Skip to main content

CAIPE Prompt Library

A prompt library is a curated collection of carefully designed prompts intended for use in multi-agent systems. These prompts guide AI agents—such as "Platform Engineer" or "Incident Engineer" personas—by providing standardized questions and instructions that facilitate effective collaboration, incident response, platform operations, and knowledge sharing.

  • Prompts are meta-level: They focus on coordination, decision-making, troubleshooting, and collaboration between agents.
  • Prompts are tested and validated: Ensuring they are effective for real-world use in roles like incident management and platform engineering.
  • A prompt library makes it easy for teams to reuse, share, and maintain high-quality instructions, leading to more reliable and efficient AI-driven workflows.

Available Prompts​

CAIPE (Community AI Platform Engineering) provides two main prompt configurations:

1. Basic CAIPE Prompt​

Configuration File: charts/ai-platform-engineering/data/prompt_config.yaml

Purpose: Simple agent routing and coordination for straightforward multi-agent operations.

Key Features:

  • Smart Routing & Coordination: Routes user requests to appropriate specialized agents (ArgoCD, AWS, Jira, GitHub, PagerDuty, Slack, Splunk, etc.)
  • Task Management: Two-phase approach for complex requests:
    • Phase 1: Planning - Creates a task plan before execution
    • Phase 2: Execution - Calls agents and tracks progress with checkmarks
  • Response Efficiency:
    • Preserves agent messages verbatim
    • Minimal wrapper around agent responses
    • Direct presentation of results
  • Simple Coordination: Focuses on routing and presenting results without complex orchestration

Use Cases:

  • Simple queries requiring single agent routing
  • Multi-step tasks that need basic coordination
  • Scenarios where straightforward agent delegation is sufficient

Example Behavior:

User: "Get the status of ArgoCD applications"
→ Routes directly to ArgoCD agent
→ Presents results cleanly

User: "Check cluster health and list open Jira tickets"
→ Creates task plan
→ Calls AWS and Jira agents
→ Presents combined results

2. CAIPE Deep Agent Prompt​

Configuration File: charts/ai-platform-engineering/data/prompt_config.deep_agent.yaml

Purpose: Advanced orchestration with comprehensive workflow management, parallel execution, and specialized incident engineering capabilities.

Key Features:

Core Orchestration Features​

  • TODO-Based Execution: Mandatory execution plan creation using write_todos tool for all operational requests
  • Parallel Execution: Maximizes parallelism by executing independent agents simultaneously
  • Agent Workspace: In-memory coordination system to prevent garbled output when multiple agents run in parallel
  • User Email Context: Smart handling of user email for first-person queries vs. third-person queries
  • Zero Hallucination: Never answers from knowledge base - always calls sub-agents first

Advanced Workflow Management​

  • Execution Modes: Declares PARALLEL, SEQUENTIAL, or HYBRID execution modes
  • Date Handling: Automatic date/time injection for relative date queries
  • Markdown Formatting: Built-in format_markdown tool for response validation
  • Error Recovery: Automatic retry logic when agents return errors with available options

User Input Handling​

  • UserInputMetaData Format: Structured JSON format for requesting user input
  • Sub-Agent Integration: Automatically formats sub-agent input requests into structured forms
  • Field Types: Supports text, textarea, number, select, and boolean input types

Incident Engineering Specialization​

Built-in support for four specialized incident management agents:

  1. Incident Investigator: Deep root cause analysis combining PagerDuty, Jira, Kubernetes/Komodor, RAG, and Confluence data
  2. Incident Documenter: Creates comprehensive post-incident reports and follow-up actions
  3. MTTR Analyst: Analyzes Mean Time To Recovery metrics and generates improvement reports
  4. Uptime Analyst: Analyzes service availability metrics and SLO compliance

Data Flow Management​

  • Explicit Data Extraction: Extracts and passes actual values between agents (not references)
  • Source Attribution: Preserves detailed information from sub-agents with provenance footers
  • Multi-Agent Correlation: Identifies relationships between data from different agents

Special Workflows​

  • OnCall Schedule & Task Analysis: Sequential workflow for PagerDuty → Jira correlation
  • Pod Investigation & Failure Analysis: Multi-agent workflow for Komodor → ArgoCD → AWS analysis
  • GitHub CI/CD Failure Analysis: Detailed CI check analysis with actionable recommendations
  • Jira Query & Data Formatting: Standardized table formatting with links and metadata

Use Cases:

  • Complex multi-agent workflows requiring parallel execution
  • Incident management and root cause analysis
  • Large-scale platform health reports
  • Operations requiring structured user input
  • Scenarios needing detailed execution planning and tracking

Example Behavior:

User: "Show me GitHub PRs and Jira tickets"
→ Creates TODO plan with PARALLEL mode
→ Calls GitHub AND Jira agents SIMULTANEOUSLY
→ Uses workspace to store results separately
→ Combines into unified table with attribution

User: "Investigate the API outage root cause"
→ Creates TODO plan
→ Routes to Incident Investigator specialist
→ Coordinates PagerDuty, Jira, Kubernetes, RAG agents
→ Synthesizes root cause analysis with evidence links

Configuration​

Both prompt configurations are located in:

  • Basic: charts/ai-platform-engineering/data/prompt_config.yaml
  • Deep Agent: charts/ai-platform-engineering/data/prompt_config.deep_agent.yaml

These YAML files contain:

  • System prompt templates
  • Agent-specific prompts (ArgoCD, AWS, Jira, GitHub, etc.)
  • Agent skill examples
  • Specialized workflow instructions

Choosing the Right Prompt​

Use Basic CAIPE Prompt when:

  • You need simple agent routing
  • Tasks are straightforward and don't require complex orchestration
  • You want minimal overhead and faster responses
  • Single-agent or simple multi-agent queries

Use CAIPE Deep Agent Prompt when:

  • You need advanced workflow management
  • Tasks require parallel execution of multiple agents
  • You need incident engineering capabilities
  • Complex multi-step operations with dependencies
  • Operations requiring structured user input
  • Large-scale platform health reports

Integration​

Both prompts integrate with the CAIPE multi-agent system and work with:

  • Specialized Agents: ArgoCD, AWS, Jira, GitHub, PagerDuty, Slack, Splunk, Komodor, Confluence, Webex, Weather, Backstage
  • RAG Knowledge Base: Documentation and process recall
  • Agent Workspace: In-memory coordination for parallel execution (Deep Agent only)
  • A2A Protocol: Agent-to-Agent communication

The prompts are used by the Platform Engineer orchestrator to route queries and coordinate agent execution based on the selected configuration.