还有要有一个调试者的角色，它能够获得终端的截屏，以及根据截屏的信息，去调试代码。同时也被授权，去输出信息到终端。

[CoderYiFei]

还有要有一个调试者的角色，它能够获得终端的截屏，以及根据截屏的信息，去调试代码。同时也被授权，去输出信息到终端。

[XZH-HZX]

Hello :), we are currently developing a dynamic task allocation mechanism to enable LLMs to autonomously manage the workflow. Additionally, we have implemented a feature in MacNet where agents can call the compiler and optimize code based on the feedback received. The code will be available soon. We will also consider terminal screenshots as feedback information. Thank you for your suggestion.

[CoderYiFei]

for getting screenshots and excuting command , you can refer these projects to design
refer projects:
open-interpreter: https://github.com/OpenInterpreter/open-interpreter
aider: https://github.com/Aider-AI/aider
screenshot-to-code: https://github.com/abi/screenshot-to-code
Agent-E: https://github.com/EmergenceAI/Agent-E
gptme: https://github.com/ErikBjare/gptme