The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of magnet links and torrent files, which can contain embedded instructions in metadata.
Ingestion points: urls (magnet links) and torrent_files parameters in automation scripts (SKILL.md).
Boundary markers: None present to delimit external content from instructions.
Capability inventory: File writing (downloads), torrents_delete, and command execution via bg run (SKILL.md).
Sanitization: No explicit sanitization or validation of torrent metadata is mentioned.
[DYNAMIC_EXECUTION]: The skill uses uv run to execute multi-line Python code provided as a string literal and bg run for background command execution.
Evidence: uv run --with qbittorrent-api python -c "..." used for status checks and automation (SKILL.md).
[EXTERNAL_DOWNLOADS]: Fetches software binaries and packages from well-known services and official repositories.
Evidence: Downloads aria2c binaries from GitHub releases and installs packages via Chocolatey and PyPI (SKILL.md).
[COMMAND_EXECUTION]: Spawns system processes to manage background daemons.
Evidence: Uses qbittorrent-nox and aria2c --enable-rpc to start long-running background tasks (SKILL.md).

torrent-manager