The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/fetch_song.py executes system commands including curl, yt-dlp, and ffprobe via subprocess.run to manage audio downloads and metadata probing. These calls are implemented using list-based arguments without a shell, which mitigates common command injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The skill connects to tonzhon.com for song searches and lyric retrieval. It also fetches audio from music.163.com and uses yt-dlp to download content from well-known services like SoundCloud and YouTube. These operations are essential for the skill's purpose of creating synchronized study materials.
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection (Category 8) because it retrieves and processes untrusted third-party content (lyrics and metadata) from the internet.
Ingestion points: External data enters the agent context via scripts/fetch_song.py output (skeleton.json).
Boundary markers: No explicit instructions or delimiters are used to warn the agent against executing instructions that might be embedded within the retrieved lyrics.
Capability inventory: The skill generates HTML files (scripts/build_study.py) and allows the agent to perform complex text transformations on the data.
Sanitization: While the generated HTML is escaped to prevent Cross-Site Scripting (XSS), the agent lacks semantic sanitization to identify or ignore malicious prompts hidden in the song data.

japanese-deep-translate