← Back to Skills Marketplace
sunlinlin-aragon

ai-medical-care-manager

by sunlinlin-aragon · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ⚠ suspicious
193
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install ai-medical-care-manager
Description
面向C端门诊就医全流程。先做症状分流和挂号科室判断,再推荐医院/医生 Top 3,并继续完成挂号引导、就医准备卡、提醒、诊后解释,以及基于高德地图的到院路线规划。
README (SKILL.md)

AI就医管家

当用户需要完成一次完整门诊就医任务,而不只是问“挂什么科”时,使用这个 skill。

这个 skill 的目标不是替代医生诊断,而是把一次就医任务拆成三个阶段并带用户走完:

  • 诊前:判断风险、推荐科室、推荐医院/医生、指导挂号
  • 诊中:解析挂号信息、生成就医准备卡、生成提醒、规划路线
  • 诊后:解释病历/处方/报告、提炼待办、提示复诊

何时使用

适合这些请求:

  • “我哪里不舒服,挂什么科?”
  • “帮我推荐医院和医生。”
  • “我已经挂好号了,帮我看看要准备什么。”
  • “帮我做就医提醒和路线。”
  • “我看完病了,帮我解释处方/报告。”
  • “帮我把这次看病的下一步待办整理出来。”

工作原则

  1. 先分阶段再行动:先判断用户处在诊前、诊中还是诊后。
  2. 先安全再推荐:任何高危情形优先急诊,不继续普通门诊推荐。
  3. 先结论再理由:先给用户下一步怎么做,再补理由。
  4. 先最小闭环再扩展:优先解决“这次看病怎么顺利完成”,不要一次堆太多边缘能力。
  5. 不替代医生诊断:只能做辅助分流、流程协助和通俗解释。

开始前先快速想清楚三件事:

  • 用户现在最需要解决的,是“判断”“执行”还是“理解”?
  • 当前最可能卡住的环节在哪一步?
  • 我这次回答里,最具体可执行的下一步是什么?

参考流程说明:{baseDir}/references/flow_playbook.md

内置资源

  • 医院数据:{baseDir}/assets/hospital_extracted_final.csv
  • 分诊与推荐:{baseDir}/scripts/triage_and_match.py
  • 挂号文本解析:{baseDir}/scripts/parse_appointment_text.py
  • 就医准备卡:{baseDir}/scripts/generate_previsit_card.py
  • 提醒生成:{baseDir}/scripts/appointment_reminders.py
  • 高德 IP 粗定位:{baseDir}/scripts/amap_ip_locate.js
  • 高德地址转坐标:{baseDir}/scripts/amap_geocode.js
  • 高德路线规划与 Web 跳转:{baseDir}/scripts/amap_route_link.js
  • 急症规则:{baseDir}/references/triage_rules.md
  • 输出模板:{baseDir}/references/response_templates.md

诊前:分流、科室判断、推荐 Top 3

第一步:收集最少必要信息

优先收集:

  • 主诉与持续时间
  • 伴随症状
  • 年龄、性别
  • 既往史/慢病/近期用药/妊娠情况
  • 想就诊的城市(默认可按北京处理)

若信息不全,也可以先初步判断,但要明确不确定性。

第二步:先做安全分流

先阅读 {baseDir}/references/triage_rules.md。 若存在明显急症信号,不要继续普通门诊推荐;直接建议急诊/120。

第三步:运行分诊与推荐脚本

python3 {baseDir}/scripts/triage_and_match.py \
  --csv {baseDir}/assets/hospital_extracted_final.csv \
  --symptoms "用户主诉与伴随症状" \
  --history "既往史或慢病,可为空" \
  --age "年龄,可为空" \
  --gender "性别,可为空" \
  --top-k 3

脚本会返回:

  • emergency_flag
  • department_candidates
  • top_matches

第四步:组织结果

最终答复中要包含:

  • 风险判断
  • 推荐科室(主推荐 + 备选)
  • Top 3 医院/科室/医生
  • 推荐理由
  • 挂号方式

输出时参考:{baseDir}/references/response_templates.md

第五步:固定给出挂号方式

默认给出:

微信内挂号更方便:

  • 方式 1:搜索“北京114预约挂号”公众号
  • 方式 2:搜索“京通”小程序 → 健康服务 → 预约挂号114

电话方式:

  • 拨打 010-114 挂号

并提示用户:挂完号后把截图或文本发上来,我会继续帮你做准备卡、提醒和路线。

诊中:准备卡、提醒、路线

第一步:解析挂号文本

当用户上传挂号截图 OCR 文本或直接贴出挂号文本时,运行:

python3 {baseDir}/scripts/parse_appointment_text.py \
  --csv {baseDir}/assets/hospital_extracted_final.csv \
  --text "挂号截图OCR文本或用户粘贴内容"

若字段缺失,继续追问医院、科室、医生、时间中的缺项。

第二步:生成就医准备卡

python3 {baseDir}/scripts/generate_previsit_card.py \
  --hospital "医院名" \
  --department "科室名" \
  --doctor "医生名,可缺省" \
  --appointment "2026-03-20 14:30" \
  --symptoms "本次主诉摘要" \
  --history "病史摘要,可为空" \
  --city "北京"

把输出整理成用户易读的“就医准备卡”:

  • 医院 / 科室 / 医生 / 时间
  • 建议到达时间
  • 需携带资料
  • 这次建议问医生什么
  • 哪些病史别漏说

第三步:生成提醒

python3 {baseDir}/scripts/appointment_reminders.py --appointment "2026-03-20 14:30"

默认列出三次提醒:T-12h、T-6h、T-2h。 如果当前环境支持闹钟/提醒工具,再在得到用户确认后创建提醒;否则明确列出时间并建议用户设置手机闹钟。

第四步:就医路线规划(高德地图)

仅在已配置 AMAP_WEBSERVICE_KEY 时执行。未配置时,给出手动高德搜索建议。

4.1 先拿到起点

优先顺序:

  1. 如果运行环境或用户上下文中能拿到真实用户 IP,则先尝试粗定位:
node {baseDir}/scripts/amap_ip_locate.js --ip="用户IP"

注意:IP 定位通常只有城市 / 区域级别,只能作为起点猜测。如果结果不够精确,必须继续向用户确认具体出发位置。

  1. 如果没有用户 IP,或 IP 无法定位,直接让用户输入当前位置,例如:
  • 我现在在朝阳大悦城
  • 我从北京西站出发
  • 我在望京 SOHO

4.2 再把起终点转成坐标

对用户起点和医院终点分别执行地址转坐标:

node {baseDir}/scripts/amap_geocode.js --address="用户当前位置描述" --city="北京"
node {baseDir}/scripts/amap_geocode.js --address="医院名称或地址" --city="北京"

脚本返回字段包括:

  • location
  • lng
  • lat

注意:高德坐标顺序是 经度,纬度

4.3 再做路线规划并生成 Web 跳转链接

node {baseDir}/scripts/amap_route_link.js \
  --mode=driving \
  --origin="116.397428,39.90923" \
  --destination="116.427281,39.903719" \
  --originName="用户当前位置" \
  --destName="医院名称" \
  --city="北京"

支持的 mode

  • driving
  • walking
  • riding
  • transfer

输出时给:

  • 预计距离
  • 预计耗时
  • 推荐出行方式
  • 可点击的 amap_link

路线规划的建议话术

  • 如果用户要尽快到院,优先 driving
  • 如果距离短且医院周边停车不便,可给 walkingriding
  • 如果用户明确想坐公共交通,使用 transfer

诊后:解释、待办、复诊

这一阶段主要依靠模型来做通俗解释,不要求额外脚本。

当用户上传病历、处方、检查报告或 OCR 文本时:

  1. 先用普通人能理解的话解释这次医生大意。
  2. 再总结“今天医生让我做的 3 件事”。
  3. 再补充“哪些情况需要尽快复诊 / 复查”。
  4. 如果文本里出现明确复查时间,再建议用户设置提醒。

解释时重点覆盖:

  • 诊断是什么意思
  • 药怎么吃
  • 检查结果重点是什么
  • 接下来要做什么

必须强调:

  • 解释仅供理解,不替代医生最终意见
  • 不建议用户自行停药、换药、延误复诊

推荐输出风格

每次答复尽量遵守:

  • 先给结论
  • 再给理由
  • 最后给下一步行动

推荐输出结构:

  1. 你的当前判断
  2. 推荐科室 / 推荐对象
  3. 你接下来该做什么
  4. 我还能继续帮你什么

不该做的事

  • 不要给出确定性的疾病诊断
  • 不要在高危症状下继续普通门诊推荐
  • 不要让用户自己去消化一大段复杂说明
  • 不要只给知识,不给可执行下一步

技能安装与放置

将此 skill 放到以下任一目录:

  • \x3Cworkspace>/skills/ai-medical-care-manager
  • ~/.openclaw/skills/ai-medical-care-manager

如果要启用高德路线规划,请在 ~/.openclaw/openclaw.json 中给该 skill 配置:

{
  "skills": {
    "entries": {
      "ai-medical-care-manager": {
        "enabled": true,
        "env": {
          "AMAP_WEBSERVICE_KEY": "你的高德 Web Service Key",
          "AMAP_KEY": "你的高德 Web Service Key"
        }
      }
    }
  }
}
Usage Guidance
This skill appears to do what it says: run local Python/Node scripts to triage symptoms, parse appointment text, produce a pre-visit card and reminders, and — if you supply an AMap key — call AMap web services to geocode and plan routes. Before installing, consider: (1) Privacy: any location string or IP you supply will be sent to AMap (restapi.amap.com) and generated map links embed route data in a public AMap demo URL — avoid sending extremely sensitive personal data to external map services. (2) Key storage: if you provide a Web Service Key via the skill helper, it may be saved to scripts/vendor/config.json in the skill folder; treat that file as sensitive. (3) IP suggestion: the skill only uses IP if you explicitly provide it; do not pass real user IPs if you do not want coarse location shared. (4) No other external credentials are requested and no hidden network hosts or obfuscated code were found. If you need stronger guarantees, ask the author to (a) declare AMAP_* env vars in the registry metadata as optional, (b) document where keys are persisted and allow opting out of on-disk storage, and (c) confirm the a.amap.com map visualization URL's use of query data.
Capability Analysis
Type: OpenClaw Skill Name: ai-medical-care-manager Version: 1.0.0 The skill facilitates medical triage and navigation, which involves handling high-risk sensitive data including medical symptoms (PHI) and user IP/location tracking (PII). Key indicators include scripts that attempt to automatically retrieve user IP addresses from the environment (amap_ip_locate.js) and a utility that writes configuration data to the local filesystem (amap_index.js). While these capabilities are aligned with the stated purpose of hospital routing and care management, the automated collection of network identifiers and the processing of health data via external APIs (restapi.amap.com) represent significant privacy and security risks that require manual oversight.
Capability Assessment
Purpose & Capability
Name/description match what is implemented: Python scripts perform triage, parsing, reminders, and pre-visit card generation; Node scripts handle AMap geocoding, IP locate and route link generation. Required binaries (python3, node) and axios dependency are expected for these tasks. The large CSV asset is appropriate for hospital/doctor matching.
Instruction Scope
SKILL.md confines runtime actions to the skill's files (scripts and references) and to calling AMap web services for mapping. It instructs optionally attempting IP-based coarse location if the agent can obtain a user IP — the node ip-locate script requires an explicit --ip argument, so there is no hidden automatic network probing. Be aware that if you provide user location/address or IP, those values are sent to AMap endpoints (external service) for geocoding/route planning.
Install Mechanism
Install spec only pulls in the npm package 'axios' (declared in package.json) — a minimal and expected runtime HTTP client. There are no downloads from arbitrary URLs or archives; no high-risk install actions detected.
Credentials
The skill does not declare required env vars in registry metadata, but the code and SKILL.md use optional AMAP_WEBSERVICE_KEY / AMAP_KEY for routing. Requesting an AMap Web Service key is proportionate to offering route planning. One thing to note: the vendor code can persist a provided key to a local config file (scripts/vendor/config.json), which stores the key on disk; this is a design choice (convenience) but worth knowing for secret handling.
Persistence & Privilege
The skill does not request always: true, does not modify other skills, and only persists its own config (vendor/config.json) if the developer helper functions are used. Files written are limited to the skill directory and are within expected scope.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install ai-medical-care-manager
  3. After installation, invoke the skill by name or use /ai-medical-care-manager
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
AI就医管家 1.0.0 - 新增面向C端门诊全流程的AI就医管家技能,从分诊判断到就医路线规划全覆盖 - 支持诊前(分流、科室判断、医院/医生推荐)、诊中(挂号解析、准备卡、提醒、路线推荐)、诊后(病历/处方解释、待办提炼、复诊提醒)全流程任务 - 集成高德地图API,实现到院路线规划及Web跳转 - 明确分阶段协作原则,优先安全分流与可执行闭环 - 提供多种挂号方式、就医材料准备建议、多轮追问与下步指引
Metadata
Slug ai-medical-care-manager
Version 1.0.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is ai-medical-care-manager?

面向C端门诊就医全流程。先做症状分流和挂号科室判断,再推荐医院/医生 Top 3,并继续完成挂号引导、就医准备卡、提醒、诊后解释,以及基于高德地图的到院路线规划。 It is an AI Agent Skill for Claude Code / OpenClaw, with 193 downloads so far.

How do I install ai-medical-care-manager?

Run "/install ai-medical-care-manager" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is ai-medical-care-manager free?

Yes, ai-medical-care-manager is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does ai-medical-care-manager support?

ai-medical-care-manager is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created ai-medical-care-manager?

It is built and maintained by sunlinlin-aragon (@sunlinlin-aragon); the current version is v1.0.0.

💬 Comments