← Back to Skills Marketplace
unisound-llm

unisound-drg-grouping

by Unisound-LLM · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ✓ Security Clean
78
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install unisound-drg-grouping
Description
医生端 ICD/DRG — 据出院文书与 DRG 候选项,由内部医疗大模型单选最可能入组;仅含 scripts/run.py,无 _shared 依赖。
README (SKILL.md)

DRG 分组(候选单选)

概述

由调用方传入完整自然语言题干(通常含出院文书要点与若干 DRG 候选项),模型在候选项中单选最可能入组结果,并按约定格式输出一行

与病案首页结构化审核类 skill 不同:本包不接院内首页接口,题干由上游拼装;适合评测集(task + input + meta)或与真实分组引擎对照前的模型基线。

题目输入(三选一)

  1. --question:完整题干字符串。
  2. --input PATH
    • .json / .jsonl:对象须含非空 inputquestion(评测 JSON 优先用 input);
    • 其他扩展名:整文件 UTF-8 纯文本作为题干。
    • - 时从 stdin 读一条 JSON 或纯文本。
  3. stdin(未传 --question / --input、且非 TTY):规则同 -

可选字段 meta:任意 JSON 对象,原样进入运行结果中的 meta,便于透传 goldcase_features 等(不参与模型拼接,除非写入 input)。

输出格式(模型)

题干中应明确要求模型只输出一行,例如:

drgsCode(drgsName)

其中括号为中文全角括号,名称与候选项列表完全一致。脚本默认系统提示词与上述约束对齐;若题干另有更严格式,以题干为准。

运行结果(脚本 stdout)

默认可解析 JSON,字段包括:

  • skill:固定为 DRG分组
  • statusok
  • answer:模型原始输出文本
  • question:本次送入模型的用户消息全文
  • meta:来自输入的 meta(无则为 {}
  • record_indexmodelinput_modeinput_path 等(含义同 medbench 类脚本)

用法示例

python3 scripts/run.py --input case.json --appkey YOUR_KEY
python3 scripts/run.py --question "……【DRG 候选】……" --appkey YOUR_KEY
python3 scripts/run.py --input case.jsonl --index 0 --batch 3 --appkey YOUR_KEY

参数

  • --question / --input:与 stdin 三选一(--question--input 不可同时使用)。
  • --appkey STRING必填--dry-run 除外)。内部医疗大模型鉴权 key。
  • --index--match-id--batch:仅对 .jsonl 生效(--match-id 匹配 meta.idother.id)。
  • --dry-run:不调用模型,仅输出解析后的题目 JSON。
  • --model--api-url--temperature--timeout--system-prompt:调模型参数。
  • --output PATH:另存完整 JSON(批量为 NDJSON)。
  • --text-only:标准输出仅打印 answer 文本。

模型

  • 默认 u1-insuremed,接口 https://maas-api.hivoice.cn/v1/chat/completions

合规说明

输出为分组辅助推断,不构成正式医保结算或院内分组结论;涉及真实患者数据须先脱敏并遵守院内流程。

Usage Guidance
Before installing or using this skill, confirm that sending the case text to the configured model API is allowed, de-identify real patient data when required, keep the app key private, and avoid overriding the API URL unless the endpoint is trusted and approved.
Capability Analysis
Type: OpenClaw Skill Name: unisound-drg-grouping Version: 1.0.0 The skill bundle is a legitimate tool for medical DRG (Diagnosis Related Group) grouping using the Unisound medical LLM API (maas-api.hivoice.cn). The Python script (scripts/run.py) is a standard API wrapper that processes local JSON/JSONL inputs and sends them to the model using an API key provided by the user. No malicious behaviors such as data exfiltration, unauthorized execution, or prompt injection were found.
Capability Assessment
Purpose & Capability
The purpose and code are coherent: the script reads a user-provided DRG question or input file, calls a medical LLM, and returns one answer. The notable issue is that the input may contain sensitive discharge or patient information.
Instruction Scope
The instructions mostly constrain output format, but the runtime options allow changing the API URL and system prompt, so the skill should be invoked with trusted parameters.
Install Mechanism
There is no install spec, package install, or remote script execution. However, internal package metadata does not fully match the submitted registry identity, so provenance is not perfect.
Credentials
Reading user-selected files/stdin, calling the stated model API, and optionally writing a user-selected output file are proportionate to the DRG grouping purpose, but the data can be medically sensitive.
Persistence & Privilege
No background persistence, privilege escalation, or autonomous mutation is shown. Persistence is limited to user-requested output files, and the app key is supplied at runtime.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install unisound-drg-grouping
  3. After installation, invoke the skill by name or use /unisound-drg-grouping
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
Initial release: provides DRG single-choice grouping for doctors based on discharge documents and candidate options. - Selects the most likely DRG from candidate options using an internal medical large language model. - Accepts input via question string, file, or stdin; supports JSON/JSONL/batch modes. - Outputs structured JSON including skill, status, answer, question, and metadata. - No dependency on _shared modules; contains only scripts/run.py. - Includes command-line arguments for model selection, API keys, and output control. - Designed for baseline evaluation or testing prior to integration with official systems; does not access internal hospital interfaces.
Metadata
Slug unisound-drg-grouping
Version 1.0.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is unisound-drg-grouping?

医生端 ICD/DRG — 据出院文书与 DRG 候选项,由内部医疗大模型单选最可能入组;仅含 scripts/run.py,无 _shared 依赖。 It is an AI Agent Skill for Claude Code / OpenClaw, with 78 downloads so far.

How do I install unisound-drg-grouping?

Run "/install unisound-drg-grouping" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is unisound-drg-grouping free?

Yes, unisound-drg-grouping is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does unisound-drg-grouping support?

unisound-drg-grouping is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created unisound-drg-grouping?

It is built and maintained by Unisound-LLM (@unisound-llm); the current version is v1.0.0.

💬 Comments