Conflict is a significant driver of global food insecurity, and agricultural assistance is a key policy tool to improve food security in such settings. However, evidence on whether and for whom such interventions are effective in conflict settings remains limited. Using panel data from 883 households in Syria and a quasi-experimental design, we estimate the average and heterogeneous treatment effects of an agricultural asset transfer on household food security. Applying the honest causal forest algorithm, we assess how gender, conflict intensity, and other household and contextual factors shape treatment heterogeneity. On average, the intervention increases the Food Consumption Score by 8.2%. The largest gains (12%) are observed among female-headed households in moderately conflict-affected areas, while households in low- or high-intensity conflict zones show limited or no improvement. These findings underscore the importance of contextualized targeting in fragile settings and caution against relying solely on average treatment effects, which may obscure critical heterogeneity in program impact.