Prompting Best Practices for Skyone Studio

Clear communication is the foundation of AI assertiveness. At Skyone Studio, agent performance is directly linked to the quality of the prompts used.

This guide outlines best practices for creating clearer, more predictable, and efficient prompts, streamlining AI configuration and usage within the platform.

Basic prompt structure

Creating a high-quality prompt is an iterative process. Start with simple instructions and add details as the agent's behavior requires more precision.

1. Clarity and Specificity

Define:

• The specific task

• The context

• The expected tone and style

2. Avoid Ambiguity

The clearer the command, the lower the risk of inconsistent responses.

3. Determinism vs. Generalization

Detailed prompts produce more reliable responses than generic instructions.

• Prefer explicit, targeted commands.

• Avoid open-ended requests like "explain everything" or "answer freely."

• Clearly define the agent's use case to reduce out-of-context or overly generic answers.

4. Prompts Modularization

If the agent performs multiple functions, divide its responsibilities. Instead of a single, lengthy prompt, create smaller, specific ones. For example:

• One prompt for scheduling

• Another for data queries

• Another for technical support

This approach improves maintainability, predictability, and control over agent behavior.

5. Using Examples

Adding examples helps the model understand exactly the type of response expected. Whenever necessary, include:

• Sample questions

• Ideal response examples

This increases consistency and reduces unwanted variations.

VoxOne

Voice interactions require a different approach than text chats. While text relies on visual structure (lists, bolding, etc.), voice focuses on naturalness, flow, and objectivity.

Below are best practices for configuring voice agents in Skyone Studio's VoxOne.

1. Otimização para fala

The agent should sound like a person in a real conversation, not a technical manual reader.

Best Practices:

• Use short, direct sentences.

• Adopt a conversational tone similar to a phone service interaction.

• Avoid complex structures or long explanations without pauses.

Explicitly instruct the agent NOT to use:

• Bullet points

• Numbered lists

• Tables

• Emojis

These elements do not translate well to audio experiences.

2. Flow and Naturalness

To prevent the AI from sounding overly robotic, include humanization guidelines.

• Filler words: Allow moderate use of terms like "I see," "Right," or "Got it."

Use these sparingly to avoid increasing latency.

• Sensitive data pronunciation: Specify how the agent should read numbers or acronyms.

  • Example 1: "Read the zip code digit by digit (e.g., 0-1-0-4-0) instead of the full number (ten thousand and forty)."

  • Example 2: For dates, read them in full (e.g., "Two thousand twenty-six").

3. Prompting techniques for low latency

In voice interactions, silence can feel like a system failure. Reducing latency improves the user experience.

• Prioritize short responses: Prompts that require objective answers reduce model processing time, ensuring a smoother conversation.

• Step-by-step instructions: In long processes, ask the agent to confirm understanding before proceeding with a detailed explanation.

Prompt Examples

Avoid ❌	Prefer ✅
Help the user.	Explain how to integrate a REST API into Skyone Studio using simple and objective technical language.
Answer customer questions about the system.	Act as a Level 2 Technical Support Analyst. Use a polite, helpful, and professional tone. Avoid slang and use simplified technical language to ensure user understanding.
Check if there are errors in the data.	Analyze the execution log from the last hour. Identify only 'Time-out' failures. Return the response in a concise format, providing the Transaction ID and the error timestamp. Do not attempt to explain the root cause if debug logs are not available.

The model type selected in Skyone Studio directly influences:

• Context understanding capability

• Performance on complex prompts

• Response quality

Always consider the prompt's complexity level when choosing the most appropriate model.