design-agent

Design Operational Agent

You are designing an operational agent based on the codebase analysis from the previous step.

Prerequisites

• The <component>-analysis.md file exists from the analyze step
• The component owner is available for Q&A

Process

Step 1: Load Analysis

Read the <component>-analysis.md file. Summarize the key findings to the owner:

"Based on the analysis, here's what I'm planning for your ops agent:

• Tools: [tool list]
• Key monitoring areas: [what the agent will watch]
• Domain knowledge: [what the agent will know about your component]"

Step 2: Determine Tool List

Always included:

• log_search — OpenSearch log querying
• metric_query — Prometheus metric querying

Add based on analysis:

• If Kafka dependencies found → suggest kafka_consumer tool (check lag, read messages)
• If database found → suggest db_query tool (health check, slow query analysis)
• If external APIs found → suggest api_health tool (endpoint health checks)

For each additional tool, ask:

"I'm suggesting a <tool_name> tool for . Should I include it? If yes, what endpoint/credentials does it need?"

Step 3: Define Config Values

Compile all config values needed for agent.yaml:

component: <from analysis>
description: <from analysis>

opensearch:
  endpoint: <ask if not in analysis>
  index_pattern: <from analysis>
  verify_certs: true

prometheus:
  endpoint: <ask if not in analysis>
  namespace: <from analysis>

Present to owner for confirmation.

Step 4: Draft System Prompt Content

Based on the analysis, prepare:

1. Component overview — what it does, in operational context
2. Domain knowledge — key error codes, business logic the agent should understand
3. Failure patterns — from analysis, formatted as agent instructions
4. Monitoring guidelines — what to check first, correlation patterns

Ask the owner:

"Here's the domain knowledge I'll give the agent: [draft content]

Is there anything to add or correct?"

Step 5: Present Design Summary

Present the complete design:

Agent Design for <component_name>
===================================

Tools:
  - log_search (OpenSearch)
  - metric_query (Prometheus)
  - <additional tools>

Config:
  <agent.yaml content>

System Prompt Sections:
  - Component: <summary>
  - Domain Knowledge: <summary>
  - Failure Patterns: <count> patterns documented
  - Guidelines: <summary>

"Does this design look good? If yes, I'll generate the agent project. Run: 03-generate-agent"

Output

Save the design as <component>-design.md with all the above details, structured so the generation skill can reference it directly.