The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a local script scripts/sfx_tool.py via the terminal tool to perform audio post-processing. While the script itself follows security best practices by using list-based subprocess.run calls (avoiding shell execution), the initial invocation by the agent involves interpolating file paths into a shell command. If the AI agent does not adequately sanitize these file paths, an attacker could potentially achieve command injection.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes user-provided prompts for sound generation and handles file metadata. Although the skill does not currently ingest high-risk external data (like web content), the capability to execute terminal commands and write to the filesystem makes this surface relevant.
Ingestion points: User-supplied prompts for sound generation and audio filenames.
Boundary markers: None identified in the prompt templates to distinguish between instructions and data.
Capability inventory: File system read/write, directory creation, ZIP archive generation, and execution of ffmpeg/ffprobe via subprocesses.
Sanitization: The sfx_tool.py script uses a _slug function to sanitize some output names and handles argument injection by suggesting the use of ./ or -- for filenames starting with dashes.

elevenlabs-sound-design