<p>To address the industry challenges of insufficient real-time diagnosis, difficulty in explaining causal mechanisms, and heavy reliance on expert knowledge in aquaculture disease diagnosis, particularly the issue of constructing structurally stable, semantically consistent, and interpretable disease causal pathways under zero-shot conditions, this study proposes a domain-constrained zero-shot multi-agent causal reasoning framework (DPAIA), in which structured prompting and agent collaboration are introduced as inductive biases to guide the reasoning process in complex aquaculture scenarios. This results in a standardized, structurally consistent, and interpretable disease causal chain generation process. Experimental results show that DPAIA achieves an F1 score of 0.773 on the Aqua-ZSCR dataset, significantly outperforming several mainstream multi-agent reasoning methods in multiple metrics, including precision, recall, and causal chain structural similarity. Additionally, DPAIA maintains stable performance on the general-domain CaLM dataset, verifying its excellent generalization ability and causal structure consistency in cross-domain scenarios. Results suggest that DPAIA provides a structurally clear and mechanistically interpretable causal path analysis solution for aquaculture disease risk tracing, early warning, and on-site farming decision support, offering an interpretable zero-shot causal inference paradigm for intelligent management and precision prevention in aquaculture.</p>