synthesia-interactive-avatar
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes live audio and text input from users to generate agent responses and avatar animations. * Ingestion points: User voice input is ingested via LiveKit STT or OpenAI Realtime API components. * Boundary markers: The instructions provide behavioral guidelines for the agent but do not define explicit input sanitization or boundary markers for untrusted user data. * Capability inventory: The skill interacts with external APIs (Synthesia, LiveKit, OpenAI) and performs network operations to stream audio/video data. * Sanitization: No specific sanitization or filtering of external content is mentioned; the skill relies on the underlying LLM's safety protocols.
- [REMOTE_CODE_EXECUTION]: The documentation suggests installing the LiveKit CLI using a piped shell script (curl -sSL https://get.livekit.io/cli | bash). This operation targets the official domain of LiveKit, which is a well-known service.
- [COMMAND_EXECUTION]: The skill involves executing shell commands for project initialization (uv init), package installation (pip install), and running the agent worker (python agent.py dev).
- [EXTERNAL_DOWNLOADS]: The skill downloads and installs Python packages from PyPI and refers to official configuration resources from Synthesia's GitHub repository and developer documentation.
Audit Metadata