llm-books/sre-agent ? reverse-engineered prompt

Build me a local demo of an AI SRE agent that can watch a small fake ecommerce system, notice when something is broken, investigate it with logs, metrics, and traces, then suggest what went wrong and how to fix it. I want the whole thing to run on Docker Desktop on a laptop, with six services, a checkout style flow, a load generator, a chaos tool that can deliberately break things, and a dashboard setup so I can see incidents happen in real time.

The agent should keep a durable record of what it has already investigated, so if I run it again on the same incident it can replay the prior result instead of starting over. It should be able to work fully offline with a scripted planner, but also support a real model if I give it an API key. Make it feel like a book companion project I can run chapter by chapter, with simple make commands to start the environment, inject a failure, and let the agent handle it. If you need to check current docs for any tools, go ahead and look them up online.