Use keyword matching for CodeAct microagents

openhands/agenthub/codeact_agent/codeact_agent.py

@@ -210,7 +210,26 @@ def step(self, state: State) -> Action:
         return self.action_parser.parse(response)
 
     def _get_messages(self, state: State) -> list[Message]:
-        messages: list[Message] = []
+        messages: list[Message] = [
+            Message(
+                role='system',
+                content=[
+                    TextContent(
+                        text=self.prompt_manager.get_system_message(),
+                        cache_prompt=self.llm.is_caching_prompt_active(),  # Cache system prompt
+                    )
+                ],
+            ),
+            Message(
+                role='user',
+                content=[
+                    TextContent(
+                        text=self.prompt_manager.get_example_user_message(),
+                        cache_prompt=self.llm.is_caching_prompt_active(),  # if the user asks the same query,
+                    )
+                ],
+            ),
+        ]
 
         for event in state.history.get_events():
             # create a regular message from an event
@@ -221,6 +240,7 @@ def _get_messages(self, state: State) -> list[Message]:
             else:
                 raise ValueError(f'Unknown event type: {type(event)}')
 
+            # add regular message
             if message:
                 # handle error if the message is the SAME role as the previous message
                 # litellm.exceptions.BadRequestError: litellm.BadRequestError: OpenAIException - Error code: 400 - {'detail': 'Only supports u/a/u/a/u...'}
@@ -230,6 +250,18 @@ def _get_messages(self, state: State) -> list[Message]:
                 else:
                     messages.append(message)
 
+        # Add caching to the last 2 user messages
+        if self.llm.is_caching_prompt_active():
+            user_turns_processed = 0
+            for message in reversed(messages):
+                if message.role == 'user' and user_turns_processed < 2:
+                    message.content[
+                        -1
+                    ].cache_prompt = True  # Last item inside the message content
+                    user_turns_processed += 1
+
+        # The latest user message is important:
+        # we want to remind the agent of the environment constraints
         latest_user_message = next(
             islice(
                 (
@@ -242,43 +274,7 @@ def _get_messages(self, state: State) -> list[Message]:
             ),
             None,
         )
-        messages = (
-            [
-                Message(
-                    role='system',
-                    content=[
-                        TextContent(
-                            text=self.prompt_manager.get_system_message(),
-                            cache_prompt=self.llm.is_caching_prompt_active(),  # Cache system prompt
-                        )
-                    ],
-                ),
-                Message(
-                    role='user',
-                    content=[
-                        TextContent(
-                            text=self.prompt_manager.get_example_user_message(
-                                ''
-                                if latest_user_message is None
-                                else latest_user_message.content[-1].text
-                            ),
-                            cache_prompt=self.llm.is_caching_prompt_active(),  # if the user asks the same query,
-                        )
-                    ],
-                ),
-            ]
-            + messages
-        )
-
         if latest_user_message:
-            reminder_text = f'\n\nENVIRONMENT REMINDER: You have {state.max_iterations - state.iteration} turns left to complete the task. When finished reply with <finish></finish>.'
-            latest_user_message.content.append(TextContent(text=reminder_text))
-
-        if self.llm.is_caching_prompt_active():
-            user_turns_processed = 0
-            for message in reversed(messages):
-                if message.role == 'user' and user_turns_processed < 2:
-                    message.content[-1].cache_prompt = True
-                    user_turns_processed += 1
+            self.prompt_manager.enhance_message(latest_user_message, state)
 
         return messages

openhands/utils/prompt.py

@@ -2,6 +2,8 @@
 
 from jinja2 import Template
 
+from openhands.controller.state.state import State
+from openhands.core.message import Message, TextContent
 from openhands.utils.microagent import MicroAgent
 
 
@@ -53,7 +55,7 @@ def get_system_message(self) -> str:
         ).strip()
         return rendered
 
-    def get_example_user_message(self, latest_user_message: str) -> str:
+    def get_example_user_message(self) -> str:
         """This is the initial user message provided to the agent
         before *actual* user instructions are provided.
 
@@ -63,11 +65,23 @@ def get_example_user_message(self, latest_user_message: str) -> str:
         These additional context will convert the current generic agent
         into a more specialized agent that is tailored to the user's task.
         """
+        return self.user_template.render().strip()
+
+    def enhance_message(self, message: Message, state: State) -> None:
+        """Enhance the user message with additional context.
+
+        This method is used to enhance the user message with additional context
+        about the user's task. The additional context will convert the current
+        generic agent into a more specialized agent that is tailored to the user's task.
+        """
         micro_agent_prompts = []
         for micro_agent in self.microagents.values():
-            if micro_agent.should_trigger(latest_user_message):
+            if micro_agent.should_trigger(message):
                 micro_agent_prompts.append(micro_agent.content)
-        rendered = self.user_template.render(
-            micro_agents=micro_agent_prompts,
-        )
-        return rendered.strip()
+        if len(micro_agent_prompts) > 0:
+            micro_text = "EXTRA INFO: the following information has been included based on a keyword match. It may or may not be relevant to the user's request.\n\n"
+            for micro_agent_prompt in micro_agent_prompts:
+                micro_text += micro_agent_prompt + '\n\n'
+            message.content.append(TextContent(text=micro_text))
+        reminder_text = f'ENVIRONMENT REMINDER: You have {state.max_iterations - state.iteration} turns left to complete the task. When finished reply with <finish></finish>.'
+        message.content.append(TextContent(text=reminder_text))