google-lyria
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted user data, including music prompts, lyrics, and image moodboards, to construct API calls to the Google Lyria service. This creates a potential surface for indirect prompt injection attacks where malicious instructions hidden in the user data could influence the agent's behavior.
- Ingestion points: Music prompts, lyrics, and image URIs are ingested into the interaction input array for Lyria 3 models (SKILL.md).
- Boundary markers: The skill suggests a 'Lyrics:' prefix for song text but lacks explicit markers to prevent the model from executing instructions potentially found within user-supplied content.
- Capability inventory: The skill constructs and sends HTTP POST requests to the Google AI Platform API (SKILL.md).
- Sanitization: The instructions focus on manual rights review and brand compliance rather than technical sanitization or escaping of the user-provided data before it is interpolated into API request payloads.
Audit Metadata