TCM Dietary Differentiation and Treatment System

Problem It Addresses

A common failure in TCM dietary recommendation is turning sparse user symptoms into an overly precise syndrome, then suggesting foods without checking contraindications. For example, a user may only say “fatigue,” yet an agent may output “lung qi deficiency”; a user with “consumptive thirst” may be recommended red dates or longan; and noisy classical entries can pollute keyword matching. tcm-dietary turns these risks into explicit rules: expand symptoms, run multiple differentiation engines in parallel, and cross-check foods against chronic-disease and ingredient constraints before generating a recommendation.

Core Capabilities and Workflow

• Differentiation SOP: when calling diagnose(), the skill runs symptom expansion, parallel scoring across Eight Principles, five organ, five phase, qi-blood, and etiology engines, then performs knowledge-base matching, conflict resolution, syndrome selection, and contraindication checks.
• Knowledge-base grounding: it retrieves from sources such as tcm-theory.json, ingredients.json, dishes.json, and chronic-diseases.json, instead of relying only on free-form generation.
• Contraindication validation: before recommending foods, it checks avoid_foods or ingredient properties, and the recommended list must not overlap with the forbidden list.
• Supporting features: personalized recipes, dish improvement, constitution-based tea recommendations, and guiding exercises make it useful for embedding TCM dietary advice into agent workflows.

Boundaries

With fewer than 3 symptoms, it should return broad Eight Principles or organ-level guidance rather than a precise syndrome name. TCM syndrome names are not equivalent to Western medical diagnoses, and any formula involving toxic herbs must include dosage limits and contraindications. It is best understood as a constrained knowledge-base workflow, not a direct source of medical conclusions.