Reconstruction of gravitational lensing effects in the CMB from current and upcoming surveys is still dominated by temperature anisotropies. Extragalactic foregrounds in temperature maps can induce significant biases in the lensing power spectrum obtained with the standard quadratic estimators. Techniques such as masking cannot remove these foregrounds fully, and the residuals can still lead to large biases if unaccounted for. In this paper, we study the "shear-only" estimator, an example of a class of geometric methods that suppress extragalactic foreground contamination while making only minimal assumptions about foreground properties. The shear-only estimator has only been formulated in the flat-sky limit and so is not easily applied to wide surveys. Here, we derive the full-sky version of the shear-only estimator and its generalisation to an
multipole estimator that has improved performance
for lensing reconstruction on smaller scales. The multipole estimator is
generally not separable, and so is expensive to compute. We explore separable
approximations based on a singular-value decomposition, which allow efficient
evaluation of the estimator with real-space methods. Finally, we apply these
estimators to simulations that include extragalactic foregrounds and verify
their efficacy in suppressing foreground biases.