Skip to main content

Prompt Configuration Feature

Status: 🟢 In-use Category: Configuration & Prompts Date: October 22, 2024

Overview​

The AI Platform Engineering Helm chart now supports flexible prompt configuration, allowing users to choose between predefined orchestration behaviors or provide custom configurations.

Feature Components​

1. Configuration Files​

The chart includes two predefined prompt configurations:

  • data/prompt_config.yaml - Default configuration

    • Standard multi-agent orchestrator
    • Balanced between flexibility and control
    • Maintains strict provenance and source attribution
    • Best for: Development, testing, general platform operations

  • data/prompt_config.deep_agent.yaml - Deep Agent configuration

    • Strict zero-hallucination mode
    • Enhanced provenance tracking
    • Enforces tool-only responses
    • Best for: Production, compliance, mission-critical systems

2. Helm Values Configuration​

Two new values control prompt behavior:

# Select predefined configuration
promptConfigType: "default"  # Options: "default" or "deep_agent"

# Or provide custom configuration (overrides promptConfigType)
promptConfig: ""

3. Template Logic​

The templates/prompt-config.yaml template implements a three-tier configuration hierarchy:

  1. Custom override (promptConfig) - Highest priority
  2. Deep Agent (promptConfigType: "deep_agent") - Second priority
  3. Default (promptConfigType: "default") - Fallback

Usage Examples​

Example 1: Deploy with Default Configuration​

helm install ai-platform charts/ai-platform-engineering/ \
  --set promptConfigType=default

Example 2: Deploy with Deep Agent Configuration​

helm install ai-platform charts/ai-platform-engineering/ \
  --set promptConfigType=deep_agent

Example 3: Deploy with Custom Configuration​

Create a custom values file:

# custom-prompt-values.yaml
promptConfig: |
  agent_name: "Custom Platform Agent"
  agent_description: |
    Specialized agent for custom workflows
  system_prompt_template: |
    Custom system prompt...
  agent_prompts:
    argocd:
      system_prompt: "Custom ArgoCD routing..."
    # ... additional configuration

Deploy:

helm install ai-platform charts/ai-platform-engineering/ \
  --values custom-prompt-values.yaml

Example 4: Switch Configuration on Upgrade​

# Upgrade from default to deep_agent
helm upgrade ai-platform charts/ai-platform-engineering/ \
  --set promptConfigType=deep_agent

Configuration Comparison​

FeatureDefaultDeep AgentCustom
Zero-hallucinationEnforcedStrictly enforcedUser-defined
Provenance trackingStandardEnhancedUser-defined
Tool response handlingVerbatim forwardingVerbatim + validationUser-defined
ComplexityLowLowHigh
CustomizabilityNoneNoneFull
Use caseGeneralMission-criticalSpecialized

Key Differences Between Default and Deep Agent​

Default Configuration​

  • Agent Name: "AI Platform Engineer"
  • Focus: Balanced orchestration with standard compliance
  • Behavioral Model: Standards-compliant orchestrator
  • Validation: ComplianceGuard and Aggregator meta-agents
  • Response Style: Professional, markdown-formatted with provenance

Deep Agent Configuration​

  • Agent Name: "AI Platform Engineer — Deep Agent"
  • Focus: Maximum adherence to zero-hallucination principles
  • Behavioral Model: Deep Agent Orchestrator with strict tool-only mode
  • Validation: Enhanced provenance validation and source verification
  • Response Style: Structured, highly traceable with explicit source attribution
  • Additional Features:
    • Explicit "Source-of-Truth Policy"
    • Stricter routing logic with RAG-first fallback
    • Enhanced tool-response handling rules
    • Explicit behavior model separation (tool-only vs RAG modes)

Implementation Details​

Template Hierarchy​

data:
  prompt_config.yaml: |
{{- if .Values.promptConfig }}
  # Use custom configuration (highest priority)
{{ .Values.promptConfig | nindent 4 }}
{{- else if eq .Values.promptConfigType "deep_agent" }}
  # Use deep agent configuration
{{ .Files.Get "data/prompt_config.deep_agent.yaml" | nindent 4 }}
{{- else }}
  # Use default configuration (fallback)
{{ .Files.Get "data/prompt_config.yaml" | nindent 4 }}
{{- end }}

Configuration Loading​

  1. Check if promptConfig is set → Use custom configuration
  2. Check if promptConfigType equals "deep_agent" → Load prompt_config.deep_agent.yaml
  3. Otherwise → Load default prompt_config.yaml

Testing​

All three configuration modes have been tested:

# Test default
helm template test charts/ai-platform-engineering/ \
  --set promptConfigType=default | grep agent_name

# Test deep_agent
helm template test charts/ai-platform-engineering/ \
  --set promptConfigType=deep_agent | grep agent_name

# Test custom
helm template test charts/ai-platform-engineering/ \
  --set-string 'promptConfig=agent_name: "Custom"' | grep agent_name

Backward Compatibility​

This feature is fully backward compatible:

  • Existing deployments without promptConfigType will use "default"
  • The original promptConfig override mechanism is preserved
  • No breaking changes to existing chart functionality

Best Practices​

Development and Testing​

promptConfigType: "default"

Production Deployments​

promptConfigType: "deep_agent"

Specialized Workflows​

promptConfig: |
  # Custom configuration
  ...

Documentation​

  • Chart README: /charts/ai-platform-engineering/README.md
  • Example Values: /charts/ai-platform-engineering/values-prompt-examples.yaml
  • Default Config: /charts/ai-platform-engineering/data/prompt_config.yaml
  • Deep Agent Config: /charts/ai-platform-engineering/data/prompt_config.deep_agent.yaml

Future Enhancements​

Potential future additions:

  • Additional predefined configurations for specific use cases
  • Configuration validation and schema enforcement
  • Dynamic configuration switching without pod restart
  • Configuration metrics and observability