The Agent Skills Directory

[COMMAND_EXECUTION]: The study.py and scaffold_skill.py scripts utilize subprocess.run to execute various system utilities like ffmpeg, ffprobe, yt-dlp, and whisper. The use of list-based arguments for these calls helps prevent shell command injection.
[EXTERNAL_DOWNLOADS]: The skill downloads video files from external URLs provided by the user using the yt-dlp tool. Additionally, it contains logic to trigger a self-update of the yt-dlp binary via the -U flag if it encounters a 403 error during download.
[DYNAMIC_EXECUTION]: scaffold_skill.py dynamically generates a new skill structure, including a SKILL.md file and a render_template.py script. It populates these files using data extracted from the video analysis and copies existing template files from other skill directories.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could contain malicious instructions designed to influence the agent's behavior during the analysis phase.
Ingestion points: The agent is instructed to read the audio/transcript.txt file (generated via Whisper) and visual frame sheets (frames/*.jpg) to perform its style analysis.
Boundary markers: There are no protective boundary markers or instructions to the agent to disregard embedded commands in the analyzed content.
Capability inventory: The skill environment allows for filesystem writes and command execution through the provided scripts.
Sanitization: The external content is not sanitized before the agent is prompted to read and act upon it.

reverse-engineer