The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/cut-silences.mjs executes ffmpeg and ffprobe to analyze and process video files. It uses spawnSync with an array of arguments, which is a secure method that prevents shell command injection.
[DATA_EXPOSURE]: The skill reads local transcript and video files and writes processed assets to a user-specified output directory. It does not perform any network operations or external data exfiltration.
[DYNAMIC_EXECUTION]: The skill dynamically generates an ffmpeg filtergraph script based on silence timings calculated from the transcript. This script is written to a temporary local file and used as an input to the ffmpeg command, which is a standard and safe practice for complex video filtering.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted transcript data (JSON). However, it extracts and validates numeric timing data and does not interpolate untrusted text into prompts or shell commands, minimizing the risk of injection attacks.

cut-silences