med-initial-record-genenration

从中文医患对话文本生成门诊初诊病历，输出结构化分段的病历正文（文本）。

Security Scan

The skill's stated privacy guarantees (automatic de‑identification, no local persistence) contradict the shipped code: the code posts dialogue text to an external API and writes output files, with no visible redaction step.

Purpose & Capability

Name/description (generate outpatient initial record from Chinese doctor–patient dialogue) match the code and instructions: the skill preprocesses various input formats and calls a backend record-generation API to produce structured medical-record text.

Instruction Scope

SKILL.md promises strict de‑identification and that inputs/intermediate data won't be persisted. The gen_initial_record.py implementation reads the input file and sends the dialogue as the 'diag' field to an external endpoint (https://shangbao.yunzhisheng.cn/skills/record-gen/gen_record_by_diag_v1) with no visible PII-redaction step. The scripts do write output files to disk (and offer an option to save preprocessed text), which contradicts the 'no local persistence' assertion.

Install Mechanism

No install spec; optional Python packages (openpyxl, pypdf) and external CLI tools (soffice, pdftotext, tesseract) are documented and reasonable for the declared input formats. No downloads from untrusted URLs or archive extraction are present.

Credentials

The skill requests no environment variables or credentials, which is consistent with the code. However, it sends (potentially sensitive) dialogue content to an external, unauthenticated HTTPS API; the lack of auth means confidential data could be transmitted to that service without further controls. SKILL.md's assurances about encryption and limited use are declarative and not enforced by code.

Persistence & Privilege

The skill is not always-enabled and does not request special platform privileges. Still, it writes outputs to disk and optionally can save preprocessed text; this contradicts the SKILL.md claim of 'no local persistence' for inputs/intermediates. There is no code that shows persistent storage of credentials or modifying other skills, but the mismatch in claimed vs actual persistence is notable.

What to consider before installing

Key points before installing or using this skill: - The README promises automatic de‑identification and that input/intermediate data won't be persisted, but the shipped code posts the dialogue text to an external HTTPS API and writes output files to disk. There is no obvious redaction step in gen_initial_record.py. - If you plan to use real patient data (PHI), do not assume the skill will protect identities. Either sanitize/anonymize the dialogue yourself before running the skill, or avoid using this skill for real patient data until the developer provides verifiable redaction logic. - The external endpoint (shangbao.yunzhisheng.cn) is unauthenticated: data will be sent over HTTPS but without an API key or per‑tenant access control. Ask the maintainer who operates that endpoint and what their data retention/policy is. - If you need stronger guarantees, request source changes: add explicit PII-redaction (names, ID numbers, phone numbers, addresses), avoid sending raw dialogues, and add an option to run fully offline or against a local model/service. - As a quick mitigation, run the tool on synthetic or redacted test data only, or disable network access for the process (or firewall the host) until you verify sanitization and the third party’s privacy commitments. If you want, I can: - Search the remaining run.py content to verify whether any redaction logic exists and where it runs (preprocessing vs. before calling the API). - Suggest specific redaction patterns or code changes to implement safe PII removal prior to upload.

门诊初诊病历生成

概述

给定一份 中文医患对话文本（通常来自 ASR 转写），本技能生成一份门诊 初诊病历 文本，包含常见病历分段（如主诉、现病史、既往史、查体、辅助检查、诊断、处理等）。

常见输出分段：

• 主诉
• 现病史
• 既往史
• 月经史（如适用）
• 查体
• 辅助检查
• 诊断
• 处理

数据安全、隐私与伦理声明

• 最小必要原则：仅处理生成病历所必需的对话内容；不要求也不鼓励包含姓名、证件号、手机号、详细地址等身份信息。
• 严格脱敏：在发送至任何模型/接口前，会对可识别个人身份的信息进行脱敏/去标识化处理（如姓名、证件号、手机号、详细地址、人脸/影像等）。仅传递脱敏后的必要信息用于本次 skill 调用。
• 不做本地持久化：不将用户输入与中间结果写入本地持久化存储（包含磁盘文件、数据库、日志）。仅在内存中短暂处理；本次调用结束即销毁。
• 第三方 API 风险提示：在功能需要时，可能会调用第三方模型/服务接口；此时仅会发送脱敏后的必要信息，并使用加密传输。除完成本次请求外，不用于任何其他用途（如训练、画像、营销）。
• 医疗边界：本技能用于病历文本整理与结构化表达的辅助生成，不构成医疗诊断或治疗建议；请由执业医生复核并承担最终医疗责任。

输入格式

纯文本对话（UTF-8），建议一行一句/一轮，例如：

患者：…… 医生：…… 患者：…… 医生：……

也支持通过统一入口 scripts/run.py 直接输入 pdf/doc/docx/xls/xlsx/csv/txt/json。 预处理成功后，会先归一化为标准医患对话文本，再调用本 skill 的原始生成逻辑。

快速开始

# 从 skills 目录运行
python3 scripts/run.py \
  --input data/med-initial-record-gen/dialogue.txt

# 或继续直接使用原始文本入口
python3 scripts/gen_initial_record.py \
  --input data/med-initial-record-gen/dialogue.txt

参数说明

• --input PATH
  • 输入对话文本路径（UTF-8）。
• --output PATH
  • 输出病历路径（默认：../runs/med-initial-record-gen/record.txt）。
• --diag-id STRING
  • 对话 ID（默认：skill-diag）。
• --timeout SECONDS
  • 超时秒数；0 表示一直等待（默认：0）。

统一入口附加参数（scripts/run.py）

• --input-type auto|pdf|doc|docx|xls|xlsx|csv|txt|json
  • 输入类型；默认 auto。
• --sheet STRING
  • 读取 Excel 时指定 sheet（可选）。
• --encoding STRING
  • txt/csv 编码（默认：utf-8）。
• --save-prepared
  • 保存预处理后的标准对话文本，便于调试。

输出约定

• 若输出路径父目录不存在，会自动创建。
• 输出为 UTF-8 文本，包含常见门诊初诊病历分段。

依赖

运行环境

• Python 3.7+

外部 API

• 后端病历生成服务：https://shangbao.yunzhisheng.cn/skills/record-gen/gen_record_by_diag_v1
  • 方法：POST，Content-Type: application/json
  • 需要网络访问 shangbao.yunzhisheng.cn

Python 第三方包（可选，按输入格式需要）

包名	用途	必要条件
openpyxl	读取 .xlsx 文件	输入为 xlsx 时必须
pypdf	提取 PDF 文本	输入为 pdf 时必须（或用 pdftotext 替代）

安装：pip install openpyxl pypdf

外部工具（可选，按输入格式需要）

工具	用途	必要条件
LibreOffice (soffice)	转换 .doc / .xls 为文本	输入为 doc/xls 时必须
pdftotext（poppler-utils）	提取 PDF 文本	输入为 pdf 且未安装 pypdf 时必须
tesseract（含 chi_sim+eng 语言包）	图片 OCR	输入为 png/jpg/bmp/tif 等图片时必须

安装（Ubuntu/Debian）：sudo apt-get install libreoffice poppler-utils tesseract-ocr tesseract-ocr-chi-sim

仅使用 TXT/JSON 输入时，无需安装任何第三方包或外部工具。

测试命令

从 skills 目录执行（网络自测）：

python3 self_tests/med-initial-record-gen/self_test_initial_record_gen.py --run-network