harness-books

Why we included this project

Once you put a coding agent inside a real repository, the question stops being what the model can write and becomes what keeps the whole system sane. These two books focus on that layer: how query loops are structured, where permission decisions happen, how context stays governed, and what happens when something fails mid-task. They also cover how verification gets split across multiple agents and the team-level rules that make an agent predictable beyond a single session. For engineers and tech leads who are starting to depend on Claude Code or Codex daily, this is a concrete way to reason about failure modes and design choices, grounded in how the tools actually behave rather than in generic best practices.