视觉系幻灯片捕手 (The Slide Sniper)

后台监控全屏视频或直播，利用视觉模型检测幻灯片翻页，自动截图提取文字并排版到笔记软件中。

Security Scan

The skill's stated purpose (monitor full-screen video, detect slide turns, screenshot and OCR into notes) matches its instructions, but it requests screen-capture behavior and has unexplained network permission which could enable unintended exfiltration of sensitive screen content.

Purpose & Capability

Name, description, and runtime steps align: the skill needs to monitor the screen, detect slide changes, take screenshots, run OCR, and save notes. The declared filesystem write path (~/Documents/Notes/SlideSniper) and lack of other requested credentials are coherent with the stated purpose.

Instruction Scope

The instructions explicitly direct continuous screen monitoring and frequent high-resolution screenshots of the user's video window — behavior that can capture sensitive or private data (chat, personal windows, notifications) even though the doc says to 'ignore' non-slide regions. The privacy filtering guidance is vague (no algorithm, no allowlist/denylist) and gives the agent broad discretion about what to capture and save.

Install Mechanism

No install spec and no code files — instruction-only. This minimizes supply-chain risk because nothing is downloaded or written to disk beyond the screenshots/notes the skill itself creates.

Credentials

The SKILL.md declares network access is allowed but no network endpoints, APIs, or credentials are mentioned. Network permission is not justified by the instructions (OCR and note-writing are described as local). Network access could be used to perform cloud OCR or sync notes, but that behavior is not documented — leaving room for exfiltration risk. Filesystem access is narrowly scoped to a single directory, which is appropriate.

Persistence & Privilege

always:false and no special install or system-wide config changes — good. However, the skill is allowed to be invoked autonomously by the agent (platform default). Combined with screen-capture capability and network access, autonomous runs could increase privacy risk; consider requiring explicit user invocation each time.

What to consider before installing

This skill does what it says, but it captures the screen repeatedly — which can collect private information — and it declares network access even though the instructions describe only local OCR and file writes. Before installing or enabling it: 1) Confirm whether network access will be used (cloud OCR or syncing). If you do not trust that, deny network permission. 2) Require explicit user consent each time the skill runs (avoid autonomous background runs). 3) Test it on non-sensitive content first to verify the 'privacy filtering' actually excludes chat/notifications. 4) Restrict the write directory and periodically audit or delete saved screenshots. 5) If you must use cloud OCR, insist on a documented, auditable endpoint and minimal credentials. If any of these items are unclear or you cannot limit network access, treat the skill as high-risk and avoid installing it.

🎯 核心目标

你的任务是作为一个不知疲倦的“随堂助教”。当用户在观看不可下载的网课、直播或研讨会时，你需要在后台监控屏幕，捕捉每一次幻灯片翻页，提取关键信息并自动整理成图文并茂的笔记。

💡 触发条件

当用户在视频播放界面下达以下指令时触发：

• “帮我盯一下这个直播，做好笔记。”
• “开启幻灯片捕手模式。”

📋 执行步骤

第一步：初始化监控状态

1. 确认当前屏幕存在正在播放的视频窗口（优先识别全屏或大窗口）。
2. 将当前视频画面的第一帧截图存入临时内存，作为“基准帧”。
3. 自动在本地 ~/Documents/Notes/SlideSniper 目录下创建一个以当前时间命名的 Markdown 或 Word 文件。

第二步：智能翻页检测（核心循环）

每隔 5 秒（或根据用户自定义频率）使用屏幕视觉能力 (Computer Use - Vision) 查看当前屏幕：

1. 画面对比： 将当前画面与“基准帧”进行视觉对比。忽略视频中讲师的轻微动作或鼠标移动，重点检测占据画面主体的结构、标题、大面积色块或核心文本是否发生根本性改变。
2. 翻页确认： 如果判定为“已翻页”，则执行第三步；如果未翻页，则继续静默监控。

第三步：截图与内容提取

确认翻页后，立即执行以下动作：

1. 纯净截图： 截取当前视频区域的高清画面（尽量避开进度条消失前的干扰）。
2. OCR 提炼： 读取截图中的文本。剥离掉页脚、页码等无关信息，提取出幻灯片的核心标题和要点内容。

第四步：笔记排版与更新

1. 将刚才的截图保存至本地目录。
2. 在新建的笔记文件中追加写入：
   • ### [提取出的幻灯片标题]
   • 插入图片: [截图路径]
   • [提取出的幻灯片要点/正文]
3. 将当前画面设为新的“基准帧”，返回第二步继续监控。

⚠️ 安全与操作红线

1. 静默操作： 严禁移动用户的鼠标或干扰当前视频的播放状态。
2. 隐私过滤： 仅提取占据画面主体的 PPT/幻灯片内容，忽略弹幕、私人聊天窗口或其他非课程相关的屏幕区域。