← Back to Skills Marketplace
unisound-llm

unisound-sm-doc

by Unisound-LLM · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ✓ Security Clean
45
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install unisound-sm-doc
Description
病历实体抽取与结构化。由调用方提供题目文本或含 `question` 字段的结构化输入,经内部医疗大模型按题干约束生成作答;仅含 `scripts/run.py`,可独立拷贝部署。
README (SKILL.md)

病历实体抽取与结构化

概述

从门诊或住院文书片段中抽取指定字段,按题干要求的 JSON 键名输出;缺失项按题干约定填「未知」等。

本能力与具体业务请求解耦:平台或调用方如何组织用户输入由集成方决定;此处只约定如何把题目交给脚本(见下)。

题目输入(三选一)

  1. --question:直接传入完整题干字符串。
  2. --input PATH
    • .jsonl:每行一个 JSON 对象,须含非空字符串字段 question;可用 --index / --batch / --match-id 选取记录;
    • .json:单个对象,须含 question
    • 其他扩展名:按 UTF-8 纯文本整文件作为题干。
    • - 时从 stdin 读取一条 JSON(含 question)或纯文本。
  3. stdin(且未传 --question / --input、且 stdin 非 TTY):读入 JSON 对象或纯文本,规则同 -

模型

  • 默认 u1-insuremed,接口 https://maas-api.hivoice.cn/v1/chat/completions(可通过参数覆盖)。

用法示例

# 命令行直接给题
python3 scripts/run.py --question "题干……" --appkey YOUR_KEY

# 从任意路径的 jsonl 取一条(与文件名无关)
python3 scripts/run.py --input /path/to/items.jsonl --index 0 --appkey YOUR_KEY

参数

  • --question / --input:见上,与 stdin 三选一(--question--input 不可同时使用)。
  • --appkey STRING必填--dry-run 时除外)。内部医疗大模型鉴权 key。
  • --index--match-id--batch:仅对 .jsonl 生效。
  • --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:标准输出仅打印模型答案文本。

输出

默认可解析 JSON:含 statusquestionanswerrecord_indexmetamodelinput_modeinput_path(来自文件时非空)等;skill 为本任务标题。

合规说明

输出为模型辅助信息,不构成正式诊疗决策;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, verify the publisher/provenance, use a scoped app key, keep the API URL trusted, and only process medical records that are authorized and preferably de-identified. The visible artifacts do not show malicious behavior.
Capability Analysis
Type: OpenClaw Skill Name: unisound-sm-doc Version: 1.0.0 The skill bundle is a legitimate tool for medical record entity extraction and structuring. The script `scripts/run.py` uses standard libraries to process local files or stdin and sends the data to a specified LLM API (maas-api.hivoice.cn). There is no evidence of data exfiltration, malicious execution, or prompt injection; the code is transparent and aligns with the documentation in `SKILL.md`.
Capability Assessment
Purpose & Capability
Purpose and implementation align: SKILL.md says it extracts structured fields from outpatient/inpatient document fragments, and scripts/run.py builds a user prompt and calls the configured medical model. This inherently may handle sensitive medical text.
Instruction Scope
Inputs are caller-directed through `--question`, `--input PATH`, or stdin, and the skill documents the external model endpoint and parameters. It also allows `--api-url` and `--system-prompt` overrides, so callers should keep these values trusted and intentional.
Install Mechanism
There is no install spec and the package is mostly a single Python script, but provenance is somewhat inconsistent: the registry slug/name differ from SKILL.md and _meta.json. No hidden installer or remote setup step is shown.
Credentials
The script reads only caller-selected input files/stdin and can optionally write caller-selected output files, which is proportionate for this task but can expose private records if broad paths or real patient data are used.
Persistence & Privilege
No background persistence or privilege escalation is evident. The main privilege is a caller-supplied `--appkey` used as a bearer token, and optional output files may persist the question, metadata, and answer.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install unisound-sm-doc
  3. After installation, invoke the skill by name or use /unisound-sm-doc
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
- Initial release of med-eval-sm-doc for medical record entity extraction and structuring. - Supports flexible input: command-line question, input file (json/jsonl/plain text), or stdin. - Outputs answers in a structured JSON format with fallback handling for missing fields. - Integrates with an internal medical large language model (default: u1-insuremed) via API. - Provides a standalone script (scripts/run.py) that can be independently deployed.
Metadata
Slug unisound-sm-doc
Version 1.0.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is unisound-sm-doc?

病历实体抽取与结构化。由调用方提供题目文本或含 `question` 字段的结构化输入,经内部医疗大模型按题干约束生成作答;仅含 `scripts/run.py`,可独立拷贝部署。 It is an AI Agent Skill for Claude Code / OpenClaw, with 45 downloads so far.

How do I install unisound-sm-doc?

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

Is unisound-sm-doc free?

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

Which platforms does unisound-sm-doc support?

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

Who created unisound-sm-doc?

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

💬 Comments