Failure Mode Review Agent

Most outages come from failures that someone could have listed beforehand. This agent lists the components in the system, such as databases, queues, third-party services and network links, from the architecture documents. For each it enumerates how it can fail: slow, down, wrong data, full or out of sync. It then checks whether handling exists, like retries, fallbacks or timeouts, and whether monitoring would raise an alert. It ranks the gaps by impact and likelihood. For the top gaps, it proposes a staging test, such as stopping a service or adding delay, and runs it only after approval. After each test it checks the result against what was expected and updates the ranking. The architect approves the final plan. Edge case: a retry setting turns a small slowdown into a flood.