Embryogenesis is commonly viewed through a tree model of cell differentiation, which does not adequately represent the spatiotemporal modulation of cell multipotency underlying morphogenesis. Here we develop an integrated approach, combining in vivo single-cell high throughput clonal lineage tracing with machine learning, to systematically decompose continuous spectra of clonal fate biases in mouse embryos traced from neurulation until mid-gestation. The reconstructed patterns of recurrent clonal variation uncovered gene programs driving dynamic positional biasing of clonal composition during axial skeletogenesis and peripheral neurogenesis. Mosaic combinatorial perturbations targeting multiple receptors, including the Hedgehog pathway, led to novel clone types, which has implications for engineering custom cell type assemblages from well-defined progenitors in vivo. Altogether, our work demonstrates an effective practical approach for interrogating programs guiding lineage specification.