The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/elevenlabs.py script utilizes the subprocess.run function to execute the ffmpeg utility for audio concatenation. The execution is handled through a manifest file generated in a temporary directory, ensuring the command construction remains isolated from direct user input strings.
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection by processing external documents provided by users.
Ingestion points: The scripts/extract.py script reads and extracts text from various file formats (PDF, DOCX, MD, TXT).
Boundary markers: There are no explicit delimiters or system-level instructions defined in the workflow to differentiate between original document content and potential malicious instructions during the podcast script generation phase.
Capability inventory: The agent context includes the ability to make network requests to the ElevenLabs API and execute ffmpeg via the command line.
Sanitization: The clean_text function in scripts/extract.py provides some mitigation by stripping metadata, URLs, code blocks, and HTML tags using regular expressions, though linguistically-based instructions might persist.

elevenlabs