The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's installation instructions recommend executing a shell script directly from a remote source: curl -fsSL https://raw.githubusercontent.com/sonpiaz/watch-cli/main/install.sh | bash. This pattern allows an untrusted third-party repository (sonpiaz/watch-cli) to execute arbitrary code on the user's system.
[COMMAND_EXECUTION]: The skill makes extensive use of CLI tools like yt-dlp, ffmpeg, jq, and curl. It defines several commands (watch, dl-video, extract-frames, transcribe, audio-q) that execute these tools using user-provided video URLs, file paths, and local cookie files.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external social media platforms (YouTube, X, TikTok, LinkedIn) which are not under the user's control.
Ingestion points: Video audio and frames are fetched from external URLs via yt-dlp and processed into text transcripts and image descriptions.
Boundary markers: None. The instructions encourage the agent to incorporate the transcript and frame descriptions directly into its reasoning process.
Capability inventory: The skill patterns suggest the agent should perform tasks like "implement the project shown" or "generate React/HTML+CSS implementation" based on the video content, which involves writing code and creating files.
Sanitization: No measures are described to sanitize or validate the content of the transcripts or frames to prevent them from containing malicious instructions for the agent.

watch-cli-video-agent