The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/loudnorm.sh executes ffmpeg and ffprobe to process media files. These commands use user-provided file paths for input and output.
[DATA_EXFILTRATION]: While the script performs file operations, it does not involve any network calls or access to sensitive system directories (e.g., .ssh, .aws). All file writes are limited to the user-specified output directory.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted media files and user-defined parameters which are interpolated into command-line arguments.
Ingestion points: Media files provided via --input and loudness targets provided via --target-i, --target-tp, and --target-lra in scripts/loudnorm.sh.
Boundary markers: None identified for parameter interpolation.
Capability inventory: File reading, writing, and directory creation via ffmpeg and mkdir in scripts/loudnorm.sh.
Sanitization: The script performs basic validation on the --platform preset but does not strictly validate the numeric format of individual target parameters before they are interpolated into the FFmpeg filter string.

normalize-loudness