The Agent Skills Directory

Prompt Injection Defenses: The skill instructions explicitly direct the agent to treat all user-provided text, manifests, and caption files as raw data. This design pattern is intended to prevent the agent from executing or following instructions that might be embedded in external content.
Subprocess Integration: The skill uses subprocess.run to invoke ffmpeg and ffprobe for video manipulation. These operations are performed using structured argument lists, which is the standard method for executing system utilities within the skill's media processing context.
Indirect Prompt Injection Awareness: The skill identifies processed media and script files as untrusted data sources. It provides specific guidance to ensure that content from these sources is handled as data only, mitigating risks associated with processing potentially malicious external inputs.

subtitles