Skip to content

Introduction

Welcome to the CI/CD for Agentic AI learning path. In this module, you will build a CI/CD pipeline that tests a Google ADK travel agent, captures Monocle traces on failure, and automatically creates GitHub issues assigned to AI coding agents for remediation.

What you will learn

  • How to instrument a Google ADK agent with Monocle
  • How to write agent tests using the MonocleValidator and Fluent API frameworks
  • How to build GitHub Actions workflows that capture traces on failure
  • How to auto-assign AI coding agents (Claude, Copilot) to investigate and fix failures
  • How to integrate Okahu Cloud evaluations into CI/CD
  • How to run the agent with either OpenAI or Google Gemini as the LLM provider

Scenario

You are building a CI/CD pipeline for a travel booking agent built with Google ADK (Agent Development Kit). When tests fail, you want the pipeline to automatically create a GitHub issue with the full Monocle trace embedded, assign it to an AI coding agent, and have the agent investigate and fix the issue.

The agent supports two LLM providers — set OPENAI_API_KEY to use OpenAI (defaults to gpt-3.5-turbo, overridable via OPENAI_MODEL), or set GOOGLE_API_KEY to use Google Gemini (defaults to gemini-2.5-flash-lite, overridable via GOOGLE_GENAI_MODEL). OpenAI takes priority when both keys are present.

Architecture overview

graph TD
    Push[Code Push / Manual Trigger] --> GHA[GitHub Actions]
    GHA --> Test[Run Agent Tests]
    Test -->|Pass| Done[Pipeline Succeeds]
    Test -->|Fail| Collect[Collect Monocle Traces]
    Collect --> Issue[Create GitHub Issue]
    Issue --> Assign[Assign AI Coding Agent]
    Assign --> Claude[Claude Code Agent]
    Assign --> Copilot[GitHub Copilot]
    Claude --> Fix[Investigate & Fix]
    Copilot --> Fix

Source code

okahu-demos/adk-travel-agent-with-cicd

Time to complete

~45 minutes