Runtime harnessing, test-time compute, and onboard reasoning for vision-language-action models - View it on GitHub