The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the execute_command tool to run a local Python script (scripts/mp-downloader.py) that acts as a bridge between the agent and various media downloader APIs.
[DATA_EXFILTRATION]: The tasks.add.direct action provides a mechanism to read arbitrary local files from the filesystem via the torrent_file parameter and the Path.read_bytes() method. This capability could be used to expose sensitive local content if the agent is directed to read non-torrent files.
[DYNAMIC_EXECUTION]: The helper script employs importlib.import_module to dynamically load provider client classes. While these imports are restricted to a hardcoded mapping of MoviePilot internal modules, this pattern involves dynamic loading of code from computed paths at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external downloader instances, such as torrent metadata and tracker information, through actions like tasks.list and tasks.trackers. Boundary markers for this ingested data are absent in the skill's instructions. The skill possesses significant capabilities including file movement and network submission. Sanitization is present in the form of credential filtering in _jsonable, but it does not filter potential natural language instructions embedded in metadata.
[DATA_EXPOSURE]: The script implements a safety mechanism (_is_sensitive_key) to detect and redact potential credentials such as passwords, tokens, and cookies from the JSON results returned to the agent, reducing the risk of accidental secret exposure.

downloader-operation