The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the command-line tools yt-dlp and ffmpeg for its core functionality. It instructs the agent to verify these tools are installed and suggests installation via brew if necessary.
[EXTERNAL_DOWNLOADS]: The skill downloads video content from external YouTube URLs. These downloads are essential to the primary purpose of the skill and target a well-known service (YouTube).
[PROMPT_INJECTION]: The skill ingests untrusted external data (YouTube videos and potentially their metadata or transcripts) which is analyzed by an LLM-based skill (watch:watch). This presents an indirect prompt injection surface.
Ingestion points: External YouTube URLs provided via the video_urls parameter in SKILL.md.
Boundary markers: No explicit delimiting or instructions to ignore embedded commands are present in the analysis workflow.
Capability inventory: The skill has the capability to write files to the local filesystem and execute video manipulation commands.
Sanitization: The skill relies on the analysis provided by the watch:watch tool and does not explicitly define secondary validation or sanitization of the extracted data.

build-brand-clip-library