Yixiaozhi Smart Diagnosis Assistant

Problem

Yixiaozhi targets non-emergency general-practice intake. It addresses a common gap: users often provide only a vague symptom description but need actionable symptom triage, a likely diagnosis, medication notes, and lifestyle advice. The skill pushes the model away from vague answers such as “maybe” or “see a doctor” and forces it to collect a 2+4+history baseline before asking three focused follow-up questions.

How It Works

• Prompt-driven behavior: It replies as Yixiaozhi without running scripts, executing commands, or reading/writing files; the diagnostic flow is driven by role rules and a fixed workflow.
• Intake collection: The first turn asks for age, sex, main symptom, onset, duration, location, character, severity, major past history, and long-term medications.
• Three-question cycle: It asks one at a time about symptom details, past medical/medication history, and lifestyle or special circumstances, then waits for each answer.
• Report generation: It produces 1. Diagnosis, 2. Cause, 3. Medication advice, and 4. Daily care notes; medication entries should include generic name, purpose, dose/frequency, and contraindications.
• Risk gating: High-risk conditions such as MI, stroke, aortic dissection, malignancy, and acute pulmonary embolism must trigger a front-of-report ⚠️ warning with recommended department, tests, and reason.
• Cancer prompt: If the report mentions malignancy, leukemia, lymphoma, sarcoma, melanoma, or precancerous lesions, it must append the fixed follow-up/screener link prompt at the end.

Boundaries

It is suitable for initial triage, symptom clarification, medication hints, and follow-up advice, not for emergency care, imaging interpretation, pathology, or final prescribing. It depends on Purico’s Chiyou CFC inference service, retrieves an API Key by default, and declares access only to jiyinjia.jinbaisen.com and ydai.jinbaisen.com. For patient-facing or production use, evaluate medical compliance, privacy controls, API Key management, and human review workflows.