WiseDiag Medical Model Consultation

Problem It Solves

Medical Q&A is not just about producing an answer. The harder parts are staying in a clinical context, exposing a reviewable reasoning path, and fitting cleanly into an existing agent stack. This skill routes topics such as medicine, diagnosis, symptoms, drugs, dosages, interactions, treatment, anatomy, physiology, and public health to the WiseDiag medical model, reducing the risk that a generic assistant answers without clinical decomposition.

How It Works

The skill calls wisediag-large-latest through an OpenAI-compatible endpoint at https://openapi.wisediag.com/v1, with credentials commonly supplied via WISEDIAG_API_KEY. Because the API is stateless, the client must maintain the messages array and manage context length. In streaming mode, the response emits delta.reasoning_content before delta.content; simple queries may omit the reasoning field. It is reasonable to use stream: true, and to tune temperature, top_p, max_tokens, and seed for reproducibility and safety. The final usage object exposes prompt_tokens and completion_tokens for cost tracking.

Boundaries and Caveats

Prompts are transmitted to WiseDiag cloud servers for processing, so this skill is not suitable for unmasked medical records, patient identity data, private health details, or business secrets. For sensitive workloads, prefer a local or offline model. Multi-turn usage accumulates tokens, so the client should truncate, summarize, or reset the conversation. Deep thinking may add several seconds to over ten seconds of latency and may not appear for every request, so UI should treat reasoning and final answer as separate stages.