Plan mode relies on parsing text from an llm response. This can be unreliable. We should likely remove opencode from the loop and use an ai sdk agent that emits structured outputs. We could also see if opencode supports returning structured output in its last message.
Plan mode relies on parsing text from an llm response. This can be unreliable. We should likely remove opencode from the loop and use an ai sdk agent that emits structured outputs. We could also see if opencode supports returning structured output in its last message.