Your agents passed every test you wrote for them. That means the tests you wrote were the problem.

Red-teaming is the discipline of finding failure modes before production does - by thinking like an attacker rather than a builder, constructing inputs designed to break rather than pass, and building a systematic record of how agents fail rather than assuming they won't. This book teaches that discipline to Claude Code operators running real multi-agent systems.

Seven chapters, seven artifacts. Chapter 1 establishes the red-team mindset: the shift from building to breaking, and why the person who built the agent is usually the worst person to test it. Chapter 2 builds a failure taxonomy - a working classification system for agent failure modes that makes testing systematic rather than random. Chapter 3 constructs adversarial inputs: parameterized prompts designed to surface specific failure categories. Chapter 4 runs boundary tests: repeatable probes at the edges of what agents can do, where degradation appears before outright failure. Chapter 5 measures instruction drift - whether an agent stays on task across an extended interaction, and how to detect the moment it stops. Chapter 6 tests prompt injection: what happens when hostile content enters the pipeline through data the agent processes, and how to probe resistance without assuming it exists. Chapter 7 automates the entire suite into a red-team pipeline that runs without manual execution and logs results for comparison across sessions.

By the last chapter you have a red-team harness that finds failure modes on demand. Production surprises come from failure modes nobody looked for. This book gives you the method for looking.