The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and processes untrusted data from the user to drive its workflow.
Ingestion points: Untrusted data enters the agent context through the user-supplied source audio path, target voice IDs, and consent metadata as described in the 'Request triage' section of SKILL.md.
Boundary markers: The instructions lack explicit boundary markers or delimiters to separate processed data from instructions, increasing the risk that the agent could misinterpret data content as commands.
Capability inventory: The skill utilizes the voice_convert tool and requests filesystem-write and network-read capabilities in its metadata, providing a mechanism for file manipulation and network interaction based on user-controlled inputs.
Sanitization: There are no instructions provided for sanitizing or validating user-provided file paths or metadata strings to prevent path traversal or other injection-based attacks.

voice-conversion-studio