How I optimized my Opencode Go credit usage
I mostly use Kimi K2.6 for development. It can burn the credits pretty fast. So, I had to optimize my token usage. I noticed, when kimi runs tests or any commands the output can quickly fill up the context and cause unnecessary credit burn. So, I defined a executor sub-agent with deepseek v4 flash. It's only purpose is to execute command and return a summarized output. Then, instructed the main agents to use this executor sub-agent for running any command, tests etc. It has reduced my credit usage significantly.
You can check my opencode configuration here: https://github.com/hossainemruz/dotfiles/tree/main/dot_config/opencode