Inappropriate antibiotic use presents a major global health challenge, particularly in resource-limited settings where access to quality care remains limited while access to antibiotics is relatively unrestricted. This study estimates the relationship between frontline primary care quality and community antibiotic use, drawing on detailed data from approximately 100 rural villages in China that link objective measures of village doctor clinical practice quality, assessed through unannounced standardized patient visits, to household-level antibiotic use data measured within the same communities. To mitigate potential endogeneity concerns, we employed an instrumental variable (IV) approach integrated within a double/debiased machine learning (DML) framework, using ensemble machine learning to combine information across multiple individually weak but jointly informative provider-level instruments. Higher clinical practice quality was associated with lower probabilities of inappropriate antibiotic receipt during healthcare encounters for cold or diarrhea symptoms (OLS: 6.2 percentage points; DML-IV: 9.6 percentage points) and households storing antibiotics at home for future use (OLS: 4.1 percentage points; DML-IV: 10.9 percentage points). These findings highlight the potential role of improving frontline primary care quality in community antibiotic stewardship. Methodologically, this study illustrates how causal machine learning can complement conventional econometric methods by addressing high-dimensional estimation challenges in global health economics research.