Intelligent Outdoor Care Monitoring & Analysis Tool | 户外看护智能监测分析工具

Detects targets such as people, vehicles, non-motorized vehicles, and pets within target areas; supports batch image analysis, suitable for outdoor surveilla...

Security Scan

The skill's code and runtime instructions disagree in multiple places: it embeds API credentials and local persistence mechanisms while the SKILL.md forbids local memory access and requests cloud-only behavior — this mismatch and the included external endpoints are concerning.

Purpose & Capability

The codebase implements outdoor monitoring and also bundles a separate 'face_analysis' subskill and a large 'smyx_common' library; those pieces provide API calls, DB access, export URLs and health/face-analysis features. Having face-analysis and database/DAO code included is not obviously disproportional for a single 'outdoor monitoring' skill and suggests code reuse or scope-creep. The presence of many unrelated utilities (DB/DAO, many scene codes, health analysis endpoints) increases the attack surface and is unexpected for a narrowly-scoped target-detection skill.

Instruction Scope

SKILL.md imposes strict runtime rules (absolute prohibition on reading local memory files and LanceDB; require cloud-only history queries; must save uploaded attachments to attachments dir). The actual code: (a) contains a local DAO/SQLite implementation that writes to disk and constructs DB paths, (b) loads config YAML files from the repository, and (c) uses ConstantEnum init to read environment variables and persist current open-id — so the implementation can read and write local state despite the SKILL.md prohibition. Also the SKILL.md references env names OPENCLAW_SENDER_ID or sender_id, but the code looks for OPENCLAW_SENDER_OPEN_ID / OPENCLAW_SENDER_USERNAME / FEISHU_OPEN_ID — a mismatch in names and behavior.

Install Mechanism

There is no install spec, but the skill ships with many Python modules and two requirements.txt files listing a very large dependency set. That is disproportionate for a lightweight image-analysis wrapper and means anyone installing this skill will likely need to install many packages. No external downloads or obscure URLs are present, but the lack of an install spec + large dependency lists is a usability/risk note.

Credentials

The skill declares no required env vars, yet the repository contains configuration YAML (skills/smyx_common/scripts/config.yaml and test/dev configs) with API keys, an API secret, database URLs, and base service endpoints (e.g., open.lifeemergence.com and other domains/IPs). Embedding service credentials and endpoints inside the codebase (instead of explicit, documented env var requirements) is a red flag: it may leak secrets or connect to third-party services unexpectedly. The skill will call external APIs and may transmit uploaded files to those endpoints.

Persistence & Privilege

The included DAO/SQLite code creates and writes a local database under a workspace/data path and the codebase contains utilities for file/directory creation. SKILL.md forbids reading local memory, yet code persists local state (and may create files). The skill is not flagged as 'always: true', but it does have local persistence abilities which contradict the stated prohibition and increase persistence/privilege concerns.

Scan Findings in Context

[HARD_CODED_CREDENTIALS] unexpected: Repository contains an API key and API secret in skills/smyx_common/scripts/config.yaml (ApiEnum.api-key and api-secret-key) and base URLs pointing to external domains/IPs. Hard-coded service credentials are present in the package rather than being requested from the environment or user, which is unexpected and risky for a monitoring skill.

What to consider before installing

Before installing or using this skill, consider the following: - There are clear mismatches between the SKILL.md and the code: the documentation forbids using local memory, but the code includes a local SQLite DAO and creates files/dirs. Ask the author to explain and reconcile this behavior. - The package includes configuration YAML with embedded API keys/secrets and external service URLs. That means the skill will contact third-party endpoints and could transmit uploaded images/videos. If you do not trust those endpoints, do not install or run the skill with real data. - The SKILL.md's open-id/environment-variable names do not match the ones the code actually reads (e.g., SKILL.md: OPENCLAW_SENDER_ID / sender_id; code: OPENCLAW_SENDER_OPEN_ID / OPENCLAW_SENDER_USERNAME / FEISHU_OPEN_ID). Confirm how open-id is obtained and whether any identifiers are sent to the remote API. - The repository bundles a very large set of Python dependencies (requirements.txt). Install in an isolated environment (container/VM) and review/test in a sandbox before granting network/file access. - Recommended precautions: request the maintainer to (1) remove hard-coded secrets or explain ownership of those credentials, (2) provide a minimal install spec and a clear list of exact external endpoints and data flows, (3) confirm whether attachments or local DBs will be written and what is stored there, and (4) run the skill in a network-restricted, ephemeral sandbox until you’re satisfied. If you need, I can list the files that contain credentials/URLs and point to the exact lines to inspect, or suggest specific safe-run steps (sandbox commands and network restrictions).

⚠️ 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：通过户外监控图片/视频进行目标检测，识别区域内出现的人、车、非机动车、宠物等闯入目标
• 能力包含：多目标检测、目标分类、数量统计、入侵判定、风险等级评估、异常闯入预警
• 支持批量处理一组图片，同时分析多帧监控画面
• 触发条件:
  1. 默认触发：当用户提供户外监控图片/视频需要检测闯入目标时，默认触发本技能进行户外看护分析
  2. 当用户明确需要进行户外看护、入侵检测时，提及庭院看护、果园监控、目标检测、户外安防等关键词，并且上传了图片或视频文件
  3. 当用户提及以下关键词时，自动触发历史报告查询功能 ：查看历史监测报告、户外看护报告清单、监测报告列表、查询历史监测报告、显示所有监测报告、户外监测分析报告，查询户外看护智能监测分析报告
• 自动行为：
  1. 如果用户上传了附件或者图片/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"查看所有监测报告"、"显示所有看护报告"、"查看历史报告"等），必须：
     • 直接使用 python -m scripts.outdoor_monitoring --list --open-id {从消息上下文获取 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
     • 如果用户未明确提供 open-id，优先从 OpenClaw 消息上下文获取 sender id（如 metadata 中的 id 字段），然后尝试从当前消息上下文的环境变量 OPENCLAW_SENDER_ID 或者 sender_id 获取，无法获取时则必须用户提供用户名或者手机号作为 open-id

前置准备

• 依赖说明:scripts 脚本所需的依赖包及版本

  requests>=2.28.0
  

操作步骤

🔒 open-id 获取流程控制（强制执行，防止遗漏）

在执行户外看护分析前，必须按以下优先级顺序获取 open-id：

第 1 步：检查用户是否在消息中明确提供了 open-id
        ↓ (未提供)
第 2 步：从当前消息上下文的环境变量中获取 OPENCLAW_SENDER_ID
        ↓ (无法获取)
第 3 步：从当前消息上下文的环境变量中获取 sender_id
        ↓ (无法获取)
第 4 步：从 OpenClaw 消息元数据中获取 id 字段（如 metadata 中的 id/session_id/user_id等）作为 open-id
        ↓ (无法获取)
第 5 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：

• 禁止自行假设或生成 open-id 值（如 outdoor123、monitor456 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（用于保存和查询监测报告记录），并询问是否继续

---

• 标准流程:
  1. 准备图片/视频输入
     • 提供本地图片/视频文件路径或网络 URL
     • 支持批量上传一组图片同时分析
     • 确保监控画面覆盖完整目标监测区域
  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. 执行户外看护智能监测分析
     • 调用 -m scripts.outdoor_monitoring 处理输入（必须在技能根目录下运行脚本）
     • 参数说明:
       • --input: 本地图片/视频文件路径（使用 multipart/form-data 方式上传）
       • --url: 网络图片/视频 URL 地址（API 服务自动下载）
       • --open-id: 当前用户的 OpenID/UserId（必填，按上述流程获取）
       • --list: 显示历史户外看护监测分析报告列表清单（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
     • 接收结构化的户外看护智能监测分析报告
     • 包含：监控基本信息、检测到的目标类型、目标数量、位置分布、是否异常闯入、风险等级、处置建议

资源索引

• 必要脚本：见 scripts/outdoor_monitoring.py(用途：调用 API 进行户外看护智能监测分析，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：jpg/jpeg/png/mp4/avi/mov，最大 100MB，支持批量图片分析
• 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
• 分析结果仅供安防参考，不能替代专业安保措施，发现可疑闯入请及时报警
• 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"、"输入类型"、"分析时间"、"检测目标数"、"风险等级"、"点击查看"六列，其中"报告名称"列使用户外看护监测分析报告-{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	输入类型	分析时间	检测目标数	风险等级	点击查看
  户外看护监测分析报告 -20260328221000001	多图	2026-03-28 22:10:00	2人+1车	中风险	🔗 查看报告

使用示例

# 分析单张监控图片（OpenClaw UI 上下文，使用 metadata id 作为 open-id）
python -m scripts.outdoor_monitoring --input /path/to/yard.jpg --open-id openclaw-control-ui

# 分析网络监控视频（OpenClaw UI 上下文，使用 metadata id 作为 open-id）
python -m scripts.outdoor_monitoring --url https://example.com/garden.mp4 --open-id openclaw-control-ui

# 显示历史分析报告/显示分析报告清单列表/显示历史监测报告（自动触发关键词：查看历史监测报告、历史报告、监测报告清单等）
python -m scripts.outdoor_monitoring --list --open-id openclaw-control-ui

# 输出精简报告
python -m scripts.outdoor_monitoring --input capture.jpg --open-id your-open-id --detail basic

# 保存结果到文件
python -m scripts.outdoor_monitoring --input capture.jpg --open-id your-open-id --output result.json