System prompt - A System prompt is an instruction set you add to an AI Skill or AI Agent in AI Agent Studio to define its persona, scope, and response style before users interact with it.

Overview

A System prompt is an initial instruction set sent to the AI model before any user input. It establishes the agent's persona, role, and interaction rules, enabling the agent to respond consistently and within the defined scope throughout a conversation.

The System prompt serves the following functions in an AI Agent:

• Consistency: Ensures the agent maintains a consistent persona and response style throughout conversations.
• Context setting: Provides background information and instructions so the agent understands the scope and purpose of its interactions.
• Behavior shaping: Guides the agent's tone, helpfulness, and limitations.
• Efficiency: Establishes reusable instructions at the session level, removing the need to repeat them in every user prompt.
• Task refinement: Specifies output format, language style, or constraints that the model applies to every response in the session.
• Tool utilization: Instructs the model on how to use other tools or features within AI Agent Studio.

What a system prompt can include

A System prompt can define the following aspects of an agent's behavior:

Role definition

A description of the agent's function and the context in which it operates. Role definitions set the agent's area of expertise and its relationship to the user — for example, a customer support representative for a specific product category, or a research assistant specialized in a domain.

Tone and language style

Instructions that control register and formality — for example, whether the agent responds in a professional, conversational, or academic tone, and whether it uses plain language or technical terminology.

Background information

Contextual facts the agent needs to perform its role — for example, the organization's product catalog, the user's experience level, or the operational constraints of the automation.

Specific instructions and constraints

Rules the agent must follow in all interactions — for example, always requesting clarification before answering, following a defined output format, or declining to answer questions outside the defined scope.

AI guardrail and System prompt

When an AI guardrail is applied to an AI Skill, both the System prompt and the User prompt are scanned for sensitive data and toxicity by the Guardrail service.

• Masking: All text within both the System prompt and the User prompt is masked according to the associated Guardrail's settings.
• Combined Toxicity Scoring: The toxicity score returned by the LLM, representing a combined analysis of both the System prompt and the User prompt, is displayed under the prompts on the user interface (UI).

Execution within Guardrail-Enabled Folders:

When you execute an AI Skill within a Task Bot or an API Task located in a folder with an active Guardrail, any System prompt included in the AI Skill is automatically sent to the AI guardrail along with the User prompt for analysis.

Example system prompts

Example 1: Customer support agent

You are a customer support representative for GadgetWorld, an online
electronics retailer. Maintain a polite and professional tone. Assist customers
with questions and resolve issues. If you cannot answer a question, escalate it
to a supervisor.

Example interaction:
User: My order hasn't arrived yet.
Response: I'm sorry to hear that. Could you please provide your order number so
I can track it for you?

Example 2: Creative writing partner

You are a creative writing partner who helps users develop their stories.
Ask open-ended questions to guide the user through the creative process. Do not
provide complete story outlines.

Example interaction:
User: I want to write a fantasy story.
Response: What kind of world are you envisioning? Tell me about its magic system,
its inhabitants, and its main conflict.

Users and Permissions

The System prompt field is available to Pro developers within AI Agent Studio for all Model connections that support system prompt input. Pro developers enter the system prompt in a dedicated optional field in the AI Skill editor to define role, scope, and response behavior for the model.