FileSurfer cannot access file.

[crenus]

Im trying to use MagenticOne currently just running an example to get it to read a file. When I run it, I am getting an error of FileSurfer not able to access files.
It seems that I cannot access the contents of requirements.txt directly at the moment. However, if you cou
ld provide the contents of the file here,

Has anyone seen this before? I dont think I am doing anything odd....
I have tried a few different things, using the prebuilt agent, and even creating my own group chat and same thing.

• Running on Ubuntu (WSL)
• Using venv
• running Python 3.12.9
• Using latest autogen version

Example code (modified from original)

import asyncio
import json
from autogen_ext.models.openai import AzureOpenAIChatCompletionClient
from app import get_access_token  # Import only the token retrieval function
from autogen_ext.teams.magentic_one import MagenticOne
from autogen_agentchat.ui import Console
from autogen_ext.agents.file_surfer import FileSurfer
from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.teams import MagenticOneGroupChat
from autogen_core.models import UserMessage
import logging

from autogen_core import TRACE_LOGGER_NAME, EVENT_LOGGER_NAME

logging.basicConfig(level=logging.WARNING)

BASE_URL = 'http://localhost:3000'

async def example_usage():
    token = get_access_token()  # Get the access token
    client = AzureOpenAIChatCompletionClient(
        model="gpt-4o-mini",
        base_url=BASE_URL,
        api_key=token,
        api_version="2024-07-01-preview",
    )
    assistant = AssistantAgent(
        "Assistant",
        model_client=client,
    )
    file_agent = FileSurfer("FileSurfer", model_client=client)
    team = MagenticOneGroupChat([assistant, file_agent], model_client=client)
    await Console(team.run_stream(task="Read the content of requirements.txt in the current directory and create a summary of the dependencies."))

# m1 = MagenticOne(client=client)
    # task = "Create a README for the current directory. It should look for any comments in files or understand the different dependencies and text to understand what this repo does. It should include how to set up the projet, description, tools used, and anything you think may be good to add."


if __name__ == "__main__":
    asyncio.run(example_usage())

[ekzhu]

It would be easier for others to assist you if you post your code, input and output.

[crenus]

THanks, yea so I updated the post with some small minor things excluded. But I tried both Magentic One and then creating my one group chat with it to see if that would change anything but it did not.

[crenus]

Ok so I have partially resolved this I think.... 1 hand was using an older version. On the other hand, we have an internal system, so I have a proxy that I created to proxy the request (Some custom work needs to happen there). Fixed that as it was missing tools so now I have it pass the whole body...
Now I have it where Im attmepting to have it create a README which the logs say nThe README file has been written to README.md in the current directory, but there is no file that gets created...

[ekzhu]

Is it LLM output? Hallucination?

[crenus]

I am not totally sure. Results are obviously inconsistent.
Sometimes it would act as if the file was created.
Other times, I could see that it would try to create a python script to write the README, but what it would do is only get part of the text in a string, and then not finish and have an unterminated triple quote string....
It looked like it would do this when there was code blocks, so I added the following to my prompt:
It should escape any code block characters to prevent unterminated triple-quoted string literal. Do not use any code blocks in the README

From here, it was still failing last night, and now this morning it actually worked.

I saw some other posts but didnt quite find what I was looking for... If the output has a code block, will it try to execute it even if its supposed to be text in like a README for example? I'm not entirely sure why the creation of a script would just stop at code blocks..

[crenus]

Yea so I ran it again without the above, and it gives me in the python file something like:

# write_readme.py

readme_content = """
# Magentic One Orchestrator

## Description

## Setup Instructions

1. **Clone the Repository**
   

And so you can see that it just stops the output. Looking at the logs, right after that last line and a new line, is a code block