Skip to main content

Module 10: Build Your Domain Agent

Duration: 120 minutes Day: Day 3, Session 4

Lab Location

The hands-on lab for this module lives in the Hermes repository. This is the primary build module of Day 3 — everything from Modules 7, 8, and 9 converges here into a working agent. Read the Concepts page first, choose your track, then open the Hermes lab guide.

What This Module Is About

This is where you stop learning about agents and start building one. Everything converges: your SKILL.md from Module 7, your tool configuration from Module 8, the design patterns from Module 9, and your impact assessment from Module 4.

The output is a working domain agent — deployed on Hermes, connected to real or simulated infrastructure, capable of executing your operational runbooks autonomously.

Choose Your Track

TrackDomainKey SkillsInfrastructure
Track A: DB Health and TuningRDS PostgreSQLSlow query analysis, parameter tuning, connection pool monitoringReal or simulated RDS (mock data provided)
Track B: Cost Anomaly and FinOpsAWS Cost ExplorerCost trend analysis, right-sizing, unused resource detectionReal or simulated Cost Explorer (mock data provided)
Track C: Kubernetes HealthKubernetesPod restart analysis, resource pressure, image pull issuesLocal KIND cluster (no cloud needed)

All three tracks produce the same type of artifact: a complete Hermes agent profile (SOUL.md + config.yaml + skills/) with simulated data for testing when real infrastructure is not available.

If you completed your Module 4 impact assessment — choose the track closest to your capstone candidate. You will extend this agent in Module 13 for your capstone.

Learning Objectives

By the end of this module, you will be able to:

  1. Build a complete domain agent from scratch by combining skills, tools, and configuration into a working Hermes profile
  2. Test against simulated infrastructure data using the provided mock datasets when real infrastructure is not available
  3. Evaluate agent output quality using a structured checklist: accuracy, completeness, confidence calibration, and actionability
  4. Identify what your agent cannot reliably do — the boundary conditions where it needs human judgment rather than autonomous action

Prerequisites

  • Modules 7 and 8 completed (you have a SKILL.md and understand tool configuration)
  • Module 9 completed (you understand design patterns)
  • Hermes installed and running
  • Track C only: KIND cluster running (kubectl cluster-info --context kind-lab)

Module Contents

SectionContentTime
ReadingConcepts: Agent Anatomy and Build Strategy20 min
ReadingReference: Agent Profile Structure and Output Evaluation10 min
LabBuild Your Domain Agent — Track A, B, or C (Hermes repo)80 min
QuizModule 10 Assessment10 min
ExploratoryStretch ProjectsOptional

What "Done" Looks Like

At the end of the lab, your agent can receive a task like "Investigate the connection pool saturation on our RDS instance" and:

  1. Load the relevant skill automatically based on the trigger
  2. Execute the diagnostic procedure against real or simulated data
  3. Reason about the evidence using the decision trees in the skill
  4. Generate a structured diagnosis with root cause hypothesis, supporting evidence, recommended actions, and escalation decision

That is a working domain agent. It will not be perfect — and the concepts reading covers what "good enough" means and how to improve from here.