feat: implement session-based output organization (#1032)

.gitignore

@@ -1,4 +1,8 @@
 ### Project-specific ###
+# Config with secrets
+config/config.toml
+config/config.daytona.toml
+
 # Logs
 logs/
 

app/agent/manus.py

@@ -1,3 +1,5 @@
+import os
+from datetime import datetime
 from typing import Dict, List, Optional
 
 from pydantic import Field, model_validator
@@ -24,6 +26,36 @@ class Manus(ToolCallAgent):
     system_prompt: str = SYSTEM_PROMPT.format(directory=config.workspace_root)
     next_step_prompt: str = NEXT_STEP_PROMPT
 
+    def get_system_prompt_with_session(self) -> str:
+        """Get system prompt with session folder information."""
+        base_prompt = self.system_prompt
+        if self.session_folder:
+            session_info = (
+                f"\n\nSESSION OUTPUT FOLDER: All research outputs should be saved to: {self.session_folder}\n"
+                f"Absolute paths for file operations:\n"
+                f"- Guides/documentation: {os.path.join(self.session_folder, 'guides')}/\n"
+                f"- Code files: {os.path.join(self.session_folder, 'code')}/\n"
+                f"- Data files: {os.path.join(self.session_folder, 'data')}/\n"
+            )
+            return base_prompt + session_info
+        return base_prompt
+
+    def get_session_path(self, subfolder: str = "") -> str:
+        """Get the full path for saving files in the session folder.
+
+        Args:
+            subfolder: One of 'guides', 'code', 'data', or empty string for session root
+
+        Returns:
+            Full path to the requested subfolder
+        """
+        if not self.session_folder:
+            return ""
+
+        if subfolder:
+            return os.path.join(self.session_folder, subfolder)
+        return self.session_folder
+
     max_observe: int = 10000
     max_steps: int = 20
 
@@ -50,6 +82,29 @@ class Manus(ToolCallAgent):
     )  # server_id -> url/command
     _initialized: bool = False
 
+    # Session output folder for organizing research outputs
+    session_folder: Optional[str] = None
+
+    @model_validator(mode="after")
+    def initialize_session_folder(self) -> "Manus":
+        """Initialize session output folder with unique timestamp."""
+        if not self.session_folder:
+            # Generate unique session folder with timestamp
+            timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+            slug = "session"  # Could be customized based on task type
+            session_path = os.path.join(config.workspace_root, "output", f"{slug}_{timestamp}")
+
+            # Create the session folder and subfolders
+            os.makedirs(session_path, exist_ok=True)
+            os.makedirs(os.path.join(session_path, "guides"), exist_ok=True)
+            os.makedirs(os.path.join(session_path, "code"), exist_ok=True)
+            os.makedirs(os.path.join(session_path, "data"), exist_ok=True)
+
+            self.session_folder = session_path
+            logger.info(f"📁 Created session output folder: {session_path}")
+
+        return self
+
     @model_validator(mode="after")
     def initialize_helper(self) -> "Manus":
         """Initialize basic components synchronously."""
@@ -143,7 +198,11 @@ async def think(self) -> bool:
             await self.initialize_mcp_servers()
             self._initialized = True
 
-        original_prompt = self.next_step_prompt
+        # Update system prompt with session folder information
+        original_system_prompt = self.system_prompt
+        self.system_prompt = self.get_system_prompt_with_session()
+
+        original_next_step_prompt = self.next_step_prompt
         recent_messages = self.memory.messages[-3:] if self.memory.messages else []
         browser_in_use = any(
             tc.function.name == BrowserUseTool().name
@@ -159,7 +218,8 @@ async def think(self) -> bool:
 
         result = await super().think()
 
-        # Restore original prompt
-        self.next_step_prompt = original_prompt
+        # Restore original prompts
+        self.system_prompt = original_system_prompt
+        self.next_step_prompt = original_next_step_prompt
 
         return result

app/agent/toolcall.py

@@ -196,10 +196,14 @@ async def execute_tool(self, command: ToolCall) -> str:
             )
 
             return observation
-        except json.JSONDecodeError:
-            error_msg = f"Error parsing arguments for {name}: Invalid JSON format"
+        except json.JSONDecodeError as je:
+            error_msg = (
+                f"Invalid JSON format for tool '{name}'. The tool arguments could not be parsed. "
+                f"Make sure all required parameters are provided and properly formatted as JSON. "
+                f"Error details: {str(je)}"
+            )
             logger.error(
-                f"📝 Oops! The arguments for '{name}' don't make sense - invalid JSON, arguments:{command.function.arguments}"
+                f"📝 JSON parsing error for '{name}': {str(je)}\nArguments: {command.function.arguments}"
             )
             return f"Error: {error_msg}"
         except Exception as e:

app/config.py

@@ -294,7 +294,7 @@ def _load_initial_config(self):
         if daytona_config:
             daytona_settings = DaytonaSettings(**daytona_config)
         else:
-            daytona_settings = DaytonaSettings()
+            daytona_settings = None
 
         mcp_config = raw_config.get("mcp", {})
         mcp_settings = None

app/prompt/manus.py

@@ -1,6 +1,19 @@
 SYSTEM_PROMPT = (
-    "You are OpenManus, an all-capable AI assistant, aimed at solving any task presented by the user. You have various tools at your disposal that you can call upon to efficiently complete complex requests. Whether it's programming, information retrieval, file processing, web browsing, or human interaction (only for extreme cases), you can handle it all."
-    "The initial directory is: {directory}"
+    "You are OpenManus, an all-capable AI assistant, aimed at solving any task presented by the user. You have various tools at your disposal that you can call upon to efficiently complete complex requests. Whether it's programming, information retrieval, file processing, web browsing, or human interaction (only for extreme cases), you can handle it all.\n"
+    "\n"
+    "IMPORTANT GUIDELINES:\n"
+    "1. Only use tools when truly necessary. Answer straightforward questions directly without tools.\n"
+    "2. python_execute: ALWAYS provide the 'code' parameter with complete, runnable Python code.\n"
+    "3. browser_use_tool: Use for web browsing, research, and information gathering from the internet.\n"
+    "4. str_replace_editor:\n"
+    "   - For 'create': MUST provide 'command', 'path', AND 'file_text' parameters. Never omit file_text.\n"
+    "   - For 'view': Provide 'command' and 'path' to view file/directory contents.\n"
+    "   - For 'str_replace'/'insert': Provide all required parameters (command, path, old_str, new_str, etc.)\n"
+    "5. ask_human: Use ONLY in extreme cases when user interaction is absolutely required.\n"
+    "6. CRITICAL: When calling ANY tool, provide ALL required parameters. Incomplete tool calls will fail.\n"
+    "7. Use the session folder for saving outputs: {directory}/output/\n"
+    "\n"
+    "The initial workspace directory is: {directory}"
 )
 
 NEXT_STEP_PROMPT = """

requirements.txt

@@ -1,8 +1,9 @@
-pydantic~=2.10.6
-openai~=1.66.3
+pydantic~=2.11.5
+openai~=1.108.1
 tenacity~=9.0.0
 pyyaml~=6.0.2
 loguru~=0.7.3
+structlog~=25.1.0
 numpy
 datasets~=3.4.1
 fastapi~=0.115.11
@@ -20,7 +21,6 @@ baidusearch~=1.0.3
 duckduckgo_search~=7.5.3
 
 aiofiles~=24.1.0
-pydantic_core~=2.27.2
 colorama~=0.4.6
 playwright~=1.51.0
 
@@ -33,10 +33,14 @@ httpx>=0.27.0
 tomli>=2.0.0
 
 boto3~=1.37.18
+daytona>=0.161.0
 
 requests~=2.32.3
 beautifulsoup4~=4.13.3
-crawl4ai~=0.6.3
+crawl4ai~=0.7.4
 
 huggingface-hub~=0.29.2
-setuptools~=75.8.0
+setuptools~=80.9.0
+
+# Dependency conflict resolution
+orjson>=3.11.5