Breaking away from hard-coding

[anonhostpi]

So a lot of work has been done lately to try to make the codebase more loosely coupled, but it has been done in a variety of inconsistent ways. I would like to foster a conversation of promoting good Inversion of Control (IoC) and encouraging contributors to stem away from "hard-coding," and possibly fix/address the issues with the current state of AutoGPT IoC

There are so many automation tools that this repo could make use of. It doesn't make sense to practice "hard-coding."

I know a lot of the re-arch is focused on making the repository more loosely-coupled, but I feel like there to needs to be a larger effort to ensure future contributors don't write code that is tightly-coupled or hard-coded.

No repository is ever going to achieve perfect IoC, but I think this repository could benefit from a higher standard for IoC. I have been reviewing a lot of the Issues and Pull Requests, and I have noticed a lot of poor IoC practices. I have also seen a lot of poor IoC practices continuously being corrected with more and more pull requests. I think elevating the IoC standard could mitigate these types of corrections.

What I had in mind for this conversation was to foster a conversation for updating the contributing guides and the manifesto to reflect the need to improve the repository's IoC.

Given the size, traction, problems, and development speed of this repo, I think we should make IoC a priority. The re-arch seems to be the first step in the right direction, but I think the second step would be to put in a standard to DISCOURAGE poor IoC practices when contributing, and emphasize to contributors why that is needed.

EDIT:

Another problem that is caused by "hard-coding" is poor localization. The repo inevitably wants to support multiple languages, but as it stands, most of the prompting and language-based features are hard-coded. I have seen a lot of efforts to support various asian and latin-based languages, however a common theme I see in the PRs is to "hard-code" the localization.

[anonhostpi]

Related PRs/Issues/Discussions Thread.

[anonhostpi]

Re-Arch Work

https://github.com/Significant-Gravitas/Auto-GPT/labels/re-arch - Re-Arch Labeled Issues and PRs
#3652 - (Re-Arch) Plugin API work
#3824 - (Re-Arch) Work to separate commands into plugins
#3790 - (Re-Arch) Planner Interface work
#3795 - (Re-Arch) Introduce the Workspace Base Class
#3805 - (Re-Arch) Introduce a Configuration Abstraction for the Re-Arch
#3717 - (Re-Arch) Agent State Encapsulation (I think this is supposed to actually describe the new architecture?)
#3717 - (Re-Arch) Message Broker (?not sure what this is supposed to be yet)
#3850 - (Re-Arch) Abstraction for the Language Model
#3853 - (Re-Arch) Work to Decouple Logging into its own Interface.

Re-Arch Related Work

#3916 - Proposal to Self-Regulate through Observer Agents

Other Decoupling Proposals

#3593 - Proposal to separate the planner and executor into 2 separate agents.
#3775 - Proposal to separate the math calculation functions into their own plugin

PRs that Hard-Code ☹

#3815 - Proposal to hardcode additional git operations
#749 - Proposal to hardcode bitcoin commands
#364 - Proposal to hardcode additional git operations and add react support

[GoMightyAlgorythmGo]

I like some part with super fast finetuning in mutlible layers or deepRL or a mix of hard coding and some more "lerning" "soft" parts

[sohrabsaran]

May be ask Auto-GPT to read its own codebase and output a review of each file. Rank the files from worst to best based on the review comments per file.