Standard classifications obscure which jobs directly drive emissions cuts. Using U.S. online vacancies data, we develop a transparent, skill-based approach that identifies low-carbon roles within occupations, leveraging NLP and text linked to established green classifications. We show that, even within the same occupation and firm, low-carbon jobs systematically demand more, and more diverse, skills than non-low-carbon jobs. Within-occupation differences account for much of the overall gap, implying occupation-level studies understate it. The transition thus requires substantial retraining within existing occupations, even if not biased towards high-skilled workers. Reskilling needs are highly occupation-specific. Returns to skill complexity are higher in low-carbon roles, yet the green wage premium is positive but modest and declining after controlling for occupation and firm heterogeneity. Low-carbon jobs are more geographically dispersed than high-carbon ones but more prevalent in wealthier areas, implying reallocation frictions and equity concerns. Our evidence supports targeted reskilling policies to support a just transition.