The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied media files, which provides a surface for indirect prompt injection attacks.
Ingestion points: avatar_image and audio_file parameters defined in the input schema in SKILL.md.
Boundary markers: No explicit delimiters or instructions are present to prevent the agent from following potential commands hidden within the ingested files.
Capability inventory: The skill is capable of executing shell commands via the refly CLI to interact with the platform API and use the open command to display generated files.
Sanitization: No input validation or sanitization of the content within the referenced files is mentioned.
[SAFE]: All external communications and file downloads are restricted to the vendor's domain (refly.ai). The local file operations and command execution are entirely consistent with the skill's primary function of generating and delivering media content.

volcengine-avatar