livekit
Pass
Audited by Gen Agent Trust Hub on Jul 17, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill establishes a voice-to-text pipeline that processes untrusted audio data from external participants, creating a surface for indirect prompt injection.
- Ingestion points: Participant audio is captured via 'ctx.wait_for_participant()' and 'assistant.start(ctx.room, participant)' in SKILL.md.
- Boundary markers: No explicit boundary markers or instructions to ignore malicious embedded commands are present in the provided agent boilerplate.
- Capability inventory: The agent is configured with 'fnc_ctx=MyFunctions()', enabling it to execute potentially sensitive tools or functions based on speech transcription.
- Sanitization: The Speech-to-Text (STT) output is passed directly to the LLM without identified sanitization or filtering logic.
Audit Metadata