Health Management Assistant

Problem

Health assessment conversations often drift into free-form advice: users provide scattered symptoms, and the agent fills in conclusions on its own. That makes it hard to trace where a risk judgment came from. This skill separates the flow into intake, backend model invocation, and result passthrough, so the local assistant acts more like a data collection and presentation layer rather than directly generating medical conclusions.

How It Works

• Structured intake: It guides users to provide name, age, gender, region, occupation, medical history, family history, lifestyle, medication, allergy, and other relevant details. For female users, it can also ask about menstrual cycle, pregnancy planning, pregnancy, lactation, and other special states.
• Lab data input: Users can submit health check reports, lab sheets, or test result text. If the platform supports images, the skill can prompt users to upload images and extract text, while preserving as much indicator information as possible.
• Backend model call: The collected data and conversation context are sent to the cyzh-cfc model endpoint with streaming enabled, temperature set to 0.7, and max_tokens set to 2048. The API key is fetched dynamically from the configured endpoint and cached in-process, avoiding plaintext secrets in local config.
• Lossless presentation: The skill requires the full model response to be shown, including staged reasoning, Markdown tables, bold text, and risk markers, without summarizing, trimming, or reinterpreting the content.
• Optional HTML report: When the user explicitly requests it, the skill can render a report template with placeholders for screening links, hotline numbers, risk score, risk level, lab data tables, disease risk cards, and intervention suggestions.

Boundaries

This skill is best used as a health information organizer and report display entry point, not as a diagnostic tool. Model output is for reference only; urgent or critical conditions should be handled by seeking immediate medical care. If the backend API times out or fails, the assistant should tell the user to retry later instead of fabricating medical advice locally.