_text += f"\nTotal Events: {event_count}\n"
+
+            # Include all captured output in the result
+            if accumulated_output:
+                result_text += f"\n{'='*60}\n"
+                result_text += f"Workflow Output:\n"
+                result_text += f"{'='*60}\n\n"
+                result_text += "\n".join(accumulated_output)
+
+            if self.stream_callback:
+                self.stream_callback(f"\n✅ Workflow execution completed in {duration:.2f}s\n")
+
+            return result_text
+
+        except Exception as e:
+            error_msg = f"JSON workflow execution failed: {str(e)}"
+            logger.error(error_msg, exc_info=True)
+            raise RuntimeError(error_msg)
+
+    def _execute_python_dsl_workflow(
+        self,
+        parameters: Dict[str, Any],
+        runner: Optional[str],
+        timeout: int
+    ) -> str:
+        """Execute a Python DSL workflow using kubiya SDK (remote execution)."""
+        if not self.python_dsl_code:
+            raise ValueError("No Python DSL code available")
+
+        if not self.kubiya_client:
+            raise RuntimeError("Kubiya SDK client not initialized - cannot execute workflow remotely")
+
+        if self.stream_callback:
+            self.stream_callback(f"🐍 Submitting Python DSL workflow for remote execution...\n\n")
+
+        try:
+            # Parse the Python DSL code to extract workflow name
+            # For now, we'll create a workflow definition that the runner can execute
+            workflow_name = "python-dsl-workflow"
+
+            # Try to extract workflow name from code
+            if "name=" in self.python_dsl_code:
+                try:
+                    import re
+                    match = re.search(r'name\s*=\s*["\']([^"\']+)["\']', self.python_dsl_code)
+                    if match:
+                        workflow_name = match.group(1)
+                except:
+                    pass
+
+            if self.stream_callback:
+                self.stream_callback(f"📋 Workflow: {workflow_name}\n")
+                self.stream_callback(f"   Runner: {runner or self.default_runner}\n")
+                self.stream_callback(f"   Parameters: {json.dumps(parameters)}\n\n")
+
+            # Create workflow definition for remote execution
+            # The runner will execute the Python DSL code
+            workflow_definition = {
+                "name": workflow_name,
+                "description": "Python DSL workflow",
+                "runner": runner or self.default_runner,
+                "steps": [
+                    {
+                        "name": "execute_python_dsl",
+                        "description": "Execute Python DSL workflow code",
+                        "executor": {
+                            "type": "python_dsl",
+                            "config": {
+                                "code": self.python_dsl_code
+                            }
+                        }
+                    }
+                ]
+            }
+
+            # Execute workflow remotely using Kubiya SDK
+            from datetime import datetime
+            start_time = datetime.utcnow()
+
+            if self.stream_callback:
+                self.stream_callback(f"▶️  Submitting to runner '{runner or self.default_runner}'...\n\n")
+
+            # ✅ Enable streaming to capture real-time workflow output
+            response = self.kubiya_client.execute_workflow(
+                workflow_definition=workflow_definition,
+                parameters=parameters,
+                stream=True
+            )
+
+            # Accumulate streaming results
+            accumulated_output = []
+            event_count = 0
+
+            # Iterate over streaming results
+            for event in response:
+                event_count += 1
+
+                # Stream to user in real-time
+                if self.stream_callback:
+                    if isinstance(event, str):
+                        self.stream_callback(f"{event}\n")
+                        accumulated_output.append(event)
+                    elif isinstance(event, dict):
+                        event_type = event.get("type", "event")
+                        event_data = event.get("data", event)
+                        formatted_event = f"[{event_type}] {json.dumps(event_data, indent=2)}"
+                        self.stream_callback(f"{formatted_event}\n")
+                        accumulated_output.append(formatted_event)
+                    else:
+                        formatted_event = str(event)
+                        self.stream_callback(f"{formatted_event}\n")
+                        accumulated_output.append(formatted_event)
+
+            end_time = datetime.utcnow()
+            duration = (end_time - start_time).total_seconds()
+
+            # Format results
+            result_text = f"\n{'='*60}\n"
+            result_text += f"Python DSL Workflow Execution Summary\n"
+            result_text += f"{'='*60}\n\n"
+            result_text += f"Workflow: {workflow_name}\n"
+            result_text += f"Runner: {runner or self.default_runner}\n"
+            result_text += f"Status: ✅ Completed\n"
+            result_text += f"Duration: {duration:.2f}s\n"
+            result_text += f"Parameters: {json.dumps(parameters, indent=2)}\n"
+            result_text += f"\nTotal Events: {event_count}\n"
+
+            # Include all captured output in the result
+            if accumulated_output:
+                result_text += f"\n{'='*60}\n"
+                result_text += f"Workflow Output:\n"
+                result_text += f"{'='*60}\n\n"
+                result_text += "\n".join(accumulated_output)
+
+            if self.stream_callback:
+                self.stream_callback(f"\n✅ Workflow execution completed in {duration:.2f}s\n")
+
+            return result_text
+
+        except Exception as e:
+            error_msg = f"Python DSL workflow execution failed: {str(e)}"
+            logger.error(error_msg, exc_info=True)
+            raise RuntimeError(error_msg)
+
+    def _inject_parameters(self, config: Dict[str, Any], parameters: Dict[str, Any]) -> Dict[str, Any]:
+        """Inject parameters into configuration values."""
+        result = {}
+
+        for key, value in config.items():
+            if isinstance(value, str):
+                # Replace {{param_name}} with parameter value
+                for param_name, param_value in parameters.items():
+                    value = value.replace(f"{{{{{param_name}}}}}", str(param_value))
+                result[key] = value
+            elif isinstance(value, dict):
+                result[key] = self._inject_parameters(value, parameters)
+            elif isinstance(value, list):
+                result[key] = [
+                    self._inject_parameters(item, parameters) if isinstance(item, dict)
+                    else str(item).replace(f"{{{{{pn}}}}}", str(pv)) if isinstance(item, str) else item
+                    for item in value
+                    for pn, pv in [(pn, pv)]
+                    for pn, pv in [(list(parameters.keys())[0] if parameters else "", list(parameters.values())[0] if parameters else "")]
+                ][:len(value)]
+                # Simplified version
+                result[key] = value
+            else:
+                result[key] = value
+
+        return result
+
+    def list_workflow_steps(self, workflow_name: Optional[str] = None) -> str:
+        """
+        List all steps in the configured workflow(s).
+
+        LEGACY METHOD: For multi-workflow skills, this lists all workflows.
+        For legacy single-workflow format, lists steps of that workflow.
+
+        Args:
+            workflow_name: Optional workflow name to filter by (multi-workflow only)
+
+        Returns:
+            str: A formatted string listing all workflow steps with their
+                 descriptions, executor types, and dependencies.
+
+        Examples:
+            # List all steps in the workflow
+            list_workflow_steps()
+        """
+        try:
+            # Multi-workflow format
+            if self.workflows:
+                if workflow_name:
+                    # Find specific workflow
+                    workflow = next((w for w in self.workflows if w.get("name") == workflow_name), None)
+                    if not workflow:
+                        return f"❌ Workflow '{workflow_name}' not found"
+                    workflows_to_show = [workflow]
+                else:
+                    workflows_to_show = self.workflows
+
+                result = f"\n📋 Workflows: {len(workflows_to_show)}\n"
+                result += "=" * 60 + "\n\n"
+
+                for wf in workflows_to_show:
+                    wf_name = wf.get("name", "unknown")
+                    wf_type = wf.get("type", "unknown")
+
+                    result += f"Workflow: {wf_name} ({wf_type})\n"
+
+                    if wf_type == "json":
+                        wf_def = wf.get("definition")
+                        if isinstance(wf_def, str):
+                            try:
+                                wf_data = json.loads(wf_def)
+                            except:
+                                result += "   ❌ Invalid JSON definition\n\n"
+                                continue
+                        else:
+                            wf_data = wf_def
+
+                        if wf_data:
+                            workflow_desc = wf_data.get("description", "No description")
+                            steps = wf_data.get("steps", [])
+
+                            result += f"   Description: {workflow_desc}\n"
+                            result += f"   Total Steps: {len(steps)}\n\n"
+
+                            if steps:
+                                result += "   Steps:\n"
+                                for idx, step in enumerate(steps, 1):
+                                    step_name = step.get("name", "unknown")
+                                    step_desc = step.get("description", "")
+                                    executor = step.get("executor", {})
+                                    executor_type = executor.get("type", "unknown")
+                                    depends_on = step.get("depends_on", [])
+
+                                    result += f"   {idx}. {step_name}\n"
+                                    if step_desc:
+                                        result += f"      Description: {step_desc}\n"
+                                    result += f"      Executor: {executor_type}\n"
+                                    if depends_on:
+                                        result += f"      Depends on: {', '.join(depends_on)}\n"
+                            else:
+                                result += "   (No steps defined)\n"
+
+                    elif wf_type == "python_dsl":
+                        result += "   Type: Python DSL\n"
+                        result += "   (To view steps, execute the workflow)\n"
+
+                    result += "\n"
+
+                return result
+
+            # Legacy single-workflow format
+            if self.workflow_type == "json":
+                if not self.workflow_data:
+                    return "❌ No workflow definition available"
+
+                workflow_name_legacy = self.workflow_data.get("name", "unknown")
+                workflow_desc = self.workflow_data.get("description", "No description")
+                steps = self.workflow_data.get("steps", [])
+
+                result = f"\n📋 Workflow: {workflow_name_legacy}\n"
+                result += f"   Description: {workflow_desc}\n"
+                result += f"   Total Steps: {len(steps)}\n\n"
+
+                if not steps:
+                    result += "   (No steps defined)\n"
+                    return result
+
+                result += "Steps:\n"
+                for idx, step in enumerate(steps, 1):
+                    step_name = step.get("name", "unknown")
+                    step_desc = step.get("description", "")
+                    executor = step.get("executor", {})
+                    executor_type = executor.get("type", "unknown")
+                    depends_on = step.get("depends_on", [])
+
+                    result += f"\n{idx}. {step_name}\n"
+                    if step_desc:
+                        result += f"   Description: {step_desc}\n"
+                    result += f"   Executor: {executor_type}\n"
+                    if depends_on:
+                        result += f"   Depends on: {', '.join(depends_on)}\n"
+
+                return result
+
+            elif self.workflow_type == "python_dsl":
+                return f"\n🐍 Python DSL Workflow\n\nTo view steps, execute the workflow.\n"
+
+            else:
+                return "❌ No workflow configured"
+
+        except Exception as e:
+            logger.error(f"Failed to list workflow steps: {e}", exc_info=True)
+            return f"❌ Error listing workflow steps: {str(e)}"
+
+    def get_workflow_info(self) -> str:
+        """
+        Get detailed information about the configured workflow(s).
+
+        This tool provides comprehensive information about the workflow
+        including its name, description, type, number of steps, triggers,
+        and configuration.
+
+        For multi-workflow skills, lists all workflows with their configurations.
+        For legacy single-workflow format, shows that workflow's information.
+
+        Returns:
+            str: A formatted string with complete workflow information.
+
+        Examples:
+            # Get workflow information
+            get_workflow_info()
+        """
+        try:
+            result = f"\n{'='*60}\n"
+            result += f"Workflow Executor Information\n"
+            result += f"{'='*60}\n\n"
+
+            result += f"Validation Enabled: {self.validation_enabled}\n"
+            result += f"Timeout: {self.timeout}s\n"
+            result += f"Default Runner: {self.default_runner or 'None'}\n"
+            result += f"Total Workflows: {len(self.workflows)}\n\n"
+
+            # Multi-workflow format
+            if self.workflows:
+                result += "Configured Workflows:\n"
+                result += "-" * 60 + "\n\n"
+
+                for idx, workflow in enumerate(self.workflows, 1):
+                    wf_name = workflow.get("name", "unknown")
+                    wf_type = workflow.get("type", "unknown")
+                    safe_name = wf_name.replace("-", "_").replace(" ", "_").lower()
+
+                    result += f"{idx}. {wf_name}\n"
+                    result += f"   Type: {wf_type}\n"
+                    result += f"   Tool: execute_workflow_{safe_name}()\n"
+
+                    if wf_type == "json":
+                        wf_def = workflow.get("definition")
+                        if isinstance(wf_def, str):
+                            try:
+                                wf_data = json.loads(wf_def)
+                            except:
+                                result += "   ❌ Invalid JSON definition\n\n"
+                                continue
+                        else:
+                            wf_data = wf_def
+
+                        if wf_data:
+                            workflow_desc = wf_data.get("description", "No description")
+                            steps = wf_data.get("steps", [])
+                            triggers = wf_data.get("triggers", [])
+                            workflow_runner = wf_data.get("runner")
+
+                            result += f"   Description: {workflow_desc}\n"
+                            result += f"   Steps: {len(steps)}\n"
+                            result += f"   Triggers: {len(triggers)}\n"
+
+                            # Show runner hierarchy
+                            if workflow_runner:
+                                result += f"   Runner: {workflow_runner} (specified in workflow)\n"
+                            elif self.default_runner:
+                                result += f"   Runner: {self.default_runner} (from skill config)\n"
+                            else:
+                                result += f"   Runner: default (no runner specified)\n"
+
+                    elif wf_type == "python_dsl":
+                        python_code = workflow.get("code", "")
+                        result += f"   Code Length: {len(python_code)} characters\n"
+
+                    result += "\n"
+
+                return result
+
+            # Legacy single-workflow format
+            result += f"Type: {self.workflow_type or 'none'}\n\n"
+
+            if self.workflow_type == "json" and self.workflow_data:
+                workflow_name = self.workflow_data.get("name", "unknown")
+                workflow_desc = self.workflow_data.get("description", "No description")
+                steps = self.workflow_data.get("steps", [])
+                triggers = self.workflow_data.get("triggers", [])
+                workflow_runner = self.workflow_data.get("runner")
+
+                result += f"Name: {workflow_name}\n"
+                result += f"Description: {workflow_desc}\n"
+                result += f"Steps: {len(steps)}\n"
+                result += f"Triggers: {len(triggers)}\n"
+
+                # Show runner hierarchy
+                if workflow_runner:
+                    result += f"Workflow Runner: {workflow_runner} (will be used for execution)\n"
+                elif self.default_runner:
+                    result += f"Workflow Runner: {self.default_runner} (from skill config)\n"
+                else:
+                    result += f"Workflow Runner: default (no runner specified)\n"
+
+            elif self.workflow_type == "python_dsl":
+                result += f"Python DSL Workflow\n"
+                result += f"Code Length: {len(self.python_dsl_code or '')} characters\n"
+
+            return result
+
+        except Exception as e:
+            logger.error(f"Failed to get workflow info: {e}", exc_info=True)
+            return f"❌ Error getting workflow info: {str(e)}"