The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes external command-line utilities yt-dlp and ffmpeg to download and process media files. These calls are made using list-based arguments which correctly prevent shell injection.
Evidence: bili_download.py (lines 42, 69, 107, 142) and vocal_extract.py (lines 18, 41, 79).
[EXTERNAL_DOWNLOADS]: The skill downloads media content from Bilibili servers. While it interacts with external URLs, this is the primary intended function of the skill.
Evidence: bili_download.py (line 148).
[CREDENTIALS_UNSAFE]: The skill supports an optional --cookies flag to allow yt-dlp to read browser session data. This is used to authenticate with Bilibili for high-quality downloads but involves access to sensitive local browser storage.
Evidence: bili_download.py (line 139) and SKILL.md (lines 47, 60).
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests untrusted metadata (such as video titles and descriptions) from external Bilibili pages during the download process.
Evidence: bili_download.py (line 151).
Ingestion points: Video metadata fetched via yt-dlp from the provided URL.
Boundary markers: None identified in the script output logic.
Capability inventory: File system write access and command execution via ffmpeg.
Sanitization: No explicit sanitization of external metadata is performed before displaying it in the agent context.

bilibili-downloader