The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes ffmpeg and ffprobe for media processing tasks using the Python subprocess.run method. Arguments are passed as lists rather than strings, which is a standard security practice that mitigates shell injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-controlled media specifications and subtitle data as part of its core functionality.
Ingestion points: Media Edit Decision Lists (JSON) and subtitle files (SRT or JSON) are loaded and parsed in scripts/montage.py and scripts/composite.py.
Boundary markers: Not present; the data is treated as structured media metadata for processing.
Capability inventory: The skill spawns local media processing tasks via the host's FFmpeg installation.
Sanitization: The skill validates numeric inputs for timing and layout (e.g., using parse_time and _num helpers), effectively restricting the input surface to valid media parameters.
[DYNAMIC_EXECUTION]: The scripts dynamically construct complex FFmpeg filter strings (such as filter_complex) at runtime to apply transitions, panning, and audio ducking. This dynamic generation is scoped to the media processing domain and is a necessary feature for programmatic video editing.

stitch-videos-ffmpeg