The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to construct and execute shell commands using ffmpeg and ffprobe. It builds these commands dynamically based on user-provided inputs such as file paths, output formats, and quality settings.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the user in the form of file paths which are then interpolated into shell command templates.
Ingestion points: The assistant accepts a user-provided file path ($INPUT_FILE) in Step 1 and Step 4 of SKILL.md.
Boundary markers: The instructions attempt to use double quotes around variables (e.g., "$INPUT_FILE", "INPUT") to delimit inputs, which provides a basic layer of protection but may be vulnerable to specific shell metacharacters or expansion techniques depending on the execution environment.
Capability inventory: The skill has the ability to read file metadata via ffprobe and write new files via ffmpeg.
Sanitization: There is no explicit sanitization or validation logic described to ensure the file path does not contain malicious shell commands or characters beyond a basic existence check.

audio-extract