Automated segmentation is routinely used as the first step in quantitative image-based clinical workflows, where predicted masks are subsequently reviewed and corrected by human operators. Automated correction of the residual structural errors is currently addressed one anatomy at a time, so the number of trained correctors scales with the number of segmentation targets. We present GRACE, a retrieval-augmented refiner that edits an input mask under context drawn from a shared knowledge base of verified prior cases. The refiner is trained once on a multi-organ corpus spanning compact organs, cavities, branching vessels, and tumour lesions. In the zero-shot dataset-transfer evaluation, three held-out datasets query a fixed, training-only knowledge base; a separate leave-one-anatomy-out experiment evaluates parameter-free adaptation by indexing verified support cases of a previously unseen compact organ. A shape-aware retrieval projector, fitted with a Dice-cosine objective on a frozen multimodal embedder, aligns the space that the knowledge base ranks by (raw multimodal similarity) with the space that the refiner consumes (shape similarity). We evaluate GRACE on 18 public datasets spanning CT, MRI and ultrasound, including three held-out CT/MRI datasets for zero-shot dataset transfer, using eight refinement configurations and seven upstream segmentors that span weak, mid, and strong operating regimes. GRACE improves weak upstream segmentations by up to +0.35 DSC while producing only small mean changes on strong segmentors. Our results suggest retrieval-conditioned refinement is a practical alternative to anatomy-specific correction models.
