There is no project structure to run the generated code correctly

[nordanz]

Bug description
No matter which target language is selected for the software project, always the dependencies will be generated in requirments.txt

Bug solved method

I would suggest to add a SeniorEngineer role or action which will update the code structure to be more target language oriented. for example in case of nodejs, having the package.json and in case of cpp having cmakefiles or conan files. ofcourse you can't implement all best practices but at least a role or agent which can be customized by the user for this specific purpose.

Environment information

• LLM type and model name: gpt4-o
• System version: ubuntu 22.-4
• Python version: 3.9
• MetaGPT version or branch: latest docker version

Screenshots or logs
reproducable by this prompt: "write a snake game in nodejs"

[iorisa]

The software company of MetaGPT is a demo used to show the plasticity of the SOP mode of the MetaGPT framework. Since the SOP is designed and developed in Python, there will be some inconsistencies when switching to other languages. For example, the dependency package configuration for Node.js should be placed in package.json, not requirements.txt.
In addition, the SOP mode is more suitable for operational scenarios where the process is already fixed.
For scenarios that are more variable, such as software development, the Data Interpreter mode is more highly recommended.