Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.sigmamind.ai/llms.txt

Use this file to discover all available pages before exploring further.

​
Introduction

Prompt engineering is the backbone of every effective AI voice agent. The way you design your prompt directly shapes how the agent:
  • Understands user intent
  • Responds in real-time conversations
  • Handles ambiguity and edge cases
  • Maintains a consistent personality
Unlike traditional programming, you’re not writing strict logic—you’re guiding behavior. Clear prompts lead to predictable, human-like conversations. Poor prompts lead to inconsistent and confusing responses. This guide focuses on practical strategies to improve reliability, clarity, and conversational quality.

​
Where to Apply

These practices can be used across different agent types:
  • Single Prompt Agents → Apply directly in your main prompt
  • Conversation Flow Agents → Apply within each node’s instructions

​
Getting Started

The fastest way to understand effective prompts is to study working examples.
  • Create a new agent in your dashboard
  • Explore pre-built templates
  • Observe how prompts are structured and constrained
Well-designed templates already include tone control, flow logic, and fallback handling.

​
Best Practice 1: Use Structured Prompts

Organize prompts into clear sections instead of long paragraphs.
Structured prompts are easier to control, update, and scale. 
## Identity
Define who the agent is and its purpose

## Style
Set tone, length, and conversational behavior

## Rules
Define constraints and response patterns

## Task
Outline step-by-step responsibilities

## Objection Handling
Handle confusion, missing info, or frustration
Example Prompt: 
## Identity
You are a helpful AI assistant that provides clear, practical guidance on business and productivity topics.

## Style
Keep responses concise, friendly, and easy to understand. Use simple language and short paragraphs.

## Rules
Do not provide false information. Ask clarifying questions if needed. Avoid unnecessary details.

## Task

Understand the user’s question
Break down the solution into clear steps
Provide actionable advice
Summarize key points briefly

## Objection Handling
If the user is unsure or disagrees, acknowledge their concern, clarify misunderstandings, and offer an alternative solution.

​
Best Practice 2: Design for Voice Conversations

Voice interactions require clarity and brevity. Guidelines:
  • Keep responses short and natural
  • Ask one question at a time
  • Acknowledge user input before responding
  • Avoid sounding scripted—opt for simple, human phrasing.

​
Best Practice 3: Be Explicit and Controlled

The agent should not guess behavior. Clearly define:
  • What actions to take
  • When to take them
  • What to avoid
Example:
  • If information is missing → ask a follow-up
  • If request is unclear → ask for clarification
  • Clarity directly improves consistency.

​
Best Practice 4: Define Actions and Tool Usage Clearly

If your agent interacts with tools or APIs, define exact triggers.

​
Example

## Tool Usage

- If user asks for order status:
  → Call `check_order_status`

- If user requests a refund:
  → Call `escalate_to_support`

- After any tool call:
  → Summarize the result
  → Ask for next step
Example Prompt 
## Identity
You are a scheduling assistant that helps users book and manage appointments.

## Style
Keep responses clear, polite, and concise.

## Rules
Do not assume availability. Always use the scheduling tool for bookings. Confirm details before finalizing.

## Task

Understand booking request
Collect required details (date, time, purpose)
Use the scheduling tool
Confirm booking with the user

## Tool Usage

If user wants to book a meeting:
→ Call book_appointment
If user wants to reschedule/cancel:
→ Call modify_appointment
After any tool call:
→ Summarize booking details
→ Ask if anything else is needed

## Objection Handling
If the requested time is unavailable, suggest alternative slots and ask the user to choose one.

​
Key Rules

  • Use clear trigger conditions
  • Define execution order when needed
  • Avoid unnecessary tool calls

​
When to Use Structured Flows

Move beyond a single prompt when:
  • Logic involves multiple decision paths
  • Several tools must be coordinated
  • Conversation state needs tracking
Breaking tasks into smaller steps improves reliability and control.

​
Summary

Well-designed prompts are:
  • Structured
  • Clear
  • Controlled
  • Conversation-aware
Strong prompts lead to predictable behavior and better user experiences.