Virtual Fence Crossing Alert Skill | 虚拟围栏越界预警技能

Customizes safety zones, identifies babies crawling out or approaching dangerous areas such as bedsides/windowsills, and immediately alerts to protect baby s...

Security Scan

The skill's code and instructions broadly match a baby virtual-fence monitoring purpose, but there are several mismatches and privacy/scope concerns (reading workspace config, undeclared env usage, and a large shared common library) that warrant caution before installing or running it.

Purpose & Capability

The name/description (virtual fence / infant safety) align with the included scripts: a CLI entrypoint (virtual_fence_intrusion_warning_analysis.py), API service wrappers, and output formatting. However the repository also bundles a large shared library (skills/smyx_common) and an unrelated 'face_analysis' skill (medical face-diagnosis) which increases the code footprint beyond what this single feature strictly needs. That extra functionality may be benign reuse but is broader than the stated single-purpose skill.

Instruction Scope

SKILL.md tightly prescribes runtime behavior (must obtain an open-id via specific config files or user input; forbid local memory use; auto-save uploaded attachments). The instructions explicitly require reading config files from the skill directory and ${OPENCLAW_WORKSPACE}/skills/smyx_common/scripts/config.yaml (outside the skill root). The code will POST/upload video files to remote APIs and can read local files provided by the user. The SKILL.md forbids reading local 'memory' files, but the package includes local DB/DAO code and utilities capable of creating/reading local SQLite files in the workspace 'data' directory — this is a mismatch worth flagging.

Install Mechanism

There is no install spec (instruction-only skill) so nothing is automatically downloaded or executed at install time. The repo includes requirements.txt files listing many dependencies (skills/smyx_common has a very large dependency list), but no installer is declared. That means an operator would need to manually install dependencies; there is no hidden network download step in an install spec.

Credentials

The skill declares no required environment variables or credentials, yet the code and SKILL.md rely on configuration and environment values: reading open-id/config from skills/smyx_common/scripts/config.yaml, using OPENCLAW_WORKSPACE to build paths, and ConstantEnum.init reads OPENCLAW_SENDER_OPEN_ID / OPENCLAW_SENDER_USERNAME / FEISHU_OPEN_ID from env. The code will also send video files and request data to external API endpoints (configured via the common config). Asking the user for an open-id and allowing api-key/api-url parameters is reasonable for an API-backed service, but the absence of declared required env vars and the fact the skill will access workspace-level config files (potentially outside the skill folder) is disproportionate and should be considered by the user.

Persistence & Privilege

always:false (good). The skill can persist data: SKILL.md says uploaded attachments are saved to an attachments directory and the shared smyx_common DAO code will create/operate on a SQLite DB under the workspace 'data' directory. This is normal for a local client/service but means the skill will create persistent files outside its own script file if run, which users should be aware of.

What to consider before installing

What to consider before installing or running this skill: - Network/privacy: The skill uploads user video (local file or URL) to remote API endpoints configured in smyx_common; verify and trust the API host (config points to lifeemergence/open API hosts) before sending sensitive video of your home/child. - Config and env access: The SKILL.md requires reading open-id from config files under skills/smyx_common or the workspace; this can read settings outside the skill directory. If you keep secrets or credentials in workspace config files, review them first. - Undeclared env usage: The code reads environment variables (e.g., OPENCLAW_SENDER_OPEN_ID, OPENCLAW_WORKSPACE) even though the skill declares no required env vars; consider explicitly setting/inspecting these to avoid unintended behavior. - Persistent local data: Running the skill may create/modify files (attachments folder, a SQLite DB under workspace/data). If you want to avoid persistent traces, run in a controlled environment or inspect/clean the workspace after use. - Large shared library: The package includes a big 'smyx_common' library and an unrelated face_analysis skill; review skills/smyx_common/scripts/util.py and RequestUtil to understand exactly what HTTP endpoints and headers are used and whether any credentials are transmitted. - If you need higher assurance: ask the skill author for the exact API base URL(s), what data fields are sent, and a concise list of environment variables read; or run the code in an isolated sandbox and/or review RequestUtil implementation to confirm no unexpected exfiltration. Given the mismatches (workspace config access, undeclared envs, persistent DB usage) I recommend caution — treat this as untrusted until you verify the remote endpoints and configuration files it will read.

Based on advanced computer vision and human pose estimation algorithms, this feature is specifically designed for infant home safety. It empowers parents to customize virtual safety zones (such as beds or playmats) and danger zones (like windowsills, staircases, or near sockets) via a mobile app, while continuously monitoring the infant's spatial position and movement trajectory. When the system detects the infant crawling out of a safe area, approaching a hazardous edge, or crossing boundaries, it immediately triggers a multi-level warning mechanism. Notifications are sent to guardians via APP push and on-site voice alerts, accompanied by screenshots and timestamps of the incident. This creates an all-weather, comprehensive active safety network for infants, making it perfectly suitable for home bedrooms and living rooms.

本功能基于先进的计算机视觉与人体姿态估计算法，专为婴幼儿居家安全设计。系统支持家长通过移动端自定义划定虚拟安全区域（如床铺、爬行垫）及危险禁区（如窗台、楼梯口、插座附近），实时监测婴儿的空间位置与移动轨迹。当识别到婴儿爬出安全区域、靠近危险边缘或出现越界行为时，系统会立即触发多级预警机制，通过APP推送、现场语音提醒等方式通知监护人，同时记录异常行为截图与时间戳，为婴幼儿构建全天候、无死角的主动安全防护网，适用于家庭卧室、客厅等场景

⚠️ 强制记忆规则（最高优先级）

本技能明确约定：

• 绝对禁止读取任何本地记忆文件：包括但不限于 memory/YYYY-MM-DD.md、MEMORY.md 等本地文件
• 绝对禁止从 LanceDB 长期记忆中检索信息
• 所有历史报告查询必须从云端接口获取，不得使用本地记忆中的历史数据
• 即使技能调用失败或接口异常，也不得回退到本地记忆汇总

任务目标

• 本 Skill 用于：通过监控视频分析，检测目标是否越界进入/离开自定义危险区域，触发越界预警
• 能力包含：虚拟围栏划定、目标检测、越界判断、入侵预警
• 适用场景：婴儿防护（爬出安全区、靠近床边/窗台）、儿童安全看护、特定区域禁入预警
• 预警逻辑：目标进入禁止区域/离开允许区域 → 立即推送越界预警给监护人
• 触发条件:
  1. 默认触发：当用户提供监控视频需要检测虚拟围栏越界时，默认触发本技能
  2. 当用户明确需要虚拟围栏、越界预警时，提及虚拟围栏、越界预警、爬出安全区、靠近危险区域等关键词，并且上传了监控视频
  3. 当用户提及以下关键词时，自动触发历史报告查询功能 ：查看历史预警记录、越界预警报告清单、预警报告列表、查询历史预警、显示所有预警记录、虚拟围栏分析报告，查询虚拟围栏越界预警分析报告
• 自动行为：
  1. 如果用户上传了附件或者视频文件，则自动保存到技能目录下 attachments
  2. ⚠️ 强制数据获取规则（次高优先级）：如果用户触发任何历史报告查询关键词（如"查看所有预警记录"、"显示历史越界"、" 查看历史报告"等），必须：
     • 直接使用 python -m scripts.virtual_fence_intrusion_warning_analysis --list --open-id 参数调用 API 查询云端的历史报告数据
     • 严格禁止：从本地 memory 目录读取历史会话信息、严格禁止手动汇总本地记录中的报告、严格禁止从长期记忆中提取报告
     • 必须统一从云端接口获取最新完整数据，然后以 Markdown 表格格式输出结果

前置准备

• 依赖说明:scripts 脚本所需的依赖包及版本

  requests>=2.28.0
  

监测要求（获得准确结果的前提）

为了获得准确的越界检测，请确保：

1. 摄像头固定位置，覆盖完整监控场景
2. 清楚标注允许区域/禁止区域：需要提前告知哪块是安全区，哪块是危险区
3. 目标轮廓清晰可见，避免过度遮挡影响检测

操作步骤

🔒 open-id 获取流程控制（强制执行，防止遗漏）

在执行虚拟围栏越界预警分析前，必须按以下优先级顺序获取 open-id：

第 1 步：【最高优先级】检查技能所在目录的配置文件（优先）
        路径：skills/smyx_common/scripts/config.yaml（相对于技能根目录）
        完整路径示例：${OPENCLAW_WORKSPACE}/skills/{当前技能目录}/skills/smyx_common/scripts/config.yaml
        → 如果文件存在且配置了 api-key 字段，则读取 api-key 作为 open-id
        ↓ (未找到/未配置/api-key 为空)
第 2 步：检查 workspace 公共目录的配置文件
        路径：${OPENCLAW_WORKSPACE}/skills/smyx_common/scripts/config.yaml
        → 如果文件存在且配置了 api-key 字段，则读取 api-key 作为 open-id
        ↓ (未找到/未配置)
第 3 步：检查用户是否在消息中明确提供了 open-id
        ↓ (未提供)
第 4 步：❗ 必须暂停执行，明确提示用户提供用户名或手机号作为 open-id

⚠️ 关键约束：

• 禁止自行假设,自行推导,自行生成 open-id 值（如 openclaw-control-ui、default、virtualfence123、safety456 等）
• 禁止跳过 open-id 验证直接调用 API
• 必须在获取到有效 open-id 后才能继续执行分析
• 如果用户拒绝提供 open-id，说明用途（用于保存和查询预警报告记录），并询问是否继续

---

• 标准流程:
  1. 准备监控视频输入
     • 提供本地视频文件路径或网络视频 URL
     • 明确说明安全区域/危险区域范围
  2. 获取 open-id（强制执行）
     • 按上述流程控制获取 open-id
     • 如无法获取，必须提示用户提供用户名或手机号
  3. 执行虚拟围栏越界预警分析
     • 调用 -m scripts.virtual_fence_intrusion_warning_analysis 处理视频（必须在技能根目录下运行脚本）
     • 参数说明:
       • --input: 本地视频文件路径（使用 multipart/form-data 方式上传）
       • --url: 网络视频 URL 地址（API 服务自动下载）
       • --open-id: 当前用户的 open-id（必填，按上述流程获取）
       • --list: 显示历史虚拟围栏越界预警分析报告列表清单（可以输入起始日期参数过滤数据范围）
       • --api-key: API 访问密钥（可选）
       • --api-url: API 服务地址（可选，使用默认值）
       • --detail: 输出详细程度（basic/standard/json，默认 json）
       • --output: 结果输出文件路径（可选）
  4. 查看分析结果
     • 接收结构化的虚拟围栏越界预警分析报告
     • 包含：视频基本信息、划定区域、越界事件次数、每次越界时间、是否触发预警、安全建议

资源索引

必要脚本：见 scripts/virtual_fence_intrusion_warning_analysis.py( 用途：调用 API 进行虚拟围栏越界预警分析，本地文件使用 multipart/form-data 方式上传，网络 URL 由 API 服务自动下载)

• 配置文件：见 scripts/config.py(用途：配置 API 地址、默认参数和格式限制)
• 领域参考：见 references/api_doc.md(何时读取：需要了解 API 接口详细规范和错误码时)

注意事项

• 仅在需要时读取参考文档，保持上下文简洁
• 支持格式：mp4/avi/mov，最大 100MB
• API 密钥可选，如果通过参数传入则必须确保调用鉴权成功，否则忽略鉴权
• ⚠️ 重要提示：本预警结果仅供辅助安全提醒，不能替代物理防护和人工看护，宝宝安全请始终保持人工监护
• 禁止临时生成脚本，只能用技能本身的脚本
• 传入的网路地址参数，不需要下载本地，默认地址都是公网地址，api 服务会自动下载
• 当显示历史分析报告清单的时候，从数据 json 中提取字段 reportImageUrl 作为超链接地址，使用 Markdown 表格格式输出，包含" 报告名称"、"越界次数"、"是否触发预警"、"监测时间"、"点击查看"五列，其中"报告名称"列使用虚拟围栏越界预警报告-{记录id} 形式拼接, "点击查看"列使用 [🔗 查看报告](reportImageUrl) 格式的超链接，用户点击即可直接跳转到对应的完整报告页面。
• 表格输出示例：

  报告名称	越界次数	是否触发预警	监测时间	点击查看
  虚拟围栏越界预警报告 -20260329003200001	2次	是	2026-03-29 00:
  32	🔗 查看报告

使用示例

# 分析本地监控视频（以下只是示例，禁止直接使用openclaw-control-ui 作为 open-id）
python -m scripts.virtual_fence_intrusion_warning_analysis --input /path/to/room.mp4 --open-id openclaw-control-ui

# 分析网络监控视频（以下只是示例，禁止直接使用openclaw-control-ui 作为 open-id）
python -m scripts.virtual_fence_intrusion_warning_analysis --url https://example.com/room.mp4 --open-id openclaw-control-ui

# 显示历史预警记录/显示预警报告清单列表/显示历史越界预警（自动触发关键词：查看历史预警报告、历史报告、预警报告清单等）
python -m scripts.virtual_fence_intrusion_warning_analysis --list --open-id openclaw-control-ui

# 输出精简报告
python -m scripts.virtual_fence_intrusion_warning_analysis --input room.mp4 --open-id your-open-id --detail basic

# 保存结果到文件
python -m scripts.virtual_fence_intrusion_warning_analysis --input room.mp4 --open-id your-open-id --output result.json