model_reasoning_effort, Claude Code maps it to --effort,
OpenCode maps it to --variant, and Grok Build maps it to
--reasoning-effort. Common values include
low, medium, high, xhigh, and
max. Unsupported agents ignore this field.