OpenCode Support

[TomLucidor]

This tool looks interesting, if it can support a more general agent, that would be great
P.S. What is your take on AI-Scientist by SakanaAI and other similar endeavors?

[mx-Liu123]

Hi @TomLucidor , thanks for your feedback! I will try to add opencode, hoping there will be a way in opencode to allow agent has full access within work directory while be limited outside.

I have used similar endeavors like R&D agent and openevolve etc, trying to apply them to my research. But I feel like they are not flexible enough to be adapted to my own research. So I create AgentCommander, focusing on the trial-and-error loop and making it convenient to monitor the agent progress and flexible+one-click to be adapt to general task. (like recently updated 'ml_autosetup_1', A one click Setup script given your data X.npy and Y.npy.)

Another key points in similar endeavors is to prevent the agent to cheat/fool us. In AgentCommander, tasks like ML optimization have a clear metric to evolve and there are a set of design like evaluator.py/metric to limit agents. For a auto-researcher goal discussed by AI-Scientist, it is hard to set a standard to let agents clearly know what is good, what is not and what shouldn't do.