Ability to Navigate Outside Working Directory.

[nathanlannan]

Currently, the best workaround is bringing the desired context into the working directory, but implementing the ability to crawl through files more gracefully would be great. Perhaps for fully autonomous users, this could be an opt-in or out type thing.

SYSTEM: Command read_file returned: Error: Attempted to access outside of working directory.

(This project rocks by the way, great work!)

[Torantulino]

Yeah, it's an interesting option.

I think the major concern from that though would be that it could, and would easily delete/overwrite files.

I have been playing about with the same approach as you, cloning the repo into the AI's working directory! 😆

[Torantulino]

Thank you so much for the kind words 😊 I'm glad you're enjoying it!

[alwinraju]

What exactly is the working directory for this? I have put my file in the same folder as the main.py file and it still gives me an error saying it's outside the Working Directory. I am working on an M1 Mac, is there other permissions I need to grant for it to have access?

[xcreateadmin]

What exactly is the working directory for this? I have put my file in the same folder as the main.py file and it still gives me an error saying it's outside the Working Directory. I am working on an M1 Mac, is there other permissions I need to grant for it to have access?

same issue on my M1 Mac

[JacobFV]

I tried making a link to the workspace from inside the AutoGPT dir:

~/github/Auto-GPT on master! ⌚ 9:42:38
$ ln -s ~/code/<my_repo>

(Note: From my use, the search_files always returns an empty list, regardless of whether accessing an inode actually under AutoGPT or just a link.)

[Nonary]

What exactly is the working directory for this? I have put my file in the same folder as the main.py file and it still gives me an error saying it's outside the Working Directory. I am working on an M1 Mac, is there other permissions I need to grant for it to have access?

It's the "auto_gpt_workspace" folder.

[findlestick]

It's the "auto_gpt_workspace" folder.

So, putting the file that you want it to read, inside of "auto_gpt_workspace" should let the program access it?

[vvp3]

I have my files in project folders. it is really cumbersome to move them around.
Also another big issue is that files have references to other files in the project folder, so auto-gpt should be optimised to read the referenced files as well, isn't it ?

[vvp3]

I am trying to search in the code where is this folder, but i think it should be like an option in the begining to choose the working folder .

Also, another suggestion would be to really do something about the agents like you wrote here ;)

# Soon this will go in a folder where it remembers more stuff about the run(s)
SAVE_FILE = os.path.join(os.path.dirname(__file__), '..', 'ai_settings.yaml')

workaround is to save each yaml file as a backup

[FatStinkyPanda]

I'm still getting the message "error: attempted to access outside of working directory". I have saved a .py file to the workspace directory and have tasked auto-gpt with completing the code. I've given auto-gpt the location of the directory "C:auto_gpt\auto_gpt_workspace" for it's goals and description prompts and "C:auto_gpt" is where I run auto-gpt in powershell. With a blank .py file or a .py program template with descriptions and context it continues to have the error message. I've seen people have auto-gpt write code and I'm very excited to try it out myself.

[styks1987]

Using a symlink works for me.

ln -s /path/to/my/app ./auto_gpt_workspace

Then in your goals you could write something like

Write documentation for all the functions in app/functions/stripe-hook/stripe.js