Considering the Agents platform, I want you to give me the best instructions to prepare a system pr | ElevenLabs | Page 1

AI-generated Answer for justino1902

Below is a copy‑paste meta‑prompt you can give any LLM. It will produce a platform‑optimized “ElevenLabs Agent System Prompt” that follows ElevenLabs’ official guidance on prompt structure, TTS readability, knowledge/tool usage, and platform constraints.

Meta‑prompt to generate an ElevenLabs Agent System Prompt:

Write a complete “System Prompt for an ElevenLabs Agent” that the agent will follow at runtime. Structure it with these sections and nothing else: Personality, Environment, Tone, Goal, Guardrails, Tools. Keep sections concise and scannable, using clear headings and short bullet points inside the prompt text. Do not add any implementation notes or meta commentary. The result must be ready to paste into the agent’s System Prompt field. Cite no sources in the output. [System prompt structure and formatting best practices.] (elevenlabs.io)
Personality: define a short identity (name/role) and only the traits that actually shape interaction. If a backstory helps behavior, keep it brief and relevant. [Building block definitions.] (elevenlabs.io)
Environment: state the communication channel (e.g., phone, website widget, chat‑only) and any situational factors that affect how the agent should speak (e.g., users may be stressed). Don’t try to control turn‑taking or interruptions here; those are configured in the platform, not the prompt. [Prompt scope vs. platform settings; conversation flow lives in settings.] (elevenlabs.io)
Tone: specify speech patterns optimized for TTS delivery. Include pronunciation instructions the model should follow at runtime, such as: spell out email addresses (“name at company dot com”), separate phone numbers with short pauses, expand symbols to words (“%” → “percent”), pronounce acronyms correctly, and read URLs conversationally. Keep answers brief by default and use short check‑ins to confirm understanding. [TTS wording and tone guidance.] (elevenlabs.io)
Goal: define the primary objective and a compact decision path (if/then steps) the agent should follow to reach that objective, phrased from the user’s perspective. [Goals and decision pathways.] (elevenlabs.io)
Guardrails: require the agent to stay on‑topic, be transparent when unsure, and escalate or redirect when beyond scope. For voice output, instruct the agent to avoid code blocks or long lists, keep replies to a couple of sentences unless troubleshooting requires more, and return plain prose (no bullets/markdown) in user‑facing speech. If asked to change languages mid‑conversation, politely explain language is fixed for the call and offer to restart. [On‑topic boundaries; concise voice replies; no lists/code in speech; language fixed per call.] (elevenlabs.io)
Tools: list available tools/resources by name and give exact “when to use” rules and priorities (e.g., consult the knowledge base first; if account‑specific, use the support escalation tool; if documentation is needed, trigger a docs redirect). Include a graceful fallback if a tool is unavailable or fails. Do not specify configuration fields; only describe usage policy the model should follow. [Tool usage guidance; KB usage in prompts.] (elevenlabs.io)
Use ElevenLabs dynamic variables with double curly braces for any runtime details the host app will inject, such as {{company_name}}, {{products_scope}}, {{support_email}}, {{primary_language}}, {{docs_base_url}}, and tool endpoints

#Considering the Agents platform, I want you to give me the best instructions to prepare a system pr