Find Reference Video via Frameset

从 frameset.app 搜索视频参考片段，找到合集页面和原视频链接。用于： (1) 根据关键词搜索广告/电影片段参考，(2) 获取原视频 YouTube/Vimeo 链接，(3) 下载视频到本地。

Security Scan

The skill's instructions mostly match its stated purpose (search frameset.app and fetch original video links) but there are inconsistencies around execution requirements (it tells the agent to run yt-dlp and perform browser hover/click actions yet declares no required binaries or install steps), so the package lacks proportional declarations and operational clarity.

Purpose & Capability

The skill's described purpose is to find reference frames and fetch original YouTube/Vimeo links and optionally download videos. The SKILL.md explicitly requires running yt-dlp for downloads and performing browser interactions (hover, click, screenshot). However the skill metadata declares no required binaries, no install steps, and no capabilities (e.g., browser automation) — a mismatch between what the skill instructs and what it declares as required.

Instruction Scope

The runtime instructions stick to the declared intent: search frameset.app, show screenshots to the user for selection, follow collection pages to get original links, list them, and optionally download chosen videos. The instructions do not request unrelated system credentials or other sensitive files. They do require taking screenshots and writing downloads to ~/Downloads, which is within scope but should be explicitly documented.

Install Mechanism

There is no install spec even though the skill explicitly calls a third‑party tool (yt-dlp) for downloads. The absence of a declared dependency or guidance on ensuring yt-dlp and a browser automation capability are present is a gap: the skill may fail at runtime or an agent may attempt to run commands that aren't available. Instruction-only skills are low-risk when they accurately describe needed tools; here they do not.

Credentials

The skill requests no environment variables or credentials, which is appropriate. It does instruct saving files to ~/Downloads and performing network requests to frameset.app / youtube.com / vimeo.com. These actions require filesystem write and outbound network access but no secrets — reasonable for the stated purpose, but should be disclosed to users and documented.

Persistence & Privilege

The skill does not request persistent presence (always: false) and does not modify other skills or system-wide settings. Allowing autonomous invocation is the platform default and is not a unique concern here.

What to consider before installing

This skill appears to do what it says (search frameset.app, show screenshots, get original YouTube/Vimeo links, and optionally download). However: (1) the SKILL.md expects the agent to run yt-dlp and perform browser hover/click/screenshot actions but the package does not declare yt-dlp or any browser automation dependency — ask the publisher to explicitly list required binaries (yt-dlp) and the browsing/automation capability (Playwright/Selenium or a platform browser tool). (2) Confirm that your agent environment has outbound network access and permission to write to ~/Downloads, and that yt-dlp is trusted/installed from a reputable source. (3) Be aware this skill downloads media: ensure you have rights/permission before downloading copyrighted content. If the publisher updates the skill to list required tools and installation steps, the inconsistencies would be resolved and my assessment would be more positive.

Find Reference Video - 找参考视频

使用方式：用户说"帮我找一个和[话题]相关的参考片"，即调用此 skill。

完整流程

1. 搜索 frameset

直接搜索关键词，不需要问用户要几个。使用正确 URL 格式：https://frameset.app/search?search=关键词

2. 发送截图给用户确认

重要：不要直接选图片！先截取搜索结果页面发给用户看，让用户告诉你需要哪些编号（编号规则：从左到右，从上到下）。

3. 处理用户选择的编号

• 如果用户给的编号超过10个 → 提醒"数量较多，是否确认？"
• 如果用户执意超过15个 → 最多只找15个（取前15个）

4. 根据用户选择的编号获取合集页面链接

不要点击图片！ 而是：

• 鼠标 hover 到图片上（图片左上角会出现可点击的链接）
• 点击链接进入合集页面

5. 获取原视频链接

• 合集页面有 "Watch" / "Open in YouTube" / "Open in Vimeo" 按钮
• 点击获取视频 URL

6. 列出表格

用 Markdown 表格列出结果：

#	合集名称	合集链接	原视频链接
1	xxx	https://frameset.app/...	https://...
2	xxx	https://frameset.app/...	https://...

7. 询问是否下载

"需要下载哪些？告诉我编号（如：1,3,5）"

8. 执行下载

根据用户给的编号，下载对应视频到 ~/Downloads/

数量限制规则

• 超过 10 个 → 提醒用户"数量较多，是否确认？"
• 超过 15 个 → 最多只找 15 个

视频下载命令

# YouTube
yt-dlp -f "best[ext=mp4]/best" -o "视频名.%(ext)s" "视频链接"

# Vimeo
yt-dlp -f "best[ext=mp4]/best" -o "视频名.%(ext)s" "视频链接"

注意事项

• frameset 里的内容都是电影/广告的静帧
• 选择静态图片，跳过 GIF 动图（URL 含 _motion）
• 关键：搜索结果页面，鼠标悬停(hover)到图片上，不要点击，图片左上角会出现链接
• 每个步骤完成后都要等页面加载完成再操作