With the rapid advancement of Generative Artificial Intelligence (AI) tools, human-agent interaction and collaboration in education have become indispensable. However, these interactions face significant challenges due to the complexity of integrating AI agents into educational contexts. To enhance our understanding of human-agent interaction and collaboration in education, we conducted a systematic review following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) workflow. This review maps the landscape of research on human-agent interaction in education, including agent types and roles, teaching modes, variables of interest, learning theories, data collection and analysis methods, and interaction patterns. Our findings indicate that the development of human-agent interaction in education is still in its early stages, with a particular focus on disembodied and virtual agents acting as assistants or tutors in one (agent)-to-one (learner) learning scenarios, especially in language learning and computer science. Key areas of interest include learning performance, behaviours, and engagement. The primary data sources are human-agent conversations, questionnaires, interviews, and log data, with content analysis often combined with other methods to analyse process-related data. Notable interaction patterns include task specific aspects transition, general learning experience of human-agent interaction, hints seeking behaviour transition, human-agent engagement and learning performance. Despite these advancements, significant gaps remain in understanding the dynamics of human-agent interaction, such as the dynamic interaction process, complex problem-solving strategies, scalability, and higher-order thinking. This review highlights the need for further research to address these gaps and improve the efficacy of human-agent collaboration in educational settings.