The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's instructions direct the agent to execute shell commands (ls -la and uv run) using variable interpolation for the $INPUT_FILE provided by the user. If the agent does not strictly validate or sanitize these inputs before execution, it could lead to command injection vulnerabilities where a malicious filename triggers unintended system commands.
[INDIRECT_PROMPT_INJECTION]: The skill transcribes external audio/video files and presents the resulting text directly to the agent's context for preview. This creates an attack surface where maliciously crafted audio content containing spoken instructions could attempt to influence or override the agent's subsequent behavior.
Ingestion points: The transcribe.py script processes external media files (audio/video) via the whisperx library, reading their content into the agent's workflow.
Boundary markers: The instructions do not define specific delimiters or "ignore embedded instructions" warnings for the agent when displaying the transcription preview.
Capability inventory: The skill possesses shell execution capabilities (via ls and uv run) and file system access (reading media and writing transcript files).
Sanitization: No sanitization, filtering, or escaping is performed on the transcribed text before it is returned to the agent context.

audio-transcribe