Video Understanding

视频理解与分析能力 - 让 AI 能够理解视频内容、提取关键信息。当用户要求分析视频、理解视频内容、总结视频、提取视频要点时触发此技能。

Security Scan

The skill is internally consistent for video metadata extraction and summarization: it references appropriate video APIs and optional tools (yt-dlp) and does not request unrelated credentials or persistent privileges.

Purpose & Capability

The name/description (video understanding, summarization, extracting titles/descriptions/duration) matches the instructions and resources which reference YouTube/Bilibili/Tencent APIs and optional download tools. Nothing requested or described is unrelated to video analysis.

Instruction Scope

Instructions remain within the stated purpose (identify platform, fetch metadata, extract subtitles/audio, summarize). They mention using third-party APIs and downloading audio/subtitles for speech-to-text, but do not instruct reading unrelated local files or exfiltrating data. Note: using download + STT may cause the agent to fetch and store video/audio data — a privacy consideration rather than an incoherence.

Install Mechanism

The skill is instruction-only and includes no install spec (low static risk). The resources advise installing yt-dlp via pip for downloads; that is an optional external package installation the user/agent would perform and is not bundled. Installing third-party packages (pip/ PyPI) is a normal choice but increases attack surface if done automatically — the bundle itself does not perform that install.

Credentials

The skill declares no required env vars or credentials. The references correctly show where a YouTube Data API key would be used if the agent opts to call that API. No unrelated secrets or config paths are requested.

Persistence & Privilege

The skill does not request persistent presence (always:false), does not modify other skills/config, and does not declare privileged system access. Autonomous invocation is allowed by platform default but is not combined with other high-risk factors here.

Assessment

This skill appears to do what it says: fetch video metadata and summarize content. Things to consider before installing or using it: - If you want full transcription/speech-to-text, the skill suggests downloading audio/subtitles and using extra tools (e.g., yt-dlp plus an STT service). Installing yt-dlp (pip) or using an external STT service can download/process user content and may send it to third-party servers — only proceed if you trust those tools/services. - The YouTube Data API example needs an API_KEY; only provide API keys you trust and limit their scope. The skill itself does not request any credentials, but you may be prompted to supply them if you want API-based metadata rather than simple web scraping. - Downloading videos can touch copyrighted or private material. Ensure you have permission before downloading or transcribing content. - If you require stricter safety, ask for an explicit statement of where transcriptions or downloaded media will be sent (local-only processing vs. remote transcription API) and whether the agent will persist files on disk. Overall this skill is coherent and proportional to its purpose, but be mindful of optional external installs and transcription service choices.

Video Understanding - 视频理解技能

概述

赋予 AI 理解和分析视频内容的能力，支持：

• 视频链接分析（YouTube、Bilibili等）
• 提取视频标题、描述、时长等信息
• 视频内容总结与关键点提取
• 语音转文字（需要额外工具）

触发场景

1. 用户发送视频链接并要求"理解"、"分析"、"总结"
2. 用户询问视频"讲了什么"、"重点是什么"
3. 用户要求"提取"视频内容
4. 需要从视频中获取信息

支持平台

平台	网址	支持情况
YouTube	youtube.com	标题、描述、时长
Bilibili	bilibili.com	标题、简介、BV号
抖音	douyin.com	标题、描述
腾讯视频	v.qq.com	标题、简介
爱奇艺	iq.com	标题、简介

获取视频信息

方法1: web_fetch

url = "https://www.youtube.com/watch?v=视频ID"
# 提取 title, description, duration

方法2: 第三方 API

• RapidAPI Video APIs
• YouTube Data API
• Bilibili API

工作流

1. 识别视频平台 → 判断使用哪个 API/方法
2. 获取基本信息 → 标题、描述、时长、作者
3. 提取关键内容 → 章节、要点、字幕（如有）
4. 整合总结 → 用中文呈现给用户

输出格式

向用户呈现视频信息时：

• 视频标题和来源
• 内容摘要（如果可以获取）
• 关键时间点（如有章节）
• 建议的观看重点

限制

• 无法直接播放视频
• 无法理解视频画面内容
• 需要平台支持或第三方 API