Do you steer your agents by manually running every single test and linter and reporting the results back to them?