The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses external command-line utilities to handle heavy media tasks. Specifically, the script scripts/prep_longform_head.py invokes ffprobe, ffmpeg, and whisper to analyze video streams, perform high-quality transcoding, and generate speech-to-text transcriptions. These operations are essential to the skill's function and are executed using list-based arguments in subprocess.run, which effectively prevents shell injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided audio and video content to generate captions and drive animation timing, which represents a potential surface for indirect injection if the source content contains malicious instructions. However, the risk is minimal as the data is only used for graphical rendering.
Ingestion points: Transcription data is read from audio/words.json and project metadata from source/meta.json in templates/longform_engine.py.
Boundary markers: There are no explicit delimiters used to isolate transcript text from the rendering logic.
Capability inventory: The skill possesses the capability to execute system-level media tools and perform file operations within the project directory and the user's Downloads folder.
Sanitization: Caption text is converted to uppercase during processing but is not filtered for prompt-like strings before being rendered into static image frames using the PIL library.

longform-animated-talking-head