Social bots are AI-based algorithms aimed at imitating (and often influencing) the behavior of users on social media. In recent years, bots have been largely used for malicious purposes, like spreading disinformation and conditioning electoral campaigns. To preserve the security and privacy of legitimate users, social media platforms need highly effective solutions to identify bot-driven accounts. Numerous approaches have been proposed to address this problem. However, to better understand how to (i) automatically recognize the specific category a bot belongs to, (ii) early detect bot-driven accounts, and (iii) design platform-independent solutions, further research is needed. In this paper, we consider a stylistic-consistency-based approach for social bot detection and assess whether such an approach can be used to bridge these research gaps. Our results demonstrate that the stylistic-consistency-based approach can (i) identify the specific bot category with an F-measure higher than 95% and (ii) enable near-early detection of bot-driven accounts, achieving high F-measure values when considering a low number of tweets/posts. Though the method is platform-independent, it needs to be trained with platform-specific data to catch the stylistic footprints of bots operating on the particular platform.