The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/ytopen.py executes the ffmpeg command to encode raw video frames into an MP4 file.
Evidence: The script uses subprocess.Popen with a fixed list of arguments to pipe raw video data to the ffmpeg binary. This is standard practice for video generation tools.
[INDIRECT_PROMPT_INJECTION]: The skill handles user-provided text for headlines, channel branding, and status chips which could theoretically contain malicious instructions.
Ingestion points: Text is ingested via command-line arguments (sys.argv) and environment variables (e.g., BRAND_URL, OPEN_EYEBROW) in both ytopen.py and ytthumb.py.
Boundary markers: None identified in the prompt templates.
Capability inventory: The skill can execute shell commands (ffmpeg) and interact with the browser via external tools mentioned in SKILL.md.
Sanitization: Input text is processed by the Pillow (PIL) library for graphical rendering. Because the data is rendered as pixels rather than being interpreted as code or logic, the risk of injection is effectively mitigated.
[DATA_EXPOSURE]: The skill documentation and scripts are designed to handle sensitive credentials safely.
Evidence: Instructions for the ELEVENLABS_API_KEY mandate the use of environment variables rather than hardcoding, which aligns with security best practices.

youtube-video-kit