erdo 0.1.4__py3-none-any.whl

erdo/types.py

@@ -0,0 +1,1142 @@
+"""
+Erdo Agent SDK - Core Types
+
+This file contains the main SDK classes for building AI agents.
+Auto-generated types are imported from the generated module.
+"""
+
+import os
+from enum import Enum
+from pathlib import Path
+from typing import Any, Callable, Dict, List, Optional, Protocol, Tuple, Union, cast
+
+from pydantic import BaseModel, Field
+
+from ._generated.types import (
+    ExecutionModeType,
+    HandlerType,
+    OutputContentType,
+    OutputVisibility,
+    ParameterDefinition,
+    ParameterHydrationBehaviour,
+    Tool,
+)
+from .template import TemplateString
+
+# Type aliases for complex types that are json.RawMessage in Go
+ExecutionCondition = Dict[str, Any]  # Complex execution condition configuration
+
+
+# Protocols for better type safety
+class ActionProtocol(Protocol):
+    """Protocol for action objects that can be converted to parameters."""
+
+    name: str
+
+    def model_dump(self) -> Dict[str, Any]: ...
+
+
+class StepLike(Protocol):
+    """Protocol for step-like objects."""
+
+    key: Optional[str]
+    id: Optional[str]
+
+
+class ConditionProtocol(Protocol):
+    """Protocol for condition objects."""
+
+    def to_dict(self) -> Dict[str, Any]: ...
+
+
+class ExecutionModeProtocol(Protocol):
+    """Protocol for execution mode objects."""
+
+    def to_dict(self) -> Dict[str, Any]: ...
+
+
+class PythonFile(BaseModel):
+    """Reference to a Python file for code execution.
+
+    Can be used in code_files parameter to reference local Python files.
+    When used, the file content will be automatically loaded during export.
+    """
+
+    filename: str = Field(
+        ..., description="Path to the file (e.g., 'analyze_file_files/analyze.py')"
+    )
+
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dict format for serialization."""
+        return {
+            "filename": self.filename,
+            "_type": "PythonFile",  # Marker to identify this as a file reference
+        }
+
+    def resolve_content(self, base_path: Optional[str] = None) -> Dict[str, str]:
+        """Resolve the file content for inclusion in code_files.
+
+        Args:
+            base_path: Base directory path to resolve relative paths
+
+        Returns:
+            Dict with filename and content
+        """
+        if base_path is None:
+            base_path = os.getcwd()
+
+        file_path = Path(base_path) / self.filename
+
+        try:
+            with open(file_path, "r", encoding="utf-8") as f:
+                content = f.read()
+            # Return just the base filename, not the full path
+            base_filename = Path(self.filename).name
+            return {"filename": base_filename, "content": content}
+        except FileNotFoundError:
+            raise FileNotFoundError(f"Python file not found: {file_path}")
+        except Exception as e:
+            raise RuntimeError(f"Error reading Python file {file_path}: {e}")
+
+
+class StepMetadata(BaseModel):
+    """Metadata for workflow steps, containing configuration and execution parameters."""
+
+    model_config = {"arbitrary_types_allowed": True}
+
+    key: Optional[str] = None
+    depends_on: Union[List[Union[Any, str]], None] = (
+        None  # Can be Step objects or strings
+    )
+    execution_mode: Optional[Union[Any, Dict[str, Any]]] = (
+        None  # Can be ExecutionMode object or dict
+    )
+    output_behavior: Optional[Dict[str, Any]] = None
+    output_channels: List[str] = Field(default_factory=list)
+    output_content_type: OutputContentType = OutputContentType.TEXT
+    history_content_type: Optional[str] = None
+    ui_content_type: Optional[str] = None
+    user_output_visibility: OutputVisibility = OutputVisibility.VISIBLE
+    bot_output_visibility: OutputVisibility = OutputVisibility.HIDDEN
+    running_message: Optional[str] = None
+    finished_message: Optional[str] = None
+    parameter_hydration_behaviour: ParameterHydrationBehaviour = (
+        ParameterHydrationBehaviour.NONE
+    )
+
+
+class Step(StepMetadata):
+    """A single step in an agent workflow."""
+
+    agent: Optional[Any] = None  # Reference to the agent this step belongs to
+    result_handler: Optional[Any] = (
+        None  # Reference to the result handler this step belongs to
+    )
+    action: Union[Any, Dict[str, Any]]  # Function call like codeexec.execute(...)
+    parameters: Dict[str, Any] = Field(default_factory=dict)
+    result_handlers: List["ResultHandler"] = Field(default_factory=list)
+
+    def __init__(self, action: Any = None, step: Optional["Step"] = None, **data: Any):
+        # Validate that only one of action or step is provided
+        if action is not None and step is not None:
+            raise ValueError("Cannot specify both 'action' and 'step' parameters")
+
+        if action is None and step is None:
+            raise ValueError("Must specify either 'action' or 'step' parameter")
+
+        if step is not None:
+            # Copy all fields from the provided step
+            # Type check is redundant due to type annotation, but kept for runtime safety
+            if not hasattr(step, "action"):
+                raise ValueError("'step' parameter must be a Step object")
+
+            # Copy fields directly to preserve object types (especially action)
+            # Don't use model_dump() as it serializes objects to dicts
+            data.update(
+                {
+                    "action": step.action,
+                    "key": step.key,
+                    "parameters": step.parameters,
+                    "depends_on": step.depends_on,
+                    "execution_mode": step.execution_mode,
+                    "output_behavior": step.output_behavior,
+                    "result_handlers": step.result_handlers,
+                    "output_channels": step.output_channels,
+                    "output_content_type": step.output_content_type,
+                    "history_content_type": step.history_content_type,
+                    "ui_content_type": step.ui_content_type,
+                    "user_output_visibility": step.user_output_visibility,
+                    "bot_output_visibility": step.bot_output_visibility,
+                    "running_message": step.running_message,
+                    "finished_message": step.finished_message,
+                    "parameter_hydration_behaviour": step.parameter_hydration_behaviour,
+                }
+            )
+        else:
+            # Standard action-based step
+            data["action"] = action
+
+        super().__init__(**data)
+        # For enhanced syntax support - use private field not tracked by pydantic
+        self._decorator_handlers: List[Tuple[Any, Callable[..., Any]]] = []
+
+        # Automatically register with agent if provided and this is not a nested step
+        if self.agent is not None and self.result_handler is None:
+            self.agent.add_step(self)
+
+    def extract_action_parameters(self) -> Dict[str, Any]:
+        """Extract parameters from the action object safely."""
+        if not self.action:
+            return {}
+
+        # Handle nested Step objects (for result handlers)
+        if isinstance(self.action, Step):
+            return self.action.extract_action_parameters()
+
+        # If action is a Pydantic model, get its dict representation
+        if hasattr(self.action, "model_dump") and callable(
+            getattr(self.action, "model_dump")
+        ):
+            try:
+                action_obj = cast(ActionProtocol, self.action)
+                params = action_obj.model_dump()
+                # Remove the redundant 'name' field since it's already known from action type
+                params.pop("name", None)
+                # Map Python SDK field names to backend field names
+                if "json_data" in params:
+                    params["json"] = params.pop("json_data")
+                # Remove None values for optional parameters (matches action function behavior)
+                params = {k: v for k, v in params.items() if v is not None}
+                # Sort parameters for deterministic output
+                return dict(sorted(params.items()))
+            except Exception:
+                return {}
+        elif hasattr(self.action, "dict") and callable(getattr(self.action, "dict")):
+            try:
+                # Legacy pydantic v1 support
+                action_dict_method = getattr(self.action, "dict")
+                params = action_dict_method()
+                # Remove the redundant 'name' field since it's already known from action type
+                params.pop("name", None)
+                # Map Python SDK field names to backend field names
+                if "json_data" in params:
+                    params["json"] = params.pop("json_data")
+                # Remove None values for optional parameters (matches action function behavior)
+                params = {k: v for k, v in params.items() if v is not None}
+                # Sort parameters for deterministic output
+                return dict(sorted(params.items()))
+            except Exception:
+                return {}
+        elif isinstance(self.action, dict):
+            # Map Python SDK field names to backend field names
+            dict_params: Dict[str, Any] = dict(self.action)
+            # Remove the redundant 'name' field since it's already known from action type
+            dict_params.pop("name", None)
+            if "json_data" in dict_params:
+                dict_params["json"] = dict_params.pop("json_data")
+            # Sort parameters for deterministic output
+            return dict(sorted(dict_params.items()))
+        else:
+            return {}
+
+    def get_action_type(self) -> str:
+        """Get the action type string safely."""
+        if not self.action:
+            raise ValueError("Action is not set")
+
+        # Handle nested Step objects (for result handlers)
+        if isinstance(self.action, Step):
+            return self.action.get_action_type()
+
+        # Action objects should have a name attribute with the action type
+        if hasattr(self.action, "name"):
+            name_attr = getattr(self.action, "name")
+            if isinstance(name_attr, str):
+                return name_attr
+            elif hasattr(name_attr, "__str__"):
+                return str(name_attr)
+
+        # Fallback for unexpected cases
+        return str(type(self.action).__name__).lower()
+
+    def get_depends_on_keys(self) -> Optional[List[str]]:
+        """Get dependency keys safely without circular references, preserving None vs [] distinction."""
+        # CRITICAL: Preserve None vs [] distinction for 100% roundtrip parity
+        if self.depends_on is None:
+            return None  # Explicitly return None for null dependencies
+
+        if not self.depends_on:  # Empty list case
+            return []
+
+        # Depends_on is guaranteed to be a non-empty list at this point
+        depends_list = self.depends_on
+
+        result: List[str] = []
+        for dep in depends_list:
+            if isinstance(dep, str):
+                # Already a string identifier
+                result.append(dep)
+            elif hasattr(dep, "key") and getattr(dep, "key", None):
+                # Prefer the step key if available
+                key_val = getattr(dep, "key")
+                result.append(str(key_val) if key_val is not None else "")
+            elif hasattr(dep, "id") and getattr(dep, "id", None):
+                # Fall back to step ID if no key
+                id_val = getattr(dep, "id")
+                result.append(str(id_val) if id_val is not None else "")
+            else:
+                # Fail if we can't get a valid identifier
+                raise ValueError(f"Step dependency has no key or id: {dep}")
+        return result
+
+    @property
+    def output(self) -> "StepOutput":
+        """Get typed output reference for this step."""
+        return StepOutput(step_key=self.key or f"step_{id(self)}")
+
+    def when(
+        self, condition: Any
+    ) -> Callable[[Callable[..., Any]], Callable[..., Any]]:
+        """Decorator for adding result handlers with conditions.
+
+        Usage:
+        @step.when(IsSuccess() & GreaterThan("confidence", 0.8))
+        def handle_high_confidence(result):
+            return store_analysis(result)
+        """
+
+        def decorator(func: Callable[..., Any]) -> Callable[..., Any]:
+            # Store the handler function and condition for later processing
+            self._decorator_handlers.append((condition, func))
+            return func
+
+        return decorator
+
+    def on(
+        self,
+        condition: Any,
+        *actions: Any,
+        handler_type: HandlerType = HandlerType.INTERMEDIATE,
+    ) -> "Step":
+        """Add a result handler with condition and action(s).
+
+        Usage:
+        # Single action
+        step.on(IsSuccess(), utils.store_analysis(data=state.analyze_step.output))
+
+        # Multiple actions (variadic arguments)
+        step.on(And(IsError(), LessThan(number='r"{{coalesce "code_retry_loops?" 0}}"', value="2")),
+            send_status(status="retrying", message="Code execution failed, attempting to fix..."),
+            utils.echo(data={"code_retry_loops": 'r"{{incrementCounter "code_retry_loops"}}'}),
+            raise_action(status="go to step", message="code", parameters={...})
+        )
+        """
+        if not actions:
+            raise ValueError("Must specify at least one action")
+
+        # Create steps for each action, handling Step objects correctly
+        step_actions: List[Step] = []
+        for action in actions:
+            if isinstance(action, Step):
+                # If it's already a Step, use step= parameter
+                step_actions.append(Step(step=action))
+            else:
+                # If it's an action function, use action= parameter
+                step_actions.append(Step(action=action))
+
+        # Create a result handler and add it to this step
+        handler = ResultHandler(
+            type=handler_type,
+            if_conditions=condition,
+            steps=step_actions,
+        )
+        self.result_handlers.append(handler)
+        return self
+
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dict format expected by backend."""
+        # Process decorator handlers before serialization
+        self._process_decorator_handlers()
+
+        # Use model_dump but exclude problematic circular reference fields
+        result = self.model_dump(
+            exclude={
+                "agent",
+                "result_handler",
+                "_decorator_handlers",
+                "action",
+                "depends_on",
+            },
+        )
+
+        # Filter out empty/None optional fields to match export behavior
+        if result.get("ui_content_type") in [None, ""]:
+            result.pop("ui_content_type", None)
+        if result.get("history_content_type") in [None, ""]:
+            result.pop("history_content_type", None)
+
+        # Add action information without circular references
+        result["action_type"] = self.get_action_type()
+        result["parameters"] = self.extract_action_parameters()
+
+        # Add depends_on only if there are actual dependencies
+        if self.depends_on:
+            # Convert Step objects to their keys
+            deps = []
+            for dep in self.depends_on:
+                if isinstance(dep, Step):
+                    # If it's a Step object, use its key
+                    if dep.key:
+                        deps.append(dep.key)
+                elif isinstance(dep, str):
+                    # If it's already a string (step key), use it directly
+                    deps.append(dep)
+            result["depends_on"] = deps
+
+        # Always include result_handlers, even if they were excluded by exclude_unset
+        # This ensures that result handlers added via .on() method are included
+        result["result_handlers"] = []
+        if self.result_handlers:
+            converted_handlers: List[Any] = []
+            for handler in self.result_handlers:
+                if hasattr(handler, "to_dict"):
+                    converted_handlers.append(handler.to_dict())
+                elif isinstance(handler, dict):
+                    # Already a dict, convert conditions if needed
+                    handler_dict = cast(Dict[str, Any], handler)
+                    if handler_dict.get("if_conditions") and hasattr(
+                        handler_dict["if_conditions"], "to_dict"
+                    ):
+                        handler_dict["if_conditions"] = handler_dict[
+                            "if_conditions"
+                        ].to_dict()
+                    converted_handlers.append(handler_dict)
+                else:
+                    converted_handlers.append(handler)
+            result["result_handlers"] = converted_handlers
+
+        # Convert ExecutionMode objects to dictionaries
+        if self.execution_mode is not None:
+            if hasattr(self.execution_mode, "to_dict") and callable(
+                getattr(self.execution_mode, "to_dict")
+            ):
+                exec_mode_obj = cast(ExecutionModeProtocol, self.execution_mode)
+                result["execution_mode"] = exec_mode_obj.to_dict()
+            elif isinstance(self.execution_mode, dict):
+                result["execution_mode"] = self.execution_mode
+            else:
+                # Handle string mode types or other formats
+                result["execution_mode"] = self.execution_mode
+
+        # Recursively convert any other condition objects to dictionaries
+        def convert_conditions(obj: Any) -> Any:
+            """Recursively convert condition objects to dictionaries."""
+
+            if obj is None or isinstance(obj, (str, int, float, bool)):
+                return obj
+            elif isinstance(obj, Enum):
+                # Convert enum values to their string values for JSON serialization
+                return obj.value
+            elif hasattr(obj, "to_dict"):
+                return obj.to_dict()
+            elif isinstance(obj, list):
+                return [convert_conditions(item) for item in obj]
+            elif isinstance(obj, dict):
+                # Handle enum keys in dictionaries
+                converted_dict: Dict[str, Any] = {}
+                obj_dict = cast(Dict[Any, Any], obj)
+                for key, value in obj_dict.items():
+                    # Convert enum keys to strings
+                    str_key: str
+                    if isinstance(key, Enum):
+                        str_key = str(key.value)
+                    else:
+                        str_key = str(key)
+                    converted_dict[str_key] = convert_conditions(value)
+                return converted_dict
+            else:
+                return obj
+
+        # Apply condition conversion to all fields
+        for key, value in result.items():
+            result[key] = convert_conditions(value)
+
+        # Recursively sort all dictionaries for deterministic serialization
+        def sort_dict_recursively(obj: Any) -> Any:
+            """Recursively sort all dictionaries to ensure deterministic JSON output."""
+            if isinstance(obj, dict):
+                # Convert keys to strings for sorting to handle mixed types (enums + strings)
+                def sort_key(item: Tuple[Any, Any]) -> str:
+                    k, v = item
+                    if hasattr(k, "value"):  # Enum
+                        return str(k.value)
+                    return str(k)
+
+                obj_dict = cast(Dict[Any, Any], obj)
+                return dict(
+                    sorted(
+                        ((k, sort_dict_recursively(v)) for k, v in obj_dict.items()),
+                        key=sort_key,
+                    )
+                )
+            elif isinstance(obj, list):
+                return [sort_dict_recursively(item) for item in obj]
+            else:
+                return obj
+
+        result = sort_dict_recursively(result)
+        return cast(Dict[str, Any], result)
+
+    def _process_decorator_handlers(self) -> None:
+        """Process decorator handlers and convert them to ResultHandler objects."""
+        if not hasattr(self, "_decorator_handlers") or not self._decorator_handlers:
+            return
+
+        # Get current result_handlers list or create new one
+        current_handlers = list(self.result_handlers) if self.result_handlers else []
+
+        for condition, _ in self._decorator_handlers:
+            # Create a simple handler that calls the function
+            # For now, we'll create a basic handler structure
+            # In a full implementation, you'd want to analyze the function and create appropriate steps
+            handler = ResultHandler(
+                type=HandlerType.FINAL,
+                if_conditions=condition,
+                output_content_type=OutputContentType.TEXT,
+                steps=[],  # Would need to convert function to steps
+            )
+            current_handlers.append(handler)
+
+        # Update the result_handlers field
+        self.result_handlers = current_handlers
+
+        # Clear processed handlers
+        self._decorator_handlers.clear()
+
+
+class StepOutput(BaseModel):
+    """Represents the output of a step for type-safe access."""
+
+    step_key: str
+
+    def __init__(self, **data: Any):
+        super().__init__(**data)
+        # Use object attribute instead of Pydantic field
+        object.__setattr__(self, "_expected_fields", {})
+
+    def __getitem__(self, key: str) -> str:
+        """Allow bracket notation for accessing output fields."""
+        # Track that this field is being accessed for validation
+        if hasattr(self, "_expected_fields"):
+            expected_fields = getattr(self, "_expected_fields", {})
+            expected_fields[key] = (
+                None  # Will be populated with actual type during execution
+            )
+        return f"{{{{{self.step_key}.{key}}}}}"
+
+    def __getattr__(self, name: str) -> str:
+        """Allow dot notation for accessing output fields."""
+        # Avoid infinite recursion for Pydantic internal attributes
+        if name.startswith("_") or name in {"step_key", "model_fields", "model_config"}:
+            raise AttributeError(
+                f"'{type(self).__name__}' object has no attribute '{name}'"
+            )
+
+        # Track that this field is being accessed for validation
+        if hasattr(self, "_expected_fields"):
+            expected_fields = getattr(self, "_expected_fields", {})
+            expected_fields[name] = (
+                None  # Will be populated with actual type during execution
+            )
+        return f"{{{{{self.step_key}.{name}}}}}"
+
+    def get_expected_fields(self) -> Dict[str, Any]:
+        """Get the fields that have been accessed on this step output."""
+        return getattr(self, "_expected_fields", {})
+
+    def validate_field_access(self, available_fields: Dict[str, Any]) -> List[str]:
+        """Validate that all accessed fields are available in the step result."""
+        errors = []
+        expected = self.get_expected_fields()
+
+        for field_name in expected.keys():
+            if field_name not in available_fields:
+                errors.append(
+                    f"Step '{self.step_key}' output field '{field_name}' is not available"
+                )
+
+        return errors
+
+
+class ResultHandler(BaseModel):
+    """Result handler definition that matches the Go backend structure."""
+
+    type: HandlerType = HandlerType.FINAL
+    if_conditions: Optional[Any] = None  # Can be condition objects or None
+    output_content_type: OutputContentType = OutputContentType.TEXT
+    history_content_type: Optional[str] = None
+    ui_content_type: Optional[str] = None
+    steps: List[Step] = Field(default_factory=list)
+
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dict format expected by backend."""
+        # Convert Step objects in steps list to dictionaries BEFORE model_dump()
+        # because model_dump() will convert Step objects to dicts but won't call to_dict()
+        converted_steps = []
+        if self.steps:
+            for step in self.steps:
+                if hasattr(step, "to_dict"):
+                    step_dict = step.to_dict()
+                    # DO NOT auto-generate keys for result handler steps
+                    # Only include keys if explicitly set to preserve roundtrip parity
+
+                    # Ensure bot_output_visibility is preserved (defaults to hidden for result handler steps)
+                    if "bot_output_visibility" not in step_dict:
+                        step_dict["bot_output_visibility"] = "hidden"
+
+                    converted_steps.append(step_dict)
+                else:
+                    # Convert step to dict if it doesn't have to_dict method
+                    if isinstance(step, dict):
+                        converted_steps.append(step)
+                    else:
+                        # Fallback: convert to dict using model_dump if available
+                        if hasattr(step, "model_dump"):
+                            converted_steps.append(step.model_dump())
+                        else:
+                            converted_steps.append({})
+
+        # Use model_dump but exclude steps, then add our properly converted steps
+        result = self.model_dump(exclude={"steps"})
+        result["steps"] = converted_steps
+
+        # Filter out empty/None optional fields to match export behavior
+        if result.get("ui_content_type") in [None, ""]:
+            result.pop("ui_content_type", None)
+        if result.get("history_content_type") in [None, ""]:
+            result.pop("history_content_type", None)
+
+        # Convert condition objects to dictionaries if needed
+        if self.if_conditions is not None and hasattr(self.if_conditions, "to_dict"):
+            result["if_conditions"] = self.if_conditions.to_dict()
+
+        # Convert enum values to their string values (both keys and values)
+        converted_result = {}
+        for key, value in result.items():
+            # Convert enum keys to strings
+            if isinstance(key, Enum):
+                key = key.value
+            # Convert enum values to strings
+            if isinstance(value, Enum):
+                value = value.value
+            converted_result[key] = value
+        result = converted_result
+
+        return result
+
+
+class SecretsDict(Dict[str, Any]):
+    """A dictionary wrapper that provides .get() method for secrets access."""
+
+    def __init__(self, secrets_data: Optional[Dict[str, Any]] = None):
+        super().__init__(secrets_data or {})
+
+    def get(self, key: str, default: Any = None) -> Any:
+        """Get decrypted secrets for a specific resource/service key."""
+        return super().get(key, default)
+
+
+class ParametersDict(Dict[str, Any]):
+    """A dictionary wrapper that provides .get() method for parameters access."""
+
+    def __init__(self, parameters_data: Optional[Dict[str, Any]] = None):
+        super().__init__(parameters_data or {})
+
+    def get(self, key: str, default: Any = None) -> Any:
+        """Get a step parameter with optional default."""
+        return super().get(key, default)
+
+
+class StepContext(BaseModel):
+    """Context available to step functions with type-safe access
+
+    Provides access to:
+    - User query and parameters
+    - Previous step results (state)
+    - Available resources (datasets, APIs)
+    - Encrypted secrets
+    - System information
+    """
+
+    # Core user input
+    query: Optional[str] = None
+
+    # Previous step results and state
+    state: Dict[str, Any] = Field(default_factory=dict)
+    steps: Dict[str, Any] = Field(default_factory=dict)  # Alias for state.steps
+
+    # Resources and data access
+    resources: List[Dict[str, Any]] = Field(default_factory=list)
+    resource_definitions: Optional[Dict[str, Any]] = None
+
+    # Security and credentials - raw data
+    secrets_data: Dict[str, Any] = Field(default_factory=dict)
+
+    # Step-specific parameters - raw data
+    parameters_data: Dict[str, Any] = Field(default_factory=dict)
+
+    # System context
+    system: Dict[str, Any] = Field(default_factory=dict)
+
+    # Raw context for advanced users
+    raw: Dict[str, Any] = Field(default_factory=dict)
+
+    def __init__(self, **data: Any):
+        # Extract secrets and parameters from input data
+        secrets_data = data.pop("secrets", {})
+        parameters_data = data.pop("parameters", {})
+
+        # Set the internal data fields
+        data["secrets_data"] = secrets_data
+        data["parameters_data"] = parameters_data
+
+        super().__init__(**data)
+
+        # Create wrapper objects for secrets and parameters
+        self._secrets_wrapper = SecretsDict(self.secrets_data)
+        self._parameters_wrapper = ParametersDict(self.parameters_data)
+
+    @property
+    def secrets(self) -> SecretsDict:
+        """Get secrets wrapper that supports .get() method."""
+        return self._secrets_wrapper
+
+    @property
+    def parameters(self) -> ParametersDict:
+        """Get parameters wrapper that supports .get() method."""
+        return self._parameters_wrapper
+
+    def __getitem__(self, key: str) -> Any:
+        """Allow bracket notation access to state"""
+        try:
+            state_value = super().__getattribute__("state")
+            return state_value.get(key) if isinstance(state_value, dict) else None
+        except AttributeError:
+            return None
+
+    def __getattr__(self, name: str) -> Any:
+        """Allow dot notation access to state"""
+        # Avoid infinite recursion for Pydantic internal attributes
+        if name.startswith("_") or name in {
+            "state",
+            "steps",
+            "resources",
+            "secrets",
+            "parameters",
+            "system",
+            "raw",
+            "query",
+            "resource_definitions",
+        }:
+            raise AttributeError(
+                f"'{self.__class__.__name__}' object has no attribute '{name}'"
+            )
+
+        try:
+            state_value = super().__getattribute__("state")
+            if isinstance(state_value, dict) and name in state_value:
+                return state_value[name]
+        except AttributeError:
+            pass
+        raise AttributeError(
+            f"'{self.__class__.__name__}' object has no attribute '{name}'"
+        )
+
+    def get_resource(self, key: str) -> Optional[Dict[str, Any]]:
+        """Get a specific resource by key"""
+        for resource in self.resources:
+            if resource.get("key") == key:
+                return resource
+        return None
+
+    def get_secret(self, key: str) -> Optional[Any]:
+        """Get decrypted secrets for a specific resource/service"""
+        return self.secrets.get(key)
+
+    def get_parameter(self, key: str, default: Any = None) -> Any:
+        """Get a step parameter with optional default"""
+        return self.parameters.get(key, default)
+
+
+class StepResult(BaseModel):
+    """Result from executing a workflow step."""
+
+    success: bool
+    data: Any = None
+    error: Optional[str] = None
+    step_name: str
+
+
+class Agent(BaseModel):
+    """Main agent class for defining AI workflows."""
+
+    name: str
+    description: Optional[str] = None
+    persona: Optional[str] = None
+    visibility: str = "private"
+    running_message: Optional[str] = None
+    finished_message: Optional[str] = None
+    version: str = "1.0"
+    timeout: Optional[int] = None
+    retry_attempts: int = 0
+    tags: List[str] = Field(default_factory=list)
+    steps: List[Step] = Field(default_factory=list)
+    parameter_definitions: List["ParameterDefinition"] = Field(default_factory=list)
+
+    def step(
+        self,
+        action: Any,
+        key: Optional[str] = None,
+        depends_on: Optional[Union[Step, List[Step], str, List[str]]] = None,
+        **kwargs: Any,
+    ) -> Step:
+        """Create a step with cleaner syntax and better defaults.
+
+        Args:
+            action: The action to execute (e.g., codeexec.execute(...))
+            key: Optional step key, auto-generated if not provided
+            depends_on: Step(s) or key(s) this step depends on
+            **kwargs: Additional step configuration
+
+        Returns:
+            Step: The created step with sensible defaults
+        """
+        # Auto-generate key if not provided
+        if key is None:
+            key = f"step_{len(self.steps) + 1}"
+
+        # Set sensible defaults
+        step_config = {
+            "agent": self,
+            "key": key,
+            "action": action,
+            "depends_on": (
+                [depends_on]
+                if depends_on and not isinstance(depends_on, list)
+                else depends_on
+            ),
+            "user_output_visibility": OutputVisibility.VISIBLE,
+            "bot_output_visibility": OutputVisibility.HIDDEN,
+            "output_content_type": OutputContentType.TEXT,
+            "parameter_hydration_behaviour": ParameterHydrationBehaviour.NONE,
+        }
+
+        # Override with any provided kwargs
+        step_config.update(kwargs)
+
+        step = Step(**step_config)
+        return step
+
+    def exec(
+        self,
+        step_metadata: Optional[StepMetadata] = None,
+        **codeexec_params: Any,
+    ) -> Callable[..., Step]:
+        """Create a codeexec step using decorator syntax.
+
+        This method is designed to be used as a decorator for functions that implement
+        code execution logic. It creates a codeexec.execute action with the provided
+        parameters and step metadata.
+
+        Args:
+            step_metadata: Step configuration (key, depends_on, etc.)
+            **codeexec_params: Parameters for codeexec.execute (entrypoint, parameters, resources, etc.)
+
+        Returns:
+            Callable: Decorator function that creates and returns a Step
+        """
+
+        def decorator(func: Callable[..., Any]) -> Step:
+            # Import here to avoid circular imports
+            from .actions import codeexec
+
+            # Create the codeexec.execute action with the provided parameters
+            action = codeexec.execute(**codeexec_params)
+
+            # Extract step metadata or use defaults
+            if step_metadata:
+                step_config = {
+                    "agent": self,
+                    "action": action,
+                    "key": step_metadata.key or func.__name__,
+                    "depends_on": step_metadata.depends_on,
+                    "execution_mode": step_metadata.execution_mode,
+                    "output_behavior": step_metadata.output_behavior,
+                    "output_channels": step_metadata.output_channels,
+                    "output_content_type": step_metadata.output_content_type,
+                    "history_content_type": step_metadata.history_content_type,
+                    "ui_content_type": step_metadata.ui_content_type,
+                    "user_output_visibility": step_metadata.user_output_visibility,
+                    "bot_output_visibility": step_metadata.bot_output_visibility,
+                    "running_message": step_metadata.running_message,
+                    "finished_message": step_metadata.finished_message,
+                    "parameter_hydration_behaviour": step_metadata.parameter_hydration_behaviour,
+                }
+            else:
+                step_config = {
+                    "agent": self,
+                    "action": action,
+                    "key": func.__name__,
+                    "user_output_visibility": OutputVisibility.VISIBLE,
+                    "bot_output_visibility": OutputVisibility.HIDDEN,
+                    "output_content_type": OutputContentType.TEXT,
+                    "parameter_hydration_behaviour": ParameterHydrationBehaviour.NONE,
+                }
+
+            # Remove None values and extract action separately
+            filtered_config = {
+                k: v for k, v in step_config.items() if v is not None and k != "action"
+            }
+
+            # Create step with the codeexec action, not the function
+            step = Step(action=action, step=None, **filtered_config)
+
+            # Store the function name on the step for runtime extraction
+            if hasattr(step, "__dict__"):
+                step.__dict__["__name__"] = func.__name__
+
+            return step
+
+        return decorator
+
+    def add_step(self, step: Step) -> None:
+        """Add a step to this agent if it's not already present."""
+        # Only add if the step is not already in the list
+        if step in self.steps:
+            return
+
+        # Create a new list with the existing steps plus the new one
+        current_steps = list(self.steps) if self.steps else []
+        current_steps.append(step)
+        self.steps = current_steps
+
+    def get_step(self, key: str) -> Optional[Step]:
+        """Get a step by its key."""
+        for step in self.steps:
+            if step.key == key:
+                return step
+        return None
+
+    def to_json(self) -> str:
+        """Export to JSON format expected by Go backend."""
+        export_data = {
+            "bot": {
+                "Name": self.name,
+                "Description": self.description or "",
+                "Visibility": self.visibility,
+                "Persona": self.persona,  # Export as string or null, not object
+                "RunningMessage": self.running_message,
+                "FinishedMessage": self.finished_message,
+                "Source": "python",
+            },
+            "parameter_definitions": self.parameter_definitions,
+            "steps": [step.to_dict() for step in self.steps],
+        }
+
+        import json
+
+        return json.dumps(export_data, indent=2)
+
+
+class ExecutionMode(BaseModel):
+    """Execution mode matching Go backend structure."""
+
+    mode: ExecutionModeType = ExecutionModeType.ALL
+    data: Optional[Any] = None
+    if_condition: Optional[Any] = None
+
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dict format expected by backend."""
+        result: Dict[str, Any] = {"mode": self.mode}
+        if self.data is not None:
+            result["data"] = self.data
+        # Always include if_condition to maintain parity with backend structure
+        result["if_condition"] = self.if_condition
+        return result
+
+
+class Prompt(BaseModel):
+    """A prompt that can be loaded from a .prompt file or defined inline.
+
+    This class provides a clean way to manage prompts in agent code,
+    supporting both inline strings and external .prompt files.
+    """
+
+    content: str = Field(..., description="The prompt content")
+    filename: Optional[str] = Field(
+        None, description="Source filename if loaded from file"
+    )
+
+    def __init__(
+        self, content: Optional[str] = None, filename: Optional[str] = None, **data: Any
+    ):
+        """Initialize a Prompt.
+
+        Args:
+            content: Direct prompt content
+            filename: Path to .prompt file to load
+            **data: Additional model data
+        """
+        if content is not None and filename is not None:
+            raise ValueError("Cannot specify both 'content' and 'filename' parameters")
+
+        if content is None and filename is None:
+            raise ValueError("Must specify either 'content' or 'filename' parameter")
+
+        if filename is not None:
+            # Load content from file
+            content = self._load_from_file(filename)
+            data.update({"content": content, "filename": filename})
+        else:
+            data.update({"content": content})
+
+        super().__init__(**data)
+
+    @classmethod
+    def from_file(cls, filename: str, base_path: Optional[str] = None) -> "Prompt":
+        """Load a prompt from a .prompt file.
+
+        Args:
+            filename: Path to the .prompt file
+            base_path: Base directory to resolve relative paths (defaults to cwd)
+
+        Returns:
+            Prompt: Loaded prompt instance
+        """
+        if base_path is None:
+            base_path = os.getcwd()
+
+        file_path = Path(base_path) / filename
+
+        try:
+            with open(file_path, "r", encoding="utf-8") as f:
+                content = f.read().strip()
+            return cls(content=content)
+        except FileNotFoundError:
+            raise FileNotFoundError(f"Prompt file not found: {file_path}")
+        except Exception as e:
+            raise RuntimeError(f"Error reading prompt file {file_path}: {e}")
+
+    def _load_from_file(self, filename: str) -> str:
+        """Load content from a .prompt file."""
+        return self.from_file(filename).content
+
+    def __str__(self) -> str:
+        """Return the prompt content when used as a string."""
+        return self.content
+
+    def __call__(self) -> str:
+        """Allow the prompt to be called like a function, returning content."""
+        return self.content
+
+    def __repr__(self) -> str:
+        """Return a helpful representation."""
+        if self.filename:
+            return f"Prompt(filename='{self.filename}')"
+        else:
+            preview = (
+                self.content[:50] + "..." if len(self.content) > 50 else self.content
+            )
+            return f"Prompt(content='{preview}')"
+
+    @classmethod
+    def load_from_directory(
+        cls, directory: str, base_path: Optional[str] = None
+    ) -> Dict[str, "Prompt"]:
+        """Load all .prompt files from a directory as Prompt objects.
+
+        Args:
+            directory: Directory containing .prompt files
+            base_path: Base directory to resolve relative paths (defaults to smart detection)
+
+        Returns:
+            Dict[str, Prompt]: Mapping of filename (without extension) to Prompt objects
+        """
+        if base_path is None:
+            # Use stack inspection to determine the calling file's directory
+            # This allows agents to load prompts from their own directory
+            # even when executed from a different working directory
+            import inspect
+
+            frame = inspect.currentframe()
+            try:
+                # Get the caller's frame (skip current frame)
+                if frame and frame.f_back:
+                    caller_frame = frame.f_back
+                    caller_file = caller_frame.f_code.co_filename
+                    caller_dir = Path(caller_file).parent
+                    # Check if the caller is in an agent directory structure
+                    # (i.e., the calling file is agent.py in a directory under erdo-agents)
+                    if caller_file.endswith("agent.py") and "erdo-agents" in str(
+                        caller_dir
+                    ):
+                        # The caller is an agent.py file, use its directory
+                        base_path = str(caller_dir)
+                    else:
+                        # Fall back to current working directory
+                        base_path = os.getcwd()
+                else:
+                    # Fall back to current working directory
+                    base_path = os.getcwd()
+            finally:
+                del frame  # Avoid reference cycles
+
+        dir_path = Path(base_path) / directory
+
+        if not dir_path.exists() or not dir_path.is_dir():
+            # Provide a more helpful error message
+            cwd = os.getcwd()
+            caller_path = None
+            import inspect
+
+            frame = inspect.currentframe()
+            try:
+                if frame and frame.f_back:
+                    caller_file = frame.f_back.f_code.co_filename
+                    caller_path = str(Path(caller_file).parent)
+            finally:
+                del frame
+
+            error_msg = f"Prompts directory not found: {dir_path}"
+            if caller_path and caller_path != str(Path(base_path)):
+                error_msg += f"\nCalling file: {caller_path}"
+            error_msg += f"\nCurrent working directory: {cwd}"
+            error_msg += f"\nBase path used: {base_path}"
+            raise FileNotFoundError(error_msg)
+
+        prompts = {}
+        for prompt_file in dir_path.glob("*.prompt"):
+            prompt_name = prompt_file.stem  # filename without extension
+            relative_path = str(prompt_file.relative_to(base_path))
+            prompts[prompt_name] = cls.from_file(relative_path, base_path)
+
+        return prompts
+
+
+# Re-export commonly used types for convenience
+__all__ = [
+    "Agent",
+    "Step",
+    "StepMetadata",
+    "StepContext",
+    "StepResult",
+    "StepOutput",
+    "ResultHandler",
+    "ExecutionMode",
+    "Tool",
+    "PythonFile",
+    "Prompt",
+    "TemplateString",
+    # Complex types
+    "ExecutionCondition",
+]