AgentSkillsCN

recommendations

基于 WebLLM 或 Gemini，针对患者的肤色定制生成可操作的临床建议（最长 25 字）。

SKILL.md
--- frontmatter
name: recommendations
description: Generates actionable clinical advice (max 25 words) calibrated for patient skin tone using WebLLM or Gemini
license: MIT
compatibility: opencode
metadata:
  audience: developers
  workflow: clinical-pipeline

What I do

I generate concise, actionable clinical recommendations based on the diagnosis, risk assessment, and web verification. I calibrate advice for the patient's specific skin tone and ensure recommendations are evidence-based.

When to use me

Use this when:

  • •Web verification is complete and you need clinical recommendations
  • •You need patient-appropriate, skin-tone-calibrated advice
  • •You're generating the final output for clinical decision support

Key Concepts

  • •25 Word Limit: Concise recommendations for easy comprehension
  • •Skin-Tone Calibration: Advice appropriate for Fitzpatrick type
  • •Evidence-Based: Grounded in verified web sources
  • •recommendations_generated: State flag after generation complete

Source Files

  • •services/vision.ts: Recommendation generation
  • •types.ts: AnalysisResult with recommendations array

Code Patterns

  • •Synthesize diagnosis, risk, and web evidence
  • •Calibrate language for patient demographics
  • •Generate concise, actionable advice

Operational Constraints

  • •Maximum 25 words per recommendation
  • •Must be calibrated to Fitzpatrick type
  • •Must clearly differentiate AI reasoning from web-grounded advice