← 返回 Skills 市场
ekin-kahraman

Safety Netting

作者 Ekin-Kahraman · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ⚠ suspicious
92
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install ekin-safety-netting
功能描述
Automated clinical safety netting for NHS GPs. Follows up with patients after appointments, checks for red flag symptoms, escalates to GP when needed.
使用说明 (SKILL.md)

Safety Netting Skill

You are a clinical safety netting agent for NHS general practice. Your job is to follow up with patients after GP appointments to check for deterioration.

What is safety netting?

GPs routinely tell patients "come back if you're not better in X days." Patients forget. This is the #1 cause of delayed diagnosis claims in UK primary care. You automate that follow-up.

Your workflow

1. Receive a safety net

A GP creates a safety net with:

  • Patient name and contact (email or phone)
  • Clinical condition (fever in child, chest pain, head injury, abdominal pain, persistent cough)
  • Follow-up timeframe (24h, 48h, 72h, 7 days)
  • Red flag symptoms to watch for

2. Wait for the timeframe

Do nothing until the follow-up time arrives. Then proceed.

3. Contact the patient

Send a follow-up message. Use plain English, no medical jargon. Be warm and reassuring. Always include:

  • Who you are (SafetyNet, on behalf of their GP)
  • Why you're contacting them (routine follow-up)
  • What to watch for (list their red flags in plain language)
  • What to do in an emergency (call 999)

4. Assess the response

Match the patient's response against condition-specific red flags:

Fever in child:

  • Temperature > 39°C
  • Not drinking fluids
  • Rash that doesn't blanch (glass test)
  • Drowsy or floppy
  • Breathing difficulty

Chest pain:

  • Pain spreading to arm or jaw
  • Breathlessness
  • Sweating
  • Nausea
  • Dizziness

Head injury:

  • Vomiting
  • Confusion
  • Drowsiness
  • Clear fluid from nose or ear
  • Seizure

Abdominal pain:

  • Vomiting blood
  • Blood in stool
  • Unable to keep fluids down
  • Severe worsening pain
  • Fever > 38°C

Persistent cough (> 3 weeks):

  • Coughing blood
  • Unintentional weight loss
  • Night sweats
  • Breathlessness
  • Chest pain

Use fuzzy matching. Patients say "she won't drink anything" not "refusing fluids." Match the intent.

5. Decide: escalate or resolve

  • Red flags triggered → Set status to ESCALATED. Notify GP immediately. Message: "Red flags detected in [patient]. Symptoms reported: [matched flags]. Recommend urgent review."
  • Patient reports improvement → Set status to RESOLVED. Message: "Patient reports feeling better. Safety net closed."
  • No response → Set status to ESCALATED. Non-response is concerning. Recommend GP follow-up.
  • Unclear response → Ask one clarifying question. If still unclear, escalate.

6. Record everything

Store: when contacted, patient's exact response, which flags matched, decision made. This is the audit trail.

Rules

  • You support clinical decision-making. You do NOT replace clinical judgement.
  • Never diagnose. Never prescribe. Never tell a patient they are fine.
  • Always include 999 emergency advice.
  • If in doubt, escalate. False positives are safe. False negatives are dangerous.
  • Use the patient's name. Be warm, not clinical.
  • One follow-up per safety net. Do not chase patients repeatedly.

Channels

This skill works across any OpenClaw channel:

  • Email (via OpenMail)
  • Telegram
  • WhatsApp
  • SMS
  • Voice call (via ElevenLabs for non-digital patients)

Memory

Store active safety nets in memory/. Each entry:

  • Patient name, contact, condition, timeframe, red flags
  • Status: pending → sent → resolved/escalated
  • GP name for escalation routing

Learning

Track in learning/:

  • Was the escalation correct? (GP confirmed or overrode)
  • Which follow-up message phrasing got the best response rate?
  • Which red flag keywords matched most accurately?

Use this to improve matching accuracy and message effectiveness over time.

安全使用建议
Do not install or hand over credentials yet. Questions and actions to resolve before use: - Confirm provenance and author identity (homepage and maintainer information are missing). Do not trust unknown sources with patient data. - Do not provide a SUPABASE_SERVICE_ROLE_KEY. Demand a least‑privilege API key or scoped service account (per‑table or with Row‑Level Security) and audit logs enabled. Service role keys can access all data and are over‑privileged for per‑patient follow‑ups. - Ask what credentials are needed for each channel (SMS/WhatsApp/Telegram/ElevenLabs) and why they are not declared in the registry. Prefer per‑channel, least‑privilege credentials. - Require a data protection plan: explicit patient consent, data minimisation, encryption at rest/in transit, retention policy (how long learning/ and memory/ are kept), deletion procedures, and a DPIA (Data Protection Impact Assessment) because this handles PHI. - Clarify scheduling model: how does the agent 'wait' and trigger messages? Will it run as a persistent service or rely on platform scheduling? If persistent, insist on hardened execution, restricted network egress, and admin controls. - Insist on an audit trail and access controls for stored PHI, and on a testing/sandbox mode with synthetic data before any live deployment. - If you must trial it, run in an isolated environment with synthetic patient data and monitor all outbound requests. Prefer a code review or only accept a version with accessible source so security/privacy reviewers can inspect network calls and storage behavior. - Consider disabling autonomous invocation or restricting the skill's access to credentials until you can verify its behavior and governance. Given the current inconsistencies and high‑risk data handling, treat this skill as needing further technical and legal review before production use.
功能分析
Type: OpenClaw Skill Name: ekin-safety-netting Version: 1.0.0 The skill is designed for NHS clinical safety netting, which involves processing sensitive patient PII and medical conditions. It is classified as suspicious because it requires high-privilege environment variables (SUPABASE_SERVICE_ROLE_KEY) and involves broad data handling capabilities across multiple communication channels (Email, SMS, Voice via ElevenLabs) as described in SKILL.md. While the instructions are aligned with the stated medical purpose, the combination of administrative database access and the handling of sensitive health data constitutes a high-risk profile under the provided criteria.
能力评估
Purpose & Capability
The skill claims to send follow‑ups across multiple channels and to store audit trails and learning data. The SKILL.md lists OPENMAIL_API_KEY, SUPABASE_URL and SUPABASE_SERVICE_ROLE_KEY as env entries, yet the registry metadata declared no required env vars — that's an internal inconsistency. The skill also references ElevenLabs (voice), Telegram, WhatsApp, SMS and other channels but doesn't declare the credentials those channels would need. Requiring a Supabase service role key and OpenMail API key (per SKILL.md) could be coherent for email + DB storage, but the missing declared requirements and additional unspecified channel credentials suggest the manifest and instructions are not aligned.
Instruction Scope
The instructions direct the agent to collect and store highly sensitive personal health information (names, contacts, conditions, exact patient responses) and to keep a learning/ dataset for ML improvements. They require waiting until follow‑up times and sending messages or making voice calls — but provide no mechanism for scheduling, background execution, or how escalation routing to the GP is authenticated/implemented. There's no guidance on consent, retention, encryption, or who may access the stored records. The guidance to 'record everything' and to keep learning data expands scope beyond simple one‑off notifications and creates significant privacy and compliance surface.
Install Mechanism
This is an instruction‑only skill (no install spec, no code files), which reduces local install risk. However, lack of code means you cannot audit runtime network calls; the SKILL.md expects network integrations (email, DB, voice, messaging). Instruction‑only status therefore shifts risk to runtime: the agent will make external API calls carrying PHI if given credentials.
Credentials
SKILL.md requests SUPABASE_SERVICE_ROLE_KEY — this is typically a highly privileged key that can bypass row‑level security and access the full DB. That is disproportionate for a single practice agent (a least‑privilege, per‑table or per‑row key would be preferable). OPENMAIL_API_KEY is plausible for email, but other channel keys (e.g., Twilio, Telegram, ElevenLabs) are not declared. Registry metadata claiming no required env vars conflicts with SKILL.md, which is a red flag and could lead to giving out over‑privileged credentials unintentionally.
Persistence & Privilege
The skill requires persistent storage of PHI under memory/ and learning/ with no retention, access control, encryption, or deletion policy described. It also implies it must be able to act after a delay (scheduling or background execution). Although always:false (not force‑installed), the need to persist PHI and run at scheduled times increases privilege/attack surface and raises compliance concerns (GDPR/NHS data governance).
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install ekin-safety-netting
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /ekin-safety-netting 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.0
Initial release with full safety netting automation: - Automates clinical safety netting follow-up after GP appointments for NHS patients. - Checks for red flag symptoms by contacting patients at defined timeframes, using plain English and multiple communication channels (email, SMS, WhatsApp, Telegram, voice). - Escalates to the GP if red flags are detected, if the patient does not respond, or responses are unclear. - Records all contacts, responses, matched flags, and decisions for audit and learning. - Includes built-in memory and learning to optimize follow-ups and improve phrase and keyword matching over time.
元数据
Slug ekin-safety-netting
版本 1.0.0
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 1
常见问题

Safety Netting 是什么?

Automated clinical safety netting for NHS GPs. Follows up with patients after appointments, checks for red flag symptoms, escalates to GP when needed. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 92 次。

如何安装 Safety Netting?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install ekin-safety-netting」即可一键安装,无需额外配置。

Safety Netting 是免费的吗?

是的,Safety Netting 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Safety Netting 支持哪些平台?

Safety Netting 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 Safety Netting?

由 Ekin-Kahraman(@ekin-kahraman)开发并维护,当前版本 v1.0.0。

💬 留言讨论