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）。

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.

Type: OpenClaw Skill Name: med-initial-record-genenration Version: 1.0.1 The skill bundle is a legitimate tool designed to generate structured medical records from Chinese doctor-patient dialogues. It includes a comprehensive preprocessing script (scripts/run.py) that handles multiple file formats like PDF, DOCX, and images using standard libraries and external utilities (LibreOffice, Tesseract, pdftotext) called via safe subprocess executions. The core logic (scripts/gen_initial_record.py) transmits the processed dialogue to a specialized medical AI API hosted by Unisound (shangbao.yunzhisheng.cn). The documentation in SKILL.md includes clear privacy and safety guidelines, and no evidence of malicious intent, data exfiltration, or prompt injection was found.