02: Multi-Prompt Packs

Build a pack containing multiple prompts for a complete application.

Learning Objectives

In this tutorial, you’ll learn to:

• Create multiple prompt configurations
• Organize prompts by feature
• Compile all prompts into a single pack
• Reference prompts within the pack

Time Required

20 minutes

Prerequisites

• Completed Tutorial 1: Your First Pack
• packc installed
• Basic understanding of prompt structure

Scenario

You’re building a customer service application that needs three types of prompts:

1. Greeting - Welcome customers
2. Support - Answer support questions
3. Escalation - Handle complex issues

Step 1: Project Setup

Create a new project:

mkdir customer-service-pack
cd customer-service-pack
mkdir -p prompts config packs

Step 2: Create Multiple Prompts

Greeting Prompt

cat > prompts/greeting.yaml <<'EOF'
apiVersion: promptkit.altairalabs.ai/v1alpha1
kind: PromptConfig
spec:
  task_type: greeting
  name: Customer Greeting
  description: Welcome customers and identify their needs

  system_prompt: |
    You are a friendly customer service representative.
    Greet the customer warmly and ask how you can help them today.

  user_template: |
    Customer:
    Previous interactions:

  template_engine: go

  parameters:
    temperature: 0.8
    max_tokens: 150
EOF

Support Prompt

cat > prompts/support.yaml <<'EOF'
apiVersion: promptkit.altairalabs.ai/v1alpha1
kind: PromptConfig
spec:
  task_type: support
  name: Technical Support
  description: Provide technical support and solutions

  system_prompt: |
    You are a knowledgeable technical support specialist.
    Help customers solve their technical issues with clear, step-by-step guidance.
    Always be patient and understanding.

  user_template: |
    Customer:
    Issue:
    Product:

  template_engine: go

  parameters:
    temperature: 0.7
    max_tokens: 500

  tools:
    - name: search_knowledge_base
      description: Search the knowledge base for solutions
    - name: create_ticket
      description: Create a support ticket
EOF

Escalation Prompt

cat > prompts/escalation.yaml <<'EOF'
apiVersion: promptkit.altairalabs.ai/v1alpha1
kind: PromptConfig
spec:
  task_type: escalation
  name: Escalation Handler
  description: Handle complex issues requiring escalation

  system_prompt: |
    You are a senior support specialist handling escalated issues.
    Document the issue thoroughly and prepare it for senior management review.
    Be empathetic and assure the customer their issue has high priority.

  user_template: |
    Customer:
    Account Type:
    Issue:
    Previous attempts:
    Urgency:

  template_engine: go

  parameters:
    temperature: 0.6
    max_tokens: 600

  tools:
    - name: create_escalation_ticket
      description: Create high-priority escalation ticket
    - name: notify_manager
      description: Notify support manager of escalation
EOF

Step 3: Create Arena Configuration

Create a configuration that includes all three prompts:

cat > config/arena.yaml <<'EOF'
prompts:
  - ../prompts/greeting.yaml
  - ../prompts/support.yaml
  - ../prompts/escalation.yaml

tools_directory: ./tools
EOF

Step 4: Compile Multi-Prompt Pack

Compile all prompts into a single pack:

packc compile \
  --config config/arena.yaml \
  --output packs/customer-service.pack.json \
  --id customer-service

Expected output:

Loaded 3 prompt configs from memory repository
Compiling 3 prompts into pack 'customer-service'...
✓ Pack compiled successfully: packs/customer-service.pack.json
  Contains 3 prompts: [greeting, support, escalation]

Step 5: Inspect the Pack

See all prompts in the pack:

packc inspect packs/customer-service.pack.json

Key sections to notice:

Prompts
=======
- ID: greeting
  Name: Customer Greeting
  ...

- ID: support
  Name: Technical Support
  Tools: [search_knowledge_base, create_ticket]
  ...

- ID: escalation
  Name: Escalation Handler
  Tools: [create_escalation_ticket, notify_manager]
  ...

Step 6: Validate the Pack

Ensure all prompts are valid:

packc validate packs/customer-service.pack.json

Understanding Multi-Prompt Packs

Pack Structure

Your pack now contains:

{
  "id": "customer-service",
  "version": "1.0.0",
  "prompts": {
    "greeting": { ... },
    "support": { ... },
    "escalation": { ... }
  }
}

Using Multiple Prompts

In your application, you can select prompts by task type:

// Load pack
pack, _ := manager.LoadPack("./packs/customer-service.pack.json")

// Use greeting prompt
conv, _ := manager.NewConversation(ctx, pack, config.ConversationConfig{
    TaskType: "greeting",
})

// Later, switch to support
supportConv, _ := manager.NewConversation(ctx, pack, config.ConversationConfig{
    TaskType: "support",
})

Organizing Large Packs

By Feature

prompts/
├── greeting/
│   ├── standard.yaml
│   └── vip.yaml
├── support/
│   ├── technical.yaml
│   └── billing.yaml
└── escalation/
    └── priority.yaml

By Environment

config/
├── arena.dev.yaml      # Dev prompts
├── arena.staging.yaml  # Staging prompts
└── arena.prod.yaml     # Production prompts

Try It Yourself

1. Add a Fourth Prompt

Create a follow-up prompt:

cat > prompts/followup.yaml <<'EOF'
apiVersion: promptkit.altairalabs.ai/v1alpha1
kind: PromptConfig
spec:
  task_type: followup
  name: Follow-up Check
  description: Check on customer satisfaction after support

  system_prompt: |
    You are conducting a satisfaction check after customer support.
    Ask if the issue was resolved and if they need any additional help.

  user_template: |
    Customer:
    Resolution:

  template_engine: go

  parameters:
    temperature: 0.8
    max_tokens: 200
EOF

Update arena.yaml:

prompts:
  - ../prompts/greeting.yaml
  - ../prompts/support.yaml
  - ../prompts/escalation.yaml
  - ../prompts/followup.yaml  # Add new prompt

Recompile:

packc compile --config config/arena.yaml --output packs/customer-service.pack.json --id customer-service

2. Create Environment-Specific Packs

# Development version (shorter responses)
cp config/arena.yaml config/arena.dev.yaml

# Production version (add more validation)
cp config/arena.yaml config/arena.prod.yaml

# Compile both
packc compile --config config/arena.dev.yaml --output packs/customer-service-dev.pack.json --id customer-service-dev
packc compile --config config/arena.prod.yaml --output packs/customer-service-prod.pack.json --id customer-service-prod

Benefits of Multi-Prompt Packs

1. Single Distribution

One pack file contains all related prompts:

# Instead of:
app.load("greeting.pack.json")
app.load("support.pack.json")
app.load("escalation.pack.json")

# You have:
app.load("customer-service.pack.json")

2. Consistent Versioning

All prompts share the same version:

packc compile --config arena.yaml --output packs/customer-service-v1.2.0.pack.json --id customer-service

3. Simplified Deployment

Deploy one file instead of many:

# Single deployment
cp packs/customer-service.pack.json /deployment/packs/

Best Practices

1. Related Prompts Together

Group prompts by application or feature:

✅ Good: customer-service.pack.json contains all customer service prompts
❌ Avoid: all-prompts.pack.json containing unrelated prompts

2. Consistent Naming

Use consistent task_type naming:

# Good
task_type: customer-greeting
task_type: customer-support
task_type: customer-escalation

# Avoid
task_type: greeting
task_type: support_thing
task_type: esc

3. Document Prompt Relationships

# arena.yaml with comments
prompts:
  - ../prompts/greeting.yaml      # Step 1: Initial contact
  - ../prompts/support.yaml       # Step 2: Problem solving
  - ../prompts/escalation.yaml    # Step 3: Complex issues
  - ../prompts/followup.yaml      # Step 4: Satisfaction check

Next Steps

Continue learning about pack management:

• Tutorial 3: Validation Workflows - Ensure pack quality
• How-To: Organize Packs - Structure your packs
• Reference: Pack Format - Understanding pack structure

Summary

You’ve learned to:

• ✅ Create multiple related prompts
• ✅ Compile them into a single pack
• ✅ Organize prompts by feature
• ✅ Understand multi-prompt pack structure
• ✅ Use prompts selectively in applications

Multi-prompt packs are ideal for:

• Complete applications
• Feature sets
• Workflow stages
• Related use cases

Great job! 🎉