elevenlabs-agents

Pass

Audited by Gen Agent Trust Hub on Jun 14, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches official SDK packages from the @elevenlabs npm scope and loads the conversational AI widget from the official elevenlabs.io domain.
  • [COMMAND_EXECUTION]: Provides shell scripts for managing AI agents via the official ElevenLabs command-line interface, supporting workflows for project initialization, deployment, and automated testing.
  • [DATA_EXFILTRATION]: Documents secure practices for handling sensitive credentials by using environment variables and configuration placeholders; it also provides implementation templates for HMAC SHA-256 signature verification to secure webhook endpoints.
  • [PROMPT_INJECTION]: Identifies a surface for indirect prompt injection where agents process untrusted data from external sources.
  • Ingestion points: Data enters the agent context through dynamic variables, user transcripts, and RAG knowledge base documents as referenced in references/api-reference.md.
  • Boundary markers: Instructions in references/system-prompt-guide.md advise defining 'Guardrails' and 'Success Criteria' within system prompts to control agent behavior.
  • Capability inventory: Shell scripts in scripts/ utilize CLI commands for platform synchronization, while SDK boilerplates perform network operations to ElevenLabs APIs and allow custom tool execution.
  • Sanitization: Provides implementation examples using the Zod library for parameter validation in client-side tool handlers in assets/react-sdk-boilerplate.tsx.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 14, 2026, 09:40 AM
Security Audit — agent-trust-hub — elevenlabs-agents