Skill flagged — suspicious patterns detected

ClawHub Security flagged this skill as suspicious. Review the scan results before using.

Medical Search

v1.0.0

Drug safety and medical information search. Use when user asks about: drug interactions, medication safety, contraindications, side effects, drug-alcohol int...

0· 98·1 current·1 all-time

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for juneyaooo/medical-search.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Medical Search" (juneyaooo/medical-search) from ClawHub.
Skill page: https://clawhub.ai/juneyaooo/medical-search
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Required binaries: curl
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install medical-search

ClawHub CLI

Package manager switcher

npx clawhub@latest install medical-search
Security Scan
VirusTotalVirusTotal
Suspicious
View report →
OpenClawOpenClaw
Suspicious
high confidence
!
Purpose & Capability
Name/description (drug safety search) aligns with using curl/web search and returning sources. However the SKILL.md only lists curl as a required binary while the runtime examples call python3 and a platform 'web_fetch' tool and also reference executing local scripts under /home/ubuntu/... — those are not declared. The requirement to always search and publish searched URLs does match the purpose, but the undeclared reliance on python3/web_fetch and optional execution of local scripts is disproportionate and inconsistent with the declared requirements.
!
Instruction Scope
Instructions mandate network queries to a hard-coded SearXNG instance at http://43.156.131.167:4000 (an external IP) and require returning exact URLs from results. They also instruct the agent to run local Python scripts at /home/ubuntu/github/openclaw-project/mediwise-health-tracker/... if present. Together these create two risks: (1) sensitive user queries (medical/PHI) will be sent to a third party by default, possibly leaking private data; (2) the agent may execute arbitrary local code if that path exists. The SKILL.md also references a 'web_fetch' tool that isn't declared in requires.bins, so the instructions rely on environment capabilities not stated in metadata.
Install Mechanism
No install spec and no code files — this is instruction-only, which minimizes disk writes. The primary runtime behavior is network I/O and invoking existing local binaries/tools. The lack of install spec itself is low-risk, but it shifts risk to the external endpoints and local execution described in the instructions.
!
Credentials
The skill does not request env variables or credentials (good), but it requires sending queries to a hard-coded external host and may invoke local scripts at a specific absolute path. Asking to execute /home/ubuntu/... is an implicit request for filesystem/execute access beyond simple search capability and is disproportionate for a search helper. Also the SKILL.md uses python3 and web_fetch without declaring them as required binaries/tools.
Persistence & Privilege
always is false and there is no install or self-modifying behavior. Autonomous invocation is allowed (platform default) but not combined with any elevated persistence or cross-skill configuration changes, so no extra privilege concerns from persistence.
What to consider before installing
This skill is coherent in purpose (it tries to force verification of drug safety via web search), but several red flags mean you should be careful before enabling it: - The skill hard-codes a third-party SearXNG instance at http://43.156.131.167:4000. Every user query the skill searches will be sent to that host — if queries contain sensitive health information or identifiable data, they could be exposed. Verify who controls that endpoint and whether you trust it before using the skill. - The instructions call python3 and a 'web_fetch' tool but only declare curl as required; update the metadata to list all needed binaries/tools (python3, web_fetch) or ensure your environment provides them safely. - The SKILL.md tells the agent to execute local scripts under /home/ubuntu/... if present. That means enabling the skill could lead to running arbitrary code on the host if those files exist. Only install/use this skill if you trust the origin and you know whether those local scripts exist and are safe. Consider removing or restricting that behavior. - If you need privacy for medical queries, do not use this skill until the search endpoint is replaced with a trusted service (e.g., a vetted official search API) or hosted locally under your control. Recommended actions before installing: 1. Ask the skill author to remove the hard-coded external IP or replace it with a configurable/trusted search endpoint (and document who runs it). 2. Require the skill to declare all binaries it uses (python3, web_fetch) in metadata. 3. Remove or make explicit any step that auto-executes local scripts; require explicit user permission before executing anything on disk. 4. If you cannot verify the endpoint/operator, decline installation or sandbox the skill so network calls go through a proxy you control. Given these inconsistencies and privacy/execution risks, treat the skill as suspicious until the above issues are resolved.

Like a lobster shell, security has layers — review code before you run it.

Runtime requirements

💊 Clawdis
Binscurl
latestvk97ak9m7as9f7d3ws3nbszja9h83r5x5
98downloads
0stars
1versions
Updated 1mo ago
v1.0.0
MIT-0

Medical Search - 药物安全与医药信息搜索

专门处理药物安全、用药禁忌、药物交互等医药健康信息搜索的技能。

核心原则(强制规则)

凡是涉及药物安全、药物交互、用药禁忌、不良反应、药物+酒精、药物+食物的问题,必须先搜索再回答,禁止仅凭训练数据回答。 这是硬性要求,不可跳过。

即使你对答案有信心,也必须搜索验证。药物安全信息错误可能危及生命。

查询方式选择

场景查询方式
两种西药之间的交互先查 DDInter(如可用),再用 web search 补充
中成药相关问题直接用 web search(DDInter 不收录中成药)
药物+酒精/食物直接用 web search(DDInter 不覆盖)
药品说明书/禁忌/不良反应直接用 web search
孕妇/哺乳期/儿童用药安全直接用 web search
药物副作用/不良反应直接用 web search

搜索方法

方式一:SearXNG Web Search(主要方式)

使用自托管 SearXNG 实例搜索医药信息。

# 基础搜索(中文医药问题推荐加 language=zh)
curl -s "http://43.156.131.167:4000/search?q=QUERY&format=json&language=zh" | python3 -c "
import sys, json
data = json.load(sys.stdin)
for i, r in enumerate(data.get('results', [])[:10], 1):
    print(f'{i}. {r.get(\"title\", \"No title\")}')
    print(f'   URL: {r.get(\"url\", \"\")}')
    print(f'   {r.get(\"content\", \"No description\")[:200]}')
    print()
"

搜索关键词构造建议:

  • 药物交互问题:"药名A 药名B 药物相互作用"、"药名A 药名B 能一起吃吗"
  • 药物+酒精:"药名 喝酒 禁忌"、"药名 酒精 相互作用"
  • 药物+食物:"药名 食物禁忌"、"药名 饮食注意事项"
  • 药物副作用:"药名 不良反应"、"药名 副作用"
  • 用药安全:"药名 禁忌症"、"药名 注意事项"、"药名 说明书"
  • 特殊人群:"药名 孕妇 能吃吗"、"药名 哺乳期"、"药名 儿童用药"

推荐搜索示例:

# 药物+酒精
curl -s "http://43.156.131.167:4000/search?q=感冒滴丸+喝酒+禁忌&format=json&language=zh"

# 药物交互
curl -s "http://43.156.131.167:4000/search?q=阿司匹林+布洛芬+能一起吃吗&format=json&language=zh"

# 药物副作用
curl -s "http://43.156.131.167:4000/search?q=奥美拉唑+不良反应+副作用&format=json&language=zh"

# 中成药安全
curl -s "http://43.156.131.167:4000/search?q=藿香正气水+禁忌+注意事项&format=json&language=zh"

方式二:DDInter 药物交互数据库(可选,仅限西药交互)

如果用户已配置 mediwise-health-tracker,可调用其 DDInter 脚本查询结构化的西药交互数据:

# 查询两种西药之间的交互
python3 /home/ubuntu/github/openclaw-project/mediwise-health-tracker/scripts/drug_interaction.py check-pair --drug-a "阿司匹林" --drug-b "华法林"

# 搜索药品(中英文)
python3 /home/ubuntu/github/openclaw-project/mediwise-health-tracker/scripts/drug_interaction.py search --name "布洛芬"

# 查看药品交互概览
python3 /home/ubuntu/github/openclaw-project/mediwise-health-tracker/scripts/drug_interaction.py lookup --name "奥美拉唑"

注意: DDInter 仅覆盖西药之间的交互,不覆盖中成药、药物+酒精、药物+食物等场景。即使 DDInter 返回了结果,仍建议用 web search 补充验证。

推荐信息源(按优先级)

搜索结果中优先采信以下来源的内容:

  1. 百度健康 (health.baidu.com) - 药品说明书、禁忌症、不良反应,内容较全
  2. 好大夫在线 (haodf.com) - 医生科普文章,用药注意事项
  3. 用药助手/丁香园 (drugs.dxy.cn) - 专业药品信息
  4. 百度百科 医学词条 - 药品基础信息
  5. 国家药品监督管理局 (nmpa.gov.cn) - 官方药品信息
  6. 医脉通 (medlive.cn) - 专业医学资讯

搜到感兴趣的结果后,可用 web_fetch 工具抓取页面详情。

结果呈现(强制要求)

回复模板(必须遵循)

回复必须包含以下结构,缺一不可:

1. 【直接回答】— 一句话回答用户问题(能/不能/需谨慎)
2. 【搜索发现】— 告诉用户你搜索了什么、找到了哪些相关信息,摘要 2-3 条最相关的搜索结果
3. 【详细分析】— 基于搜索结果给出具体的风险分析和建议
4. 【参考来源】— 列出 2-3 个搜索到的具体 URL 链接(从搜索结果中提取)
5. 【免责声明】— 固定文本

示例回复:

**不建议。** 服用感冒滴丸期间不要饮酒。

**搜到以下相关信息:**

- 药智新闻报道:酒精在体内代谢为乙醛,部分感冒药会抑制乙醛代谢,造成蓄积中毒(来源:https://news.yaozh.com/archive/30637.html)
- 39健康网:感冒药含对乙酰氨基酚,与酒精同服会增加肝毒性风险(来源:http://askar.39.net/a/230714/k490n3e.html)
- 健康之路:滴丸类药物服用期间饮酒会刺激加重症状(来源:https://m.jkzl.com/...)

**具体风险分析:**
...(基于以上搜索结果展开)

**参考来源:**
1. 药智新闻 https://news.yaozh.com/archive/30637.html
2. 39健康网 http://askar.39.net/a/230714/k490n3e.html
3. 健康之路 https://m.jkzl.com/...

⚕️ 以上信息来自公开医药数据库和医学百科,仅供参考,不构成医疗建议。具体用药请咨询医生或药师。

关键要求

  • 必须展示搜索到的 URL:从搜索结果中挑选 2-3 个最相关的链接,直接展示给用户
  • 必须告诉用户搜了什么:让用户知道你确实查过资料,不是凭空回答的
  • URL 必须是搜索结果中实际返回的:不要编造 URL,直接用搜索结果里的 url 字段

风险等级提示

  • 严重风险:明确警告,建议立即咨询医生,不要自行合用
  • 中等风险:提示注意,建议咨询药师确认是否需要调整
  • 轻微风险:告知存在轻微交互,一般无需特殊处理
  • 无明确风险:告知未发现明确禁忌,但仍建议遵医嘱

来源引用(强制要求)

每次药物安全相关回复必须包含具体的参考来源:

  • 如果是 DDInter 查询结果 -> 注明"来源:DDInter 药物相互作用数据库"
  • 如果是 web search 查询结果 -> 必须附带搜索到的具体 URL 链接,例如"参考来源:百度健康 https://health.baidu.com/..."
  • 禁止只给出泛泛的建议而不附带任何来源

免责声明(强制要求)

每次回复末尾必须附带以下免责声明:

以上信息来自公开医药数据库和医学百科,仅供参考,不构成医疗建议。具体用药请咨询医生或药师。

查不到数据时

提示"该药物/问题未查到可靠的参考信息,建议咨询药师或医生确认",不要沉默跳过,也不要编造信息。

完整工作流示例

用户问:"感冒滴丸吃了能喝酒吗?"

  1. 识别意图:药物+酒精安全问题 -> 必须搜索
  2. 执行搜索
    curl -s "http://43.156.131.167:4000/search?q=感冒滴丸+喝酒+禁忌&format=json&language=zh" | python3 -c "
    import sys, json
    data = json.load(sys.stdin)
    for i, r in enumerate(data.get('results', [])[:10], 1):
        print(f'{i}. {r.get(\"title\", \"No title\")}')
        print(f'   URL: {r.get(\"url\", \"\")}')
        print(f'   {r.get(\"content\", \"No description\")[:200]}')
        print()
    "
    
  3. 抓取详情(如有需要):用 web_fetch 获取排名靠前的页面内容
  4. 整理回复:综合搜索结果,给出明确建议 + 来源链接 + 免责声明

反模式

  • 不要仅凭训练数据回答药物安全问题 -> 必须先搜索
  • 不要省略来源链接 -> 每次回复必须有参考来源
  • 不要省略免责声明 -> 每次回复必须有免责提示
  • 不要编造药物信息 -> 查不到就说查不到
  • 不要将搜索结果作为确定性医疗建议呈现 -> 始终注明"仅供参考"
  • 回复用户时使用中文,不要用英文回复中文用户

Comments

Loading comments...