The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the douyin-downloader tool from an external GitHub repository (github.com/jiji262/douyin-downloader.git) which is not a recognized trusted vendor.- [REMOTE_CODE_EXECUTION]: The installation instructions in SKILL.md and logic in scripts/video_downloader.py perform a git clone followed by uv sync, which involves executing code from a third-party repository to set up the environment. Additionally, scripts/video_downloader.py uses yaml.load() to process configuration files, which can be an unsafe operation if the source is untrusted.- [COMMAND_EXECUTION]: The skill uses subprocess.run to call system binaries like git, uv, and yt-dlp with arguments derived from user-supplied URLs.- [CREDENTIALS_UNSAFE]: The skill includes functionality to automate the harvesting of sensitive session cookies (specifically hy_token and hy_user) from a browser session during the WeChat Channels login process.- [PROMPT_INJECTION]: The skill processes untrusted user-supplied URLs which are passed to downstream command-line tools, creating a surface for indirect prompt injection. 1. Ingestion points: URLs provided by users via the download command. 2. Boundary markers: None; input URLs are passed directly to subprocesses. 3. Capability inventory: Subprocess execution, file system writes, and network downloads. 4. Sanitization: Output filenames are sanitized, but input URLs are not validated for malicious command flags or payloads.

video-downloader