T05 · Unauthorized Access and Privilege Escalation
- Location
SKILL.md:29- Finding
Mandatory Browser Cookie Access for YouTube Downloads
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This video downloader is coherent, but it needs review because it tells agents to use browser session cookies for YouTube by default and requests overly broad shell permissions.
Install only if you are comfortable with an agent running yt-dlp/ffmpeg, writing downloaded files locally, and potentially reading browser cookies for YouTube. Prefer trying downloads without cookies first, approve cookie access only when needed, use a dedicated browser profile if possible, and review any pip, brew, or sudo install commands before running them.
SKILL.md:29Mandatory Browser Cookie Access for YouTube Downloads
SKILL.md:126Download Workflow Requests Unrestricted Shell Permissions
SKILL.md:16Unpinned Third-Party Package Installation and Update Instructions
The README explicitly recommends --cookies-from-browser chrome, which causes yt-dlp to read browser cookies from the local profile to authenticate requests. Although this is a legitimate yt-dlp feature rather than malware, in a skill context it is sensitive because it encourages access to browser-stored session tokens without clear warning, potentially exposing account sessions or private content if mishandled.
m/watch?v=xxx" | | Extract audio | "Extract audio from https://youtube.com/watch?v=xxx" | | Download with subtitles | "Download with subtitles https://youtube.com/watch?v=xxx" | | Specific quality | "Download in 720p https://youtube.com/watch?v=xxx" |
You: Download this video https://www.youtube.com/watch?v=xxx
Cursor:
yt-dlp -P "~/Downloads/yt-dlp" --cookies-from-browser chrome "https://www.youtube.com/watch?v=xxx"
yt-dlp supports thousands of sites including:
| Error | Solution |
|---|---|
| HTTP 403 Forbidden (YouTube) | Use --cookies-from-browser chrome |
| yt-dlp not found | Run pip install yt-dlp |
| ffmpeg not found | Run brew install ffmpeg |
| Download fails | Run `pip install -U yt-dl |
The documented use of --cookies-from-browser chrome matches credential/session access behavior because it instructs the tool to read browser cookies from the user's Chrome profile. In this skill context, that is more dangerous because an AI agent may normalize or automate the command, causing users to expose authenticated session material without understanding the sensitivity or downstream handling.
��载这个视频 https://youtube.com/watch?v=xxx" | | 提取音频 | "提取音频 https://youtube.com/watch?v=xxx" | | 下载字幕 | "下载视频和字幕 https://youtube.com/watch?v=xxx" | | 指定画质 | "下载 720p https://youtube.com/watch?v=xxx" |
你: 下载这个视频 https://www.youtube.com/watch?v=xxx
Cursor:
yt-dlp -P "~/Downloads/yt-dlp" --cookies-from-browser chrome "https://www.youtube.com/watch?v=xxx"
yt-dlp 支持上千个网站,包括:
| 错误 | 解决方案 |
|---|---|
| HTTP 403 Forbidden (YouTube) | 使用 --cookies-from-browser chrome |
| yt-dlp 未找到 | 运行 pip install yt-dlp |
| ffmpeg 未找到 | 运行 brew install ffmpeg |
| 下载失败 | 运行 pip install -U yt-dlp 更� |
YARA rule matched a known malware signature (reverse shell, backdoor, ransomware, C2 framework, or info stealer).
If not installed, install them first:
```bash
pip install yt-dlp
brew install ffmpeg # macOS
yt-dlp -P "~/Downloads/yt-dlp" "VIDEO_URL"
YouTube often blocks direct downloads with 403 errors. Always use browser cookies for YouTube:
yt-dlp -P "~/Downloads/yt-dlp" --cookies-from-browser chrome "YOUTUBE_URL"
Supported browsers: chrome, firefox, safari, edge, brave, opera
yt-dlp -P "/path/to/save" -o "%(title)s.%(ext)s" "VIDEO_URL"
yt-dlp -P "~/Downloads/yt-dlp" "VIDEO_URL"
yt-dlp -P "~/Downloads/yt-dlp" -x --audio-format mp3 "VIDEO_URL"
yt-dlp -P "~/Downloads/yt-dlp" --write-subs --sub-langs all "VIDEO_URL"
The README advertises automatic handling of YouTube 403 errors via browser cookies but does not warn that --cookies-from-browser accesses locally stored browser session cookies, which may include authenticated session data. In an agent-skill context, users may follow examples without understanding the privacy and credential exposure implications, increasing the chance of unintended sensitive data access.
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.
mkdir -p ~/.cursor/skills/yt-dlp-downloader/scripts
The README explicitly promotes using browser cookies to bypass YouTube 403 errors but does not warn that this accesses authenticated browser session data. In an agent skill context, this is risky because users may not realize the command reads sensitive browser state, which could expose account/session information or normalize unsafe access to local credentials.
Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.
brew install ffmpeg # macOS
## 安装方法
Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.
brew install ffmpeg # macOS
## 安装方法
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.
创建 skill 目录:
mkdir -p ~/.cursor/skills/yt-dlp-downloader/scripts
从本仓库下载 SKILL.md 和 scripts/download.sh
The skill tells the agent to use browser cookies for YouTube downloads but does not disclose to the user that this reads authenticated browser data. That omission prevents informed consent and can lead to unexpected access to private session material, especially in an agentic environment where users may not realize local browser state is being touched.
The skill repeatedly instructs use of --cookies-from-browser, which causes the agent to access authenticated browser cookie stores from the local machine. For a downloader skill, that reaches beyond the minimally necessary data access and can expose session-bearing credentials or private account context if the command output, logs, or downstream tooling mishandles them.
The workflow requires Shell permissions ['all', 'network'], where all is broader than necessary for downloading media. Overbroad execution privileges increase blast radius: if the skill is misused or combined with malicious input, it could read, modify, or execute unrelated local resources beyond downloading a file.
The workflow directs the agent to perform network downloads and local file writes without a user-facing notice about those side effects. In an automated toolchain, hidden network and filesystem actions reduce transparency and make it easier for a user to authorize behavior they did not fully understand.
The markdown instructs users to run pip install, brew install, and git clone commands that modify the local system, but it does not include any warning or disclosure about making system changes. For markdown files, potentially integrity-affecting behavior should be accompanied by a brief warning so users understand the impact before execution.
No suspicious patterns detected.