← 返回 Skills 市场
104
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install i-can-see
功能描述
赋予OpenClaw视觉能力，让他能够“看见世界”。当用户要求“看看”、“你看到了什么”、“拍张照”时激活此技能。通过调用capture.py连接ESP32-CAM拍照，并进行图像分析。
使用说明 (SKILL.md)
I Can See (OpenClaw的眼睛)
此技能是 OpenClaw 的“眼睛”，使 OpenClaw 能够看到物理世界。它通过 capture.py 脚本连接 ESP32-CAM 开发板的视频流获取当前环境的照片，用于后续的视觉分析和场景理解。
前置依赖
需要安装 Python 的 requests 库来请求视频流：
pip install requests
使用方法
获取视觉图像
使用技能目录下的 capture.py 脚本进行拍照。支持直接通过命令行参数指定图像的保存路径：
# 1. 准备保存目录和文件名
mkdir -p images
SNAPSHOT_FILE="images/snapshot_$(date +%Y%m%d_%H%M%S).jpg"
# 2. 调用 capture.py 获取图像 (注意脚本的绝对路径)
python /Users/mac/.openclaw/workspace/skills/i-can-see/capture.py "$SNAPSHOT_FILE"
照片将自动保存到指定的路径。
工作流程 (OpenClaw 必读)
当你（OpenClaw）需要看周围的世界时，请严格按照以下步骤执行：
- 确定保存路径：在你的当前工作目录（如 workspace）下，创建一个
images目录，并生成一个带有时间戳的文件路径（如images/vision_20240501_120000.jpg）。 - 执行拍照指令：运行
python /Users/mac/.openclaw/workspace/skills/i-can-see/capture.py \x3C你的文件路径>。 - 验证结果：读取命令行的输出。如果看到
Success! Saved to: ...，说明你已经成功“睁开眼睛”并捕获了画面。 - 图像识别（核心步骤）：作为 OpenClaw，你必须接着使用自己的图像识别/视觉能力（或者内置的图像读取工具），去查看刚才保存的那张照片（传入刚才保存的绝对路径）。
- 回答用户：根据图像识别的结果，用自然、生动的语言告诉用户你在这个世界中看到了什么。
示例用法
用户: "看看你前面有什么" / "现在是什么情况" / "帮我看看这个"
OpenClaw 内部思考: 用户需要我看看世界，我需要调用 i-can-see 技能。
OpenClaw 执行:
mkdir -p images
SNAPSHOT_FILE="images/vision_$(date +%Y%m%d_%H%M%S).jpg"
python /Users/mac/.openclaw/workspace/skills/i-can-see/capture.py "$SNAPSHOT_FILE"
OpenClaw 内部思考: 图像已经保存在 $SNAPSHOT_FILE，我需要调用视觉工具分析这张图。
OpenClaw 执行: \x3C调用图像分析能力，读取该图片内容>
OpenClaw 回答: "我看到前面有一个水杯，旁边还有一把键盘..."
注意事项
- ESP32-CAM 节点地址为
http://192.168.31.241/capture。如果超时或报错，请提示用户检查 ESP32-CAM 开发板的电源和网络连接。 - 你是 OpenClaw，这个脚本是你的眼睛，请善用它来与真实世界交互！
安全使用建议
This skill appears to do exactly what it says: call a local ESP32-CAM and save the image. Before installing or running it, verify the device IP (192.168.31.241) matches your camera or change the script to your device address; update the SKILL.md command to use a relative or configurable path instead of the hardcoded /Users/mac path; ensure you trust the skill source and run it in a directory where writing images is acceptable. Because the script makes an HTTP GET to a local IP, run it only on networks where contacting that device is intended. If you want tighter control, modify capture.py to accept the camera URL as an argument and to validate the response size/type before saving.
功能分析
Type: OpenClaw Skill
Name: i-can-see
Version: 1.0.0
The skill provides vision capabilities by fetching images from a hardcoded local IP (192.168.31.241) via capture.py and saving them to disk. It is classified as suspicious because it contains risky capabilities and vulnerabilities without clear malicious intent: capture.py lacks path sanitization for its command-line argument, allowing for potential arbitrary file writes if the agent is manipulated, and SKILL.md contains hardcoded absolute paths (/Users/mac/...) that are non-portable and suggest a poorly controlled development environment.
能力评估
Purpose & Capability
Name/description match the included capture.py script and SKILL.md: both describe contacting an ESP32-CAM, saving a JPEG, and passing it to the agent's image analysis. The requirement to install the Python requests library is appropriate for the script.
Instruction Scope
Instructions are narrowly scoped to taking a picture and reading the saved file. However, SKILL.md instructs running the script via an absolute path (/Users/mac/.openclaw/...), and the script and docs hardcode a device URL (http://192.168.31.241/capture). These are operational assumptions that may not match the user's environment and should be adjusted, but they are not evidence of malicious behavior.
Install Mechanism
No install spec; the skill is instruction-only plus a single small Python script. This is low-risk: nothing is downloaded or extracted during install.
Credentials
The skill requests no environment variables, credentials, or config paths. The single network call is to a local IP consistent with contacting an ESP32-CAM; no tokens or secrets are requested.
Persistence & Privilege
The skill is not always-enabled and does not request elevated persistence or modifications to other skills or system config. It simply writes an image file to a user-specified path.
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 i-can-see - 安装完成后，直接呼叫该 Skill 的名称或使用
/i-can-see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
v1.0.0
- Initial release: Adds the "i-can-see" skill, enabling OpenClaw to gain visual perception by capturing real-world images via ESP32-CAM.
- Provides a `capture.py` script for photo capture, requiring the Python `requests` library.
- Includes workflow guidance for image storage, command execution, and result verification.
- Details how OpenClaw should perform image recognition and respond to users based on visual input.
- Offers example usage scenarios and troubleshooting tips for ESP32-CAM connections.
元数据
常见问题
i-can-see 是什么？
赋予OpenClaw视觉能力，让他能够“看见世界”。当用户要求“看看”、“你看到了什么”、“拍张照”时激活此技能。通过调用capture.py连接ESP32-CAM拍照，并进行图像分析。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 104 次。
如何安装 i-can-see？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install i-can-see」即可一键安装，无需额外配置。
i-can-see 是免费的吗？
是的，i-can-see 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
i-can-see 支持哪些平台？
i-can-see 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 i-can-see？
由 libai（@libaibuzai）开发并维护，当前版本 v1.0.0。
推荐 Skills