The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface where user-supplied text prompts are passed to the voice synthesis engine. If untrusted data is processed without sanitization, it could attempt to influence the agent's behavior through instructions embedded in the speech content. However, this is a baseline risk inherent to the primary functionality of the skill.\n
Ingestion points: Untrusted text enters the context through the prompt and text fields in the belt app run command inputs within SKILL.md.\n
Boundary markers: No specific delimiters (e.g., triple backticks or XML-style tags) are demonstrated in the examples to isolate the text being synthesized.\n
Capability inventory: The skill is authorized to use the belt CLI via the Bash tool to perform network-based AI inference.\n
Sanitization: The documentation does not specify sanitization or escaping steps for input prompts.\n- [EXTERNAL_DOWNLOADS]: The skill references external installation instructions and related skills from the vendor's repository on GitHub.\n
Evidence: Provides a link to cli-install.md on the inference-sh GitHub organization and suggests adding platform-specific skills using npx skills add to extend functionality.

ai-voice-cloning