Learning Analytics (LA) is collecting and analyzing traces of learners' activities in order to understand and improve learning. This paper focuses on traces generated using the version control system Git. Existing works on the topic have limitations regarding the quality of the traces they analyze: (1) the quantity and content are not always sufficient for in-depth analysis of student behavior, (2) their limited reliability can lead to a loss of exploitable data, and (3) the method of generating these traces is not generic. We propose a new interactive system based on Git and the observation of file modifications to generate automatically reliable and rich traces. This interactive system will soon be experimented with in an ecological context and is intended for diversified teaching contexts.