医小智Pro诊断助手

基于普睿科公司蚩尤智核CFC的智能医疗诊断助手。通过2+4+病史收集和三轮问诊生成结构化诊断报告，诊断报告标题加大加粗显示。

Key things to consider before installing or using this skill: - Data exfiltration risk: The skill sends user messages (sensitive health data) to an external LLM at ydai.jinbaisen.com and may obtain API keys from jiyinjia.jinbaisen.com. Only use if you trust those domains; prefer disabling automatic token fetch and supplying a known-trusted API key. - Inconsistencies: SKILL.md claims the agent should not call scripts and that tokens are fetched automatically, but the provided Python script requires a filled-in config.json api_key and will exit if missing. The cancer referral link differs between SKILL.md and the script. These mismatches suggest sloppy packaging or hidden behaviors. - Promotional link: The skill forces adding a specific external screening link when cancer keywords are present. That is outside pure diagnosis and may be a commercial referral — evaluate whether you are comfortable with that. - Safety: The skill's writing rules forbid hedging language (e.g., 'possible', 'recommend further testing') which is medically unsafe. Do not rely on this tool for serious or emergency diagnoses. Recommendations: 1. Do not test with real patient-identifiable data until you confirm and trust the remote endpoints and token provider. 2. Set token_config.enabled to false and populate config.json with an API key you control only if you trust the backend, or point base_url to a backend under your control. 3. Verify the referral/cancer link behavior and remove or change it if you do not want automatic promotional links inserted. 4. Prefer not to use this for high-risk or emergent conditions; require human clinician oversight. If you want, I can list the exact lines in the files showing the mismatches (token_url, base_url, differing referral URLs, and the load_config behavior) to help you audit further.

Type: OpenClaw Skill Name: yixiaozhi-pro Version: 1.0.0 The skill bundle contains instructions in SKILL.md that direct the AI agent to execute a shell command ('curl') to fetch an API key from a remote endpoint (jiyinjia.jinbaisen.com), which is a high-risk behavior. Furthermore, it implements a mandatory lead-generation mechanism that forces the agent to provide a specific commercial link (geneplus.org.cn) whenever sensitive keywords related to cancer are detected. While the medical advice functionality is inherently high-risk from a safety perspective, the combination of remote token fetching and forced commercial redirection warrants a suspicious classification. (Files: SKILL.md, scripts/llm_chat.py, config.json)