Skip to main content

Prompt Engineering

The Prompt Panel provides tools for crafting effective system prompts and context configurations for your agents.

Overview​

Prompts define agent behavior:

  • System Prompt: Core instructions and personality
  • Context Prompt: Dynamic context added at runtime
  • Model Settings: Temperature, max tokens, etc.

System Prompt​

The system prompt is the foundation of your agent's behavior:

const agent = await client.agents.update('agent-123', {
  systemPrompt: `You are a helpful IT support agent for Acme Corp.

Your responsibilities:
- Answer common IT questions
- Guide users through troubleshooting
- Create support tickets when needed

Guidelines:
- Be professional and patient
- Ask clarifying questions
- Escalate complex issues to human support

Available tools: memory, http_request, create_ticket`
});

Context Prompt​

Context prompts add dynamic information:

const agent = await client.agents.update('agent-123', {
  contextPrompt: `Current user information:
- Name: {{user_name}}
- Department: {{user_department}}
- Role: {{user_role}}

Recent tickets: {{recent_tickets}}`
});

Model Settings​

Configure the underlying model behavior:

const agent = await client.agents.update('agent-123', {
  model: 'claude-3-5-sonnet-20241022',
  temperature: 0.7,
  maxTokens: 4096
});

Prompt Structure Best Practices​

1. Clear Role Definition​

You are [role] for [organization].

Your primary purpose is [purpose].

2. Explicit Instructions​

Your responsibilities:
- [Task 1]
- [Task 2]
- [Task 3]

You should NOT:
- [Anti-pattern 1]
- [Anti-pattern 2]

3. Context Awareness​

You have access to:
- User's conversation history
- Company knowledge base
- Ticketing system

Use this information to provide personalized support.

4. Escalation Rules​

Escalate to human support when:
- User explicitly requests human
- Issue involves sensitive data
- Problem persists after 3 attempts

Transparency Principle​

All system prompts in MeetLoyd are fully transparent and editable. This is a core governance principle:

  • No black box: Users can always see how their agents are instructed
  • Full auditability: Every prompt change is tracked in version history
  • User control: Organization owners can view and modify any agent's prompt

This transparency ensures:

  • Compliance with AI regulations (EU AI Act, etc.)
  • Trust and accountability in AI decisions
  • Easy debugging and improvement
Historical Note

Previous versions supported a systemPromptProtected flag. This has been deprecated in favor of full transparency. All prompts are now visible to authorized users.

Testing Prompts​

Use the preview feature to test prompts:

  1. Open the Prompts tab
  2. Make changes to the prompt
  3. Click "Preview" to test
  4. Iterate based on results

Next: Learn about Schema Builder for configuration fields.