Update Planner Interface

autogpt/core/planning/base.py

@@ -1,5 +1,108 @@
 import abc
+import enum
+import typing
+from dataclasses import dataclass
+
+if typing.TYPE_CHECKING:
+    from autogpt.config.ai_config import (  # Should this be replaced with Configuration?
+        AIConfig,
+    )
+    from autogpt.core.configuration.base import Configuration
+    from autogpt.core.logging.base import Logger
+    from autogpt.core.plugin.base import PluginManager
+    from autogpt.core.workspace.base import Workspace
+
+
+class LLMModel(enum.StrEnum):
+    # This will be defined probably in the LLM interface module
+    GPT3 = "gpt3"
+    GPT4 = "gpt4"
+
+
+class Role(enum.StrEnum):
+    USER = "user"
+    SYSTEM = "system"
+    ASSISTANT = "assistant"
+
+
+@dataclass
+class Message:
+    role: Role
+    message: str
+
+
+@dataclass
+class Context:  # This will be defined probably in the Memory interface module
+    progress: str
+    result: str
+    memories: list(str)
+    feedback: str
+
+
+@dataclass
+class Thoughts:
+    system: str
+    reasoning: str
+    plan: list(str)
+    thoughts: str
+    criticism: str
 
 
 class Planner(abc.ABC):
-    pass
+    """Build prompts based on inputs, can potentially store and retrieve planning state from the workspace"""
+
+    @abc.abstractmethod
+    def __init__(
+        self,
+        configuration: Configuration,
+        logger: Logger,
+        plugin_manager: PluginManager,
+        workspace: Workspace,
+    ) -> None:
+        self.configuration = configuration
+        self.logger = logger
+        self.plugin_manager = plugin_manager
+        self.workspace = workspace
+        pass
+
+    @abc.abstractmethod
+    def construct_objective_prompt_from_user_input(
+        self,
+        user_objective: str,
+    ) -> list[Message]:  #
+        """
+        This method is called upon the creation of an agent to refine its goals based on user input.
+
+        Args:
+            user_objective (str): The user-defined objective for the agent.
+
+        Returns:
+            List[Dict]: A list of message dictionaries that define the refined goals for the agent.
+        """
+        pass
+
+    @abc.abstractmethod
+    def get_self_feedback_prompt(
+        self,
+        context: Context,
+        thoughts: Thoughts,
+        llm_model: LLMModel,  # Should this be a model object?
+    ) -> str:
+        """
+        Generates a self-feedback prompt for the language model, based on the provided context and thoughts.
+
+        This method takes in a Context object containing information about the agent's progress, result,
+        memories, and feedback. It also takes in a Thoughts object containing keys such as 'reasoning',
+        'plan', 'thoughts', and 'criticism'. The method combines these elements to create a prompt that
+        facilitates self-assessment and improvement for the agent.
+
+        Args:
+            context (Context): An object containing information about the agent's progress, result,
+                               memories, and feedback.
+            thoughts (Thoughts): An object containing the agent's reasoning, plan, thoughts, and criticism.
+            llm_model (str): The identifier for the language model to be used. 
+
+        Returns:
+            str: A self-feedback prompt for the language model based on the given context and thoughts.
+        """
+        pass