Agent-Specific Prompting

Agent-specific prompts define how individual agents should behave, what knowledge they have access to, and how they should respond to user requests within their domain of expertise.

What is Agent-Specific Prompting?

Agent-specific prompts are the instructions that determine an individual agent's:

  • Identity and personality
  • Knowledge and expertise
  • Response patterns and behaviors
  • Conversation style and tone
  • Specialized procedures or workflows

Why Use Multiple Agents?

Multiple agents allow you to:

  • Isolate different functionalities
  • Maintain cleaner, more focused prompts
  • Specialize agents for different types of interactions
  • Provide different tools and capabilities to different agents
  • Create more manageable and maintainable systems

How to Create and Edit Agent Prompts

  1. Navigate to your team's agents section
  2. Select the agent you want to edit (or create a new one)
  3. Locate the "Prompt" or "System Prompt" field
  4. Structure your prompt following the three-part format (description, notes, actions)
  5. Use XML tags to organize different sections of your prompt
  6. Save your changes

How to Structure an Effective Agent Prompt

The best agent prompts typically follow a three-part structure:

  1. Description Section - Explains what the agent is and its purpose
  2. Notes/Guidelines Section - Provides additional context or constraints
  3. Action/Response Section - Details specific behaviors, questions, or responses

Example agent prompt structure:

<description>
You are Coach Ed Baxter, a skilled conversation coach specializing in helping users improve their communication skills in professional settings.
</description>

<notes>
- Keep responses concise and actionable
- Use examples from real-world scenarios when possible
- Avoid technical jargon unless specifically requested
</notes>

<actions>
1. Begin by understanding the user's specific communication challenge
2. Offer practical advice with concrete examples
3. Suggest specific exercises or techniques to improve
4. Check if the user has questions about your recommendations
</actions>

How to Include Knowledge in Agent Prompts

  1. Identify the essential information your agent needs
  2. Add this information in a dedicated section of your prompt
  3. For larger bodies of knowledge, consider using the knowledge base feature instead
  4. Format complex information (like conversation samples) clearly with appropriate spacing
  5. Use XML tags to separate different types of knowledge

Example of adding knowledge to a prompt:

<conversation_samples>
User: I struggle with public speaking anxiety.
Agent: I understand how challenging public speaking anxiety can be. Many professionals experience this. Let's start with a simple technique: the 4-7-8 breathing method. Inhale for 4 seconds, hold for 7 seconds, and exhale for 8 seconds. Try this before your next presentation. Would you like more techniques specific to your situation?

User: Yes, please. I especially freeze when faced with unexpected questions.
Agent: For handling unexpected questions, I recommend the "bridge" technique. Acknowledge the question, bridge to what you know, and answer confidently. For example: "That's an interesting perspective. It relates to what we found in our research..." Would you like to practice this with a scenario?
</conversation_samples>

Specialized Agent Types

Different use cases may require different types of specialized agents:

Conversational Agents

These handle general chat interactions, focusing on natural dialogue and information provision.

Deep Dive Agents

These guide users through structured exploration of specific topics, often following predefined question sequences.

Technical Support Agents

These help troubleshoot issues through diagnostic questions and step-by-step solutions.

Sales or Lead Qualification Agents

These qualify prospects and guide them through initial stages of a sales process.

Best Practices for Agent Prompts

  1. Follow the three-part structure: description, notes, actions
  2. Focus each agent on a specific domain or function
  3. Include examples of ideal responses when possible
  4. Be specific about the tone, style, and approach you want
  5. Use XML tags to organize different response types
  6. Test with various inputs to ensure consistent performance

Next Steps

Once you've created your agent prompts, learn how to effectively use XML tags in prompting to better structure your prompts.