The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs the agent to extract and utilize sensitive browser cookies from Chrome, Firefox, or Safari using the --cookies-from-browser flag to bypass bot detection. This grants the agent access to the user's active session credentials and authentication tokens.
[EXTERNAL_DOWNLOADS]: The script scripts/download_video.py automatically performs dynamic installation of third-party Python packages (bgutil-ytdlp-pot-provider, yt-dlp-getpot-wpc) and pulls a remote Docker image (brainicism/bgutil-ytdlp-pot-provider) during execution. These resources originate from unverified third-party developers.
[COMMAND_EXECUTION]: The skill heavily relies on executing shell commands and system binaries via a Python wrapper. It manages external tools including yt-dlp, ffmpeg, docker, and pip through subprocess.run calls, which can be misused if input sanitization fails.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: Untrusted data from YouTube, including video titles, thumbnails, and transcripts (subtitles), are fetched and processed in the agent's context in scripts/download_video.py and references/text-first.md.
Boundary markers: The skill does not use delimiters or instructions to prevent the agent from obeying instructions that might be embedded in the retrieved metadata or transcripts.
Capability inventory: The agent has the capability to write files, execute shell commands, and install software, making it vulnerable to multi-step attacks if the ingested data is malicious.
Sanitization: There is no evidence of sanitization or filtering of external content before it is processed by the agent.

youtube-downloader