Almost half of Earth's surface is threatened by agriculture, which has extensively degraded ecosystems and resulted in significant biodiversity loss. Remnant ecosystems in fragmented agricultural landscapes are threatened by past and present grazing and land-clearing. Declines in native diversity are common in these ecosystems, and their restoration is a key conservation goal globally. Understanding the drivers of change in floristic condition, reflecting continuity in floristic composition towards native plant communities, is fundamental to inform effective restoration practice. Previous investigations have demonstrated abiotic and biotic drivers of floristic condition independently. However, few consider the combined influence of these drivers on floristic condition, or the interactions between them, which may mediate indirect effects (e.g. plant-soil interactions). Despite this, ecological interactions may underpin changes in floristic condition, and provide critical insights needed to inform restoration. Here, we use structural equation modelling to disentangle the relationships between plants, soils and grass and litter biomass (leaf litter and fine woody debris) to elucidate the direct and indirect drivers of floristic condition in some of the most degraded landscapes globally: the critically endangered box-gum grassy woodlands in south-eastern Australia. We identify divergent plant-soil interactions between native versus exotic plants to key soil properties including soil nitrate and phosphorus. Specifically, native plants were negatively associated with increasing soil fertility, which favored exotic species. We also found evidence of indirect effects on floristic condition, mediated through interactions between litter biomass, soils and the basal area of overstorey trees. Our findings highlight the major role of soils in shaping floristic condition through direct and indirect pathways, and the role of multivariate interactions in mediating these pathways in a highly degraded, critically endangered ecosystem. Effective restoration must therefore consider the multivariate direct and indirect drivers of ecological condition to maximise positive outcomes in these landscapes and those similar.