The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to install official ElevenLabs packages via npm (e.g., @elevenlabs/react, @elevenlabs/client, @elevenlabs/agents-cli). It also includes a template in assets/widget-embed-template.html that loads a conversation widget from ElevenLabs' official domain. These references are to well-known, trusted service providers.
[INDIRECT_PROMPT_INJECTION]: As a framework for building conversational AI agents, the skill inherently manages untrusted user input (voice and text transcripts).
Ingestion points: Data enters the system via SDK event handlers (onEvent) and the conversation widget as documented in SKILL.md and assets/react-sdk-boilerplate.tsx.
Boundary markers: The skill specifically instructs developers to include a 'Guardrails' section in system prompts to define boundaries and ethical constraints (see assets/system-prompt-template.md).
Capability inventory: The agents can execute client-side tools (e.g., navigate using window.location.href) and server-side webhooks (e.g., fetching weather or order data).
Sanitization: The skill demonstrates best practices by using Zod schema validation for tool parameters (e.g., z.string().url()) in both SKILL.md and assets/react-sdk-boilerplate.tsx to ensure input conforms to expected formats before processing.

elevenlabs-agents