← 返回 Skills 市场
844
总下载
2
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install cameras
功能描述
Connect to security cameras, capture snapshots, and process video feeds with protocol support.
使用说明 (SKILL.md)
Scope
This skill:
- ✅ Generates camera capture commands
- ✅ Guides integration with security systems
- ✅ Provides troubleshooting for camera issues
User-driven model:
- User provides camera credentials (RTSP URLs, passwords)
- User runs capture commands
- User installs required tools
This skill does NOT:
- ❌ Store camera credentials
- ❌ Run captures automatically without user request
- ❌ Access cameras without user-provided access info
Requirements
Required:
ffmpeg— for capture and recording
Optional (user installs if needed):
gphoto2— for DSLR/mirrorless controlv4l2-ctl— for USB cameras on Linux
Quick Reference
| Topic | File |
|---|---|
| Security camera integration | security-integration.md |
| USB/webcam capture | capture.md |
| DSLR control | photography-control.md |
| Video processing | processing.md |
Core Rules
1. User Provides Camera Access
When user requests capture:
User: "Snapshot from my front door camera"
Agent: "I need the RTSP URL. Format: rtsp://user:pass@ip/stream
Provide it or set CAMERA_FRONT_URL in env."
User: "rtsp://admin:[email protected]/stream1"
→ Agent generates: ffmpeg -i "URL" -frames:v 1 snapshot.jpg
2. Common Commands
# Snapshot from RTSP (user provides URL)
ffmpeg -i "$RTSP_URL" -frames:v 1 snapshot.jpg
# Record 10s clip
ffmpeg -i "$RTSP_URL" -t 10 -c copy clip.mp4
# Webcam snapshot (macOS)
ffmpeg -f avfoundation -i "0" -frames:v 1 webcam.jpg
# Webcam snapshot (Linux)
ffmpeg -f v4l2 -i /dev/video0 -frames:v 1 webcam.jpg
3. Protocol Reference
| Protocol | Use Case | URL Format |
|---|---|---|
| RTSP | IP cameras | rtsp://user:pass@ip:554/stream |
| HTTP | Simple cams | http://ip/snapshot.jpg |
| V4L2 | USB cameras | /dev/video0 |
4. Integration Patterns
With Home Assistant:
GET /api/camera_proxy/camera.front_door
User provides HA URL and token.
With Frigate:
- MQTT:
frigate/eventsfor alerts - HTTP:
/api/events/{id}/snapshot.jpg
5. Security
- Never log camera URLs with credentials
- Recommend user stores URLs in env vars
- RTSP streams may be unencrypted — warn about LAN security
安全使用建议
This skill mostly does what it says (ffmpeg-based capture and local processing) but you should verify a few things before installing: 1) Ask the publisher to list exactly which environment variables and tokens the skill will ask for and why (Anthropic/Google/AWS/HomeAssistant/Ring/Frigate/MQTT). 2) Confirm whether the agent will ever run periodic monitoring or subscribe to MQTT autonomously — require explicit opt-in for continuous captures. 3) Prefer local processing (Frigate/OpenCV) and avoid sending raw snapshots to cloud services unless you understand where they go; if you must use cloud APIs, create limited-scope service accounts / short-lived tokens and rotate them. 4) Test in an isolated network or VLAN with test cameras first, and ensure the agent (or platform) will not log or persist RTSP URLs or credentials. 5) If you need stronger assurances, request an explicit list of outbound endpoints the skill will contact and a privacy/security policy from the author.
功能分析
Type: OpenClaw Skill
Name: cameras
Version: 1.0.1
The skill bundle is generally benign, providing functionality for camera integration and video processing using standard tools like ffmpeg, gphoto2, and cloud vision APIs. The SKILL.md explicitly promotes user-driven credential handling and warns against logging sensitive data. However, the `security-integration.md` file contains a `curl -k` command for UniFi Protect integration, which disables SSL certificate validation. This creates a Man-in-the-Middle (MITM) vulnerability, allowing potential interception and manipulation of communication with the camera system. This is a significant security flaw, classifying the skill as suspicious due to this vulnerability.
能力评估
Purpose & Capability
Declared purpose (connect to cameras, capture snapshots, process video) matches required binary ffmpeg and optional camera tooling. However the documentation also instructs use of multiple cloud vision APIs (Anthropic, Google Cloud Vision, AWS Rekognition), Ring/Nest APIs, Home Assistant tokens, and Frigate/MQTT — none of these credentials or endpoints are declared in the skill metadata. Examples also show ONVIF with plaintext creds. The presence of many external integrations without declared credential requirements is an incoherence.
Instruction Scope
SKILL.md and included files provide concrete commands and agent code that will capture images, run subprocesses, discover networked cameras (ONVIF), subscribe to MQTT, and send images to external vision services. While that is within a camera skill's remit, the instructions reference environment variables and tokens (e.g., CAMERA_FRONT_URL, HA_TOKEN, Ring refreshToken, AWS/Google creds, Anthropic client) that are not declared. The skill claims it will not run captures automatically, yet processing.md contains sample loops for periodic monitoring and event-driven MQTT handling — a scope contradiction that could enable continuous capture if implemented by an agent.
Install Mechanism
Instruction-only skill with no install spec and no code files. That keeps the on-disk footprint small and is lower risk than arbitrary downloads or install scripts.
Credentials
The skill requests no environment variables in metadata but the runtime instructions repeatedly reference env vars and external service tokens (Anthropic, Google, AWS, Home Assistant, Ring, Frigate, MQTT credentials). This mismatch means sensitive credentials are implied by the instructions but not declared or scoped, which increases the chance of accidental credential exposure or unclear authorization boundaries.
Persistence & Privilege
Skill metadata does not request always: true and uses normal model-invocation defaults. Still, the documentation includes patterns for continuous monitoring and MQTT subscription; if an agent implements those autonomously, that increases blast radius. The skill claims 'does NOT run captures automatically' but supplies example code that would — a small but important contradiction to resolve.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install cameras - 安装完成后，直接呼叫该 Skill 的名称或使用
/cameras触发 - 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
v1.0.1
User-driven credential model, declared tool requirements
v1.0.0
Initial release - camera integration for agents
元数据
常见问题
Cameras 是什么？
Connect to security cameras, capture snapshots, and process video feeds with protocol support. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 844 次。
如何安装 Cameras？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install cameras」即可一键安装，无需额外配置。
Cameras 是免费的吗？
是的，Cameras 完全免费（开源免费），可自由下载、安装和使用。
Cameras 支持哪些平台？
Cameras 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（linux, darwin）。
谁开发了 Cameras？
由 Iván（@ivangdavila）开发并维护，当前版本 v1.0.1。
推荐 Skills