The Agent Skills Directory

[COMMAND_EXECUTION]: The script uses subprocess.run with list-based arguments for ffmpeg and ffprobe calls. This is the recommended security practice to prevent shell injection, as it bypasses the shell and treats arguments as literal strings.
[DATA_EXPOSURE]: The skill uses a manifest file (manifest.json) to track operations. It logs local file paths and status, which is standard for build/processing tools and does not constitute unauthorized data exposure.
[REMOTE_CODE_EXECUTION]: All code is executed locally. The skill depends on ffmpeg and ffprobe being pre-installed on the system, but does not attempt to download or execute external code at runtime.
[PROMPT_INJECTION]: The instructions are clearly scoped to audio editing tasks. There are no patterns suggesting attempts to override agent behavior or bypass safety filters.

insert-silence-beat