feat: simplify state.history with to_memory call in micro-agent. Or the call to LLM may exceed the token limit. (All-Hands-AI#1806)

* feat: simplify state.history with to_memory call in micro-agent.

* feat: merge master and replace to_memory with event_to_memory.

---------

Co-authored-by: aaren.xzh <[email protected]>

Author: assertion

agenthub/micro/agent.py

@@ -5,6 +5,7 @@
 from opendevin.core.utils import json
 from opendevin.events.action import Action
 from opendevin.events.serialization.action import action_from_dict
+from opendevin.events.serialization.event import event_to_memory
 from opendevin.llm.llm import LLM
 
 from .instructions import instructions
@@ -26,6 +27,18 @@ def to_json(obj, **kwargs):
     return json.dumps(obj, **kwargs)
 
 
+def history_to_json(obj, **kwargs):
+    """
+    Serialize and simplify history to str format
+    """
+    if isinstance(obj, list):
+        # process history, make it simpler.
+        processed_history = []
+        for action, observation in obj:
+            processed_history.append((event_to_memory(action), event_to_memory(observation)))
+        return json.dumps(processed_history, **kwargs)
+
+
 class MicroAgent(Agent):
     prompt = ''
     agent_definition: dict = {}
@@ -44,6 +57,7 @@ def step(self, state: State) -> Action:
             state=state,
             instructions=instructions,
             to_json=to_json,
+            history_to_json=history_to_json,
             delegates=self.delegates,
             latest_user_message=latest_user_message,
         )

agenthub/micro/coder/prompt.md

@@ -21,7 +21,7 @@ Do NOT finish until you have completed the tasks.
 
 ## History
 {{ instructions.history_truncated }}
-{{ to_json(state.history[-10:]) }}
+{{ history_to_json(state.history[-10:]) }}
 
 ## Format
 {{ instructions.format.action }}

agenthub/micro/commit_writer/prompt.md

@@ -18,7 +18,7 @@ the `reject` action with `outputs.answer` set to the reason.
 
 ## History
 {{ instructions.history_truncated }}
-{{ to_json(state.history[-10:]) }}
+{{ history_to_json(state.history[-10:]) }}
 
 If the last item in the history is an error, you should try to fix it.
 

agenthub/micro/manager/prompt.md

@@ -17,7 +17,7 @@ provide the correct inputs for the delegate you select.
 
 ## History
 {{ instructions.history_truncated }}
-{{ to_json(state.history[-10:]) }}
+{{ history_to_json(state.history[-10:]) }}
 
 ## Available Actions
 {{ instructions.actions.delegate }}

agenthub/micro/math_agent/prompt.md

@@ -10,7 +10,7 @@ and call the `finish` action with `outputs.answer` set to the answer.
 
 ## History
 {{ instructions.history_truncated }}
-{{ to_json(state.history[-10:]) }}
+{{ history_to_json(state.history[-10:]) }}
 
 If the last item in the history is an error, you should try to fix it.
 

agenthub/micro/postgres_agent/prompt.md

@@ -18,7 +18,7 @@ You may take any of the following actions:
 
 ## History
 {{ instructions.history_truncated }}
-{{ to_json(state.history[-10:]) }}
+{{ history_to_json(state.history[-10:]) }}
 
 ## Format
 {{ instructions.format.action }}

agenthub/micro/repo_explorer/prompt.md

@@ -20,7 +20,7 @@ When you're done, put your summary into the output of the `finish` action.
 
 ## History
 {{ instructions.history_truncated }}
-{{ to_json(state.history[-10:]) }}
+{{ history_to_json(state.history[-10:]) }}
 
 ## Format
 {{ instructions.format.action }}

agenthub/micro/study_repo_for_task/prompt.md

@@ -19,7 +19,7 @@ When you're done, put your summary in `outputs.summary` in the `finish` action.
 
 ## History
 {{ instructions.history_truncated }}
-{{ to_json(state.history[-10:]) }}
+{{ history_to_json(state.history[-10:]) }}
 
 ## Format
 {{ instructions.format.action }}

agenthub/micro/typo_fixer_agent/prompt.md

@@ -23,7 +23,7 @@ Do NOT finish until you have fixed all the typos and generated a summary.
 
 ## History
 {{ instructions.history_truncated }}
-{{ to_json(state.history[-5:]) }}
+{{ history_to_json(state.history[-5:]) }}
 
 ## Format
 {{ instructions.format.action }}

agenthub/micro/verifier/prompt.md

@@ -21,7 +21,7 @@ explaining what the problem is.
 
 ## History
 {{ instructions.history_truncated }}
-{{ to_json(state.history[-10:]) }}
+{{ history_to_json(state.history[-10:]) }}
 
 ## Format
 {{ instructions.format.action }}