The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references official ElevenLabs SDKs (@elevenlabs/elevenlabs-js, elevenlabs) and well-known libraries such as openai, dotenv, and livekit-client. These are trusted dependencies for the described real-time audio functionality.
[DATA_EXFILTRATION]: No unauthorized data exfiltration was detected. The skill correctly instructs users to manage sensitive API keys via environment variables and explicitly warns against exposing credentials in client-side browser code.
[COMMAND_EXECUTION]: The documentation mentions standard utility commands for local development, such as ngrok for creating secure tunnels and package managers for installation. No malicious shell patterns or arbitrary command executions were identified.
[PROMPT_INJECTION]: The skill processes user audio transcripts as input for an LLM. While this constitutes a standard indirect prompt injection attack surface inherent to chat and voice interfaces, the provided implementation follows standard patterns and does not expose high-risk tools or bypass safety guidelines.

speech-engine