pass a returned file from a function to another function with agent

[Zohreh6384NKH]

hi could you please help me wih this. i have a function which return a dictionary of files .now i want to get an agent to pass this dictionary to another function as input and want an agent to call the fucntion. how agent can get the output and pass it as input of another function

[ekzhu]

I am not sure what's the blocker here is? You can call one agent to list the files, and then call other agent to read the files. Or, having the same agent with both tools, and call the agent twice.

Could you describe a bit more what is the context and scenario? By just reading the description it feels like this should be accomplished by just calling two agents or one agent two times.

[Zohreh6384NKH]

the scenario is that I want to read a bunch of python files and extract helper functions or utility functions from them. for this i decided to use agentic workflow. first task is working perfect.one agent can call a function and get the list of files. now i was thinking for extracting the helper functions first i need to read the content of each file . and after reading them probably an llm can easily recognise helper functions on its own.

[Zohreh6384NKH]

the blocker is i should give list or dictionary of files to second function and call the function .how we can give the list of file to second agent

[ekzhu]

@Zohreh6384NKH I believe group chat is a wrong choice for your scenario. You have deterministic sequence of steps, but group chat cannot provide that deterministic sequence as it uses an LLM to choose the next step.

f"Please follow these steps sequentially:\n"

If your steps are sequential, why using group chat?

What will work well for you, is to call each agent in sequence manually, and pass relevant chat history as context from one agent to the next one.

Group chat is for handling ad-hoc scenarios when for each task, the steps are decided on the fly. This is similar to a single agent with all the tools -- the order of tool invocation is different for different tasks. If you have such scenario, choose GroupChat. You can add some determinism by using a selector function.

I can see you are using an outdated v0.2 API. I highly recommend upgrading to v0.4. Make sure you install autogen-agentchat~=0.4.0 to upgrade your AutoGen.

[Zohreh6384NKH]

thank you so much Eric. sure I will do sequential

[Zohreh6384NKH]

Hi Eric . I tried to use sequential chat but still I can only receive the dictionary of python files which is the first step. for the second step I should somehow give this dictionary and pass it through another agent to call read_file_content but it can not get. here is my code could you tell me what is the problem with my prompt:

…

On Tue, Jan 28, 2025 at 10:50 PM Eric Zhu ***@***.***> wrote: @Zohreh6384NKH <https://github.com/Zohreh6384NKH> I believe group chat is a wrong choice for your scenario. You have deterministic sequence of steps, but group chat cannot provide that deterministic sequence as it uses an LLM to choose the next step. f"Please follow these steps sequentially:\n" If your steps are sequential, why using group chat? What will work well for you, is to call each agent in sequence manually, and pass relevant chat history as context from one agent to the next one. Group chat is for handling ad-hoc scenarios when for each task, the steps are decided on the fly. This is similar to a single agent with all the tools -- the order of tool invocation is different for different tasks. If you have such scenario, choose GroupChat. You can add some determinism by using a selector function. I can see you are using an outdated v0.2 API. I highly recommend upgrading to v0.4. Make sure you install autogen-agentchat~=0.4.0 to upgrade your AutoGen. — Reply to this email directly, view it on GitHub <#5238 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BFLSJZJHIQPVIPSUD5F526L2NACSJAVCNFSM6AAAAABWBFV372VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCOJYHE3TGOA> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[Zohreh6384NKH]

---

EXECUTING FUNCTION read_files_content...
user_proxy (to read_files_agent):

***** Response from calling tool (ollama_func_8990) *****
Error: 'str' object has no attribute 'items'

---

---

read_files_agent (to user_proxy):

{"name": "read_files_content", "parameters": {"files_by_folder": "{'folder1': ['file1.py', 'file2.py'], 'folder2': ['file3.py', 'file4.py']}", "repo_path": "/path/to/repo"}}

---

user_proxy (to read_files_agent):
as you can see read_files_agent can not get the dictionary .It simply get one string instead of dictionary of files so i get this error

[ekzhu]

@Zohreh6384NKH I believe you shouldn't be using AI agents. Use VSCode + GitHub Copilot I can quickly generate the following code that does what you want, 100 % of the time.

import os
import tempfile
from typing import Dict, List

def list_python_files_by_directory(dir: str) -> Dict[str, List[str]]:
    """
    Get all the python files in the given directory,
    grouped by the directory they are in.
    """
    python_files = dict()
    for root, _, files in os.walk(dir):
        python_files[root] = [file for file in files if file.endswith(".py")]
    return python_files

def read_files_content(files: Dict[str, List[str]]) -> Dict[str, str]:
    """
    Reads the content of the given files and returns a dictionary
    with the full file name as the key and the content as the value.
    """
    files_content = dict()
    for directory, files in files.items():
        for file in files:
            with open(os.path.join(directory, file), "r") as f:
                files_content[os.path.join(directory, file)] = f.read()
    return files_content

if __name__ == "__main__":
    with tempfile.TemporaryDirectory() as tmpdirname:
        print(f"Created temporary directory: {tmpdirname}")

        # Put some files in the temporary directory
        with open(os.path.join(tmpdirname, "file1.py"), "w") as f:
            f.write("print('Hello, World!')")
        with open(os.path.join(tmpdirname, "file2.py"), "w") as f:
            f.write("print('Hello, Python!')")

        python_files = list_python_files_by_directory(tmpdirname)
        files_content = read_files_content(python_files)
        for file, content in files_content.items():
            print(f"File: {file}\nContent:\n{content}\n")

Maybe you can add a prompting step at the end just to get some natural language response.

[raimondasl]

@ekzhu : I believe @Zohreh6384NKH wants to send the .py file contents to an LLM to extract helper functions from them. I agree with you though that finding all files and reading their content does not require agents or LLMs and should be done as you illustrated. Then the content of each can be sent to LLM for helper function extraction. Even this can be done without agent, but I guess an agent can be used for it. Just send the file content to an agent and get result back?

[Zohreh6384NKH]

exactly @raimondasl , I want to extract utility or helper function from python codes. so my question is how can give the content of files to an agent. can i pass it in a prompt for an agent?

[ekzhu]

Yes. Just pass it to the prompt.

Though I think you should think about ways to do this without using LLM. E.g., AST parser

[Zohreh6384NKH]

Thanks Eric . I just wanted to do it in an agentic workflow. Now that I have the content of the files, do you think for extracting helper functions in a code I can create a conversable agent that does the task for me?

…

On Wed, Jan 29, 2025 at 5:19 PM Eric Zhu ***@***.***> wrote: @Zohreh6384NKH <https://github.com/Zohreh6384NKH> I believe you shouldn't be using AI agents. Use VSCode + GitHub Copilot I can quickly generate the following code that does what you want, 100 % of the time. import osimport tempfilefrom typing import Dict, List def list_python_files_by_directory(dir: str) -> Dict[str, List[str]]: """ Get all the python files in the given directory, grouped by the directory they are in. """ python_files = dict() for root, _, files in os.walk(dir): python_files[root] = [file for file in files if file.endswith(".py")] return python_files def read_files_content(files: Dict[str, List[str]]) -> Dict[str, str]: """ Reads the content of the given files and returns a dictionary with the full file name as the key and the content as the value. """ files_content = dict() for directory, files in files.items(): for file in files: with open(os.path.join(directory, file), "r") as f: files_content[os.path.join(directory, file)] = f.read() return files_content if __name__ == "__main__": with tempfile.TemporaryDirectory() as tmpdirname: print(f"Created temporary directory: {tmpdirname}") # Put some files in the temporary directory with open(os.path.join(tmpdirname, "file1.py"), "w") as f: f.write("print('Hello, World!')") with open(os.path.join(tmpdirname, "file2.py"), "w") as f: f.write("print('Hello, Python!')") python_files = list_python_files_by_directory(tmpdirname) files_content = read_files_content(python_files) for file, content in files_content.items(): print(f"File: {file}\nContent:\n{content}\n") Maybe you can add a prompting step at the end just to get some natural language response. — Reply to this email directly, view it on GitHub <#5238 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BFLSJZJJF3264SHG7T64VL32NEERPAVCNFSM6AAAAABWBFV372VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCOJZHAZDANI> . You are receiving this because you were mentioned.Message ID: ***@***.***>