Optional reasoning effort for agents that support it. Codex maps this to model_reasoning_effort, Claude Code maps it to --effort, and OpenCode maps it to --variant. Common values include low, medium, high, xhigh, and max. Unsupported agents ignore this field.