Most impedance-based walking controllers use a finite state machine (FSM) with dozens of user-specific parameters that need to be manually tuned by technical experts. These parameters are only optimal near the task (\eg walking speed and incline) at which they were tuned, resulting in decreased performance as task inevitably varies. This paper presents a tuning-free, phase-based controller that uses a hybrid combination of continuously-variable impedance control during stance and kinematic control during swing to enable biomimetic locomotion over a continuum of tasks. After generating a data-driven model of variable joint impedance with convex optimization, we implement a novel task-invariant phase variable and real-time estimates of speed and incline to enable the controller to autonomously adapt to task variation. Experiments with an amputee participant using a powered knee-ankle prosthesis show that our tuning-free controller 1) features highly-linear phase estimates and accurate task estimates, 2) produces more biomimetic joint work trends compared to a hand-tuned FSM impedance controller, and 3) achieves lower kinematic and kinetic error than the FSM impedance controller in 7 of 8 tested metrics. Our data-driven control approach may allow easier clinical implementation of variable-activity powered knee-ankle prostheses by replicating biological behavior across tasks without expert tuning.