Cancer buddy vault

Skill CancerDAO/cancer-buddy-skill/skills/cancer-buddy-vault

为患者的本地肿瘤资料建立清单、访问政策、导出与审计流程。分享必须依赖宿主鉴权和患者的明确、可撤销、限目的授权；不把 patient_code 或亲属关系当权限。Triggers on 数据保险箱, 健康档案, 数据分享, 导出病历, 撤销授权.From its SKILL.md

Install

npx -y skills add CancerDAO/cancer-buddy-skill --skill cancer-buddy-vault

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

5 stars5 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

5.1 KB, as published. Nobody here has run it

cancer-buddy-vault

这是数据治理工作流，不是云存储、身份系统或法律授权替代品。详细规则见 references/data-vault.md。

Preconditions

宿主必须完成身份认证、授权校验和接收方确认；skill 不能自行发放 signed URL、密钥或访问权。
患者是默认控制者。照护者/家属访问需要可验证的权限范围、目的、期限和撤销状态。
patient_code 只是目录定位符；知道它不代表可访问。

Workflow

盘点资料类型、PII、临床敏感性、来源和现有访问策略。
所有对象默认 private。为每个授权记录 subject、recipient、scope、purpose、created_at、expires_at、revoked_at、legal/consent basis 和 host authorization reference。
分享前生成清单：具体文件、接收方、目的、期限、是否含原始影像/病理/遗传信息、撤销与留存方式。患者/授权代表逐项确认。
导出使用宿主提供的加密和密钥管理；密码/密钥通过独立安全通道传递。skill 不宣称本地 zip 等同合规传输。
每次查看、导出、修改和撤销写入不可追加修改的审计事件；并发更新使用版本检查。
撤销阻止未来访问，但诚实说明不能远程收回接收方已下载的副本；按机构政策请求删除并记录结果。

De-identification

“去掉姓名”不等于匿名。导出前评估日期、罕见病、机构、地理、自由文本、影像 DICOM、基因组和文件元数据的重识别风险。研究/跨境用途需要适用法域下的伦理、数据保护和人类遗传资源审查；skill 不自行判定合规。

保留原始临床字符串；翻译/规范化是带 provenance 的附加层。脱敏不得改写剂量、单位、病理或分子结论。原件与派生件分别标识。

Role behavior

Role = patient：在宿主认证后管理自己的授权。
Role = caregiver：仅限有效授权 scope；不能因曾被标为 caregiver 获得永久权限。
Role = family：无授权时只能获得一般流程说明，不能查看患者记录或匿名化视图。

披露偏好不能阻止一个有能力的患者在认证后访问自己的信息。涉及能力、法定代理或争议时，停止分享并交由医疗机构隐私/伦理/法律流程处理。

Disclosure

共享范围由当前 viewer 权限决定。family suppression 不能阻止有能力患者访问或导出自己的资料；也不能授权家属查看患者资料。每次导出独立确认 scope 和 recipient。

Outputs

vault-manifest.md
sharing-settings.json
access.log
经明确确认生成的导出包与 manifest

所有患者可见内容按 resolved locale 输出，同时保留源临床字符串与来源。遵循 ../../references/roles.md、../../references/i18n.md 和 ../../references/clinical-content-governance.md。

另见 ../../references/safety-guardrails.md 与 ../../references/disclosure-behavior.md。

Charting an indicator the user asked about

When the user asks about a specific named lab value or observation (CEA, 白蛋白, 体重…), check longitudinal_observations.json / labs.json for that analyte BEFORE answering in prose:

≥2 comparable points → answer in text and attach a chart: python3 ../cancer-buddy-charts/scripts/render_chart.py --chart trend --from-longitudinal <patient_dir>/longitudinal_observations.json --metric <analyte> --out-html <patient_dir>/charts/<analyte>_趋势.html
fewer than 2, or not comparable → answer in text and say in one line why there is no chart

Volunteering a chart covers only the analyte the user named. A general question ("我的化验单怎么样") does not auto-chart — list which indicators form a series and let them pick. When the user explicitly asks for several ("都画出来"), chart them all; there is no cap.

Answer the question, do not wall it off. What the indicator is, what it generally reflects, why reference ranges differ between hospitals, what guidelines generally say about follow-up intervals — all answerable (verify a current primary source at answer time and cite it; route to cancer-buddy-education). Only a verdict on this person's numbers — response, progression, whether to change regimen or add imaging, prognosis — routes to the treating team, in a sentence or two woven into the answer rather than a standing disclaimer block.

Keep implementation detail out of the reply: no script names, exit codes, rule numbers, or your own verification steps.