The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill handles external speech timing data (JSON, SRT, or VTT files) that could potentially contain malicious content meant to influence downstream tools.
Ingestion points: The scripts/duck_envelope.py utility reads timing data from files specified via a command-line path argument.
Boundary markers: The utility operates as a standalone data transformer with no direct LLM instruction interface.
Capability inventory: The script's output is designed to be used as a filter string within ffmpeg shell commands.
Sanitization: The implementation is highly robust; all data extracted from the input files (start times, end times, and gain levels) is explicitly converted to numeric types (float or int) before being interpolated into the output string. This prevents any non-numeric content or potential command injection sequences from reaching the ffmpeg execution environment.

audio-ducking