Abstract: In the context of digital and intelligent territorial spatial governance, urban health examination and evaluation is shifting from indicator monitoring toward comprehensive diagnosis. Yet indicator-centered assessment still tends to describe anomalies superficially, while the integration of data analytics and professional planning knowledge remains weak. This paper understands indicator-based assessment as a systematic cognitive process for interpreting the operating state of complex urban systems, and proposes an S-E-D diagnostic ontology composed of Symptom, Explanation, and Decision. On this basis, it develops a knowledge-enhanced large language model method that supports indicator analysis, causal explanation, and strategy generation under conditions of constrained planning knowledge through the structured processing of planning documents and technical standards. A case study of Nanjing shows that the proposed method performs well in multi-indicator integration, reasoning consistency, and the interpretability of strategic recommendations. The study offers a methodological reference for transforming urban indicator assessment from experience-dependent practice into reproducible and traceable intelligent diagnosis.