cartesia-line

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The skill’s Line VoiceAgentApp runtime processes incoming call/user speech via UserTextSent events (and call metadata/system prompt fields) and forwards that text into the LLM as conversation history, so outsider-authored user prompts can directly reach the agent.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The guide includes an installation command that at runtime fetches and pipes a remote shell script into sh (curl -fsSL https://cartesia.sh | sh), which downloads and executes external code required to install the Cartesia CLI.