The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill assembles Python code blocks and shell commands at runtime by interpolating user-supplied arguments (e.g., <input>, <min-speech>, <min-silence>). This generated code is intended for execution in the user's local environment to perform audio analysis.
[COMMAND_EXECUTION]: The skill relies on external calls to ffmpeg for audio processing and file segmentation. These commands are constructed dynamically, creating a dependency on the system's shell environment and the security of the provided file paths.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection because it ingests untrusted data that influences its execution flow.
Ingestion points: User-provided audio file paths and potential audio metadata accessed via $ARGUMENTS in SKILL.md.
Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands within the processed file paths.
Capability inventory: The skill possesses capabilities for file system read/write operations, Python script execution (via torch and silero-vad), and shell command execution via ffmpeg.
Sanitization: The instructions do not specify any sanitization, escaping, or validation logic for the input strings before they are interpolated into executable contexts.

vad