agents
Pass
Audited by Gen Agent Trust Hub on Nov 14, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs users to download and install official packages from ElevenLabs, a well-known service provider, using standard package managers.
- Evidence includes installation of
@elevenlabs/cli,@elevenlabs/elevenlabs-js, and theelevenlabsPython package. - The widget integration fetches the conversation embed script from a well-known CDN (unpkg.com).
- [COMMAND_EXECUTION]: The skill documents the use of the ElevenLabs CLI for project initialization and agent management.
- Examples include
elevenlabs agents init,elevenlabs agents push, and authentication commands. - [INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of AI agents that ingest untrusted user audio and text during real-time conversations, which are then processed by an LLM.
- Ingestion points: Agents receive real-time voice and text input from users during active sessions, as documented in
SKILL.mdandreferences/agent-configuration.md. - Boundary markers: The configuration schema in
references/agent-configuration.mdincludes specific safety controls underplatform_settings.guardrails, allowing developers to enable prompt injection detection. - Capability inventory: The agents can execute arbitrary webhooks (
references/client-tools.md), run client-side JavaScript in the user's browser, and initiate outbound phone calls via Twilio integration (references/outbound-calls.md). - Sanitization: The platform provides a
sanitizeoption for dynamic variables to remove sensitive data from tool responses before they reach the LLM or transcript, as noted inreferences/outbound-calls.md. - [DATA_EXPOSURE]: The documentation in
references/installation.mdnotes that the ElevenLabs CLI stores API keys in a local configuration file at~/.agents/api_keys.json. This is a description of the tool's standard credential management behavior.
Audit Metadata