The Agent Skills Directory

[COMMAND_EXECUTION]: The download_article_video.py script uses subprocess.run to execute the agent-browser CLI for page interaction and ffprobe for media metadata extraction. The arguments passed to these commands are either static or undergo validation (e.g., article URL host check).
[EXTERNAL_DOWNLOADS]: The skill downloads media files from mpvideo.qpic.cn using Python's urllib.request. The script strictly validates that the media URL is an HTTPS link targeting the allowed hostname. It also verifies that the downloaded content has a valid MP4 (ftyp) header before finalizing the file write.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from external WeChat articles, which represents a potential surface for indirect prompt injection. This risk is addressed in SKILL.md by instructions that explicitly warn the agent to treat page content as untrusted and not to execute instructions found within the page. The backend script further mitigates this by restricting media extraction to a specific domain (mpvideo.qpic.cn).

wechat-article-video-download