Habits are an essential part of everyday decision-making. However, the mechanisms underlying habit formation and expression in humans are difficult to study in the laboratory, owing to a dearth of convenient experimental paradigms that reliably exhibit a key marker of habits - training-induced inflexibility - under ecologically valid conditions. This difficulty is often attributed to the fact that habits are identified in contrast to goal-directed (GD) control, which research participants typically engage strongly in laboratory experiments. To address this gap, we develop a new, short habit learning paradigm that incorporates several features we hypothesized would encourage participants to disengage GD control, enabling habits to exert greater influence over behavior: a hierarchical multi-step trial structure, opportunities for self-correction, and frequent switches between extensively and moderately practiced behaviors. Through a series of experiments, we demonstrate that overtraining amplifies habitual control, as evidenced by errors biased toward the overtrained context and away from the moderately-trained context at early response times, while later responses remain dominated by GD control. The reliability of this overtraining effect depended on the inclusion of task features designed to dampen GD control. In addition to providing a practical, robust, and flexible tool for studying the cognitive processes underlying habit formation and habitual control, our paradigm moves us beyond the traditional stimulus-response conception of habits, expanding the definition to include more complex, hierarchical behaviors that better reflect naturalistic human habits.