Claude Code You Can Trust: Verification, Skills and Hooks for Working Developers
Kroes, Fernando Silva
Synopsis "Claude Code You Can Trust: Verification, Skills and Hooks for Working Developers"
Your coding agent writes good code most of the time. You still read every line, because you cannot tell which time is the bad one. Claude Code You Can Trust is about the work that lets you stop. It is written for developers who already use Claude Code or another coding agent, by a developer who runs one every day. What you build A verify script and skill, so the agent checks the running app, not only the tests Hooks that block the few actions that are never right, and the tests that keep them working Mutation checks that find tests that cannot fail A critic subagent that judges finished work against a rubric it did not write An outer loop that turns bug reports into checked, queued findings without you A weekly sample of landed changes, so you review a few instead of everything Why this book is different Every example runs. Every code block is a file from the companion repository or a command with its output, and a script checks all of them before each edition. Real cases, with numbers. A deploy that printed success while the old build kept serving. A test that encoded the bug it should catch. A stop hook that let red tests through, found by an eval. Dated on purpose. Checked against Claude Code 2.1.289. An appendix lists every statement about Claude Code the book relies on, with its documentation page. For Claude Code 2.1.289 and later, and Node.js 22. The companion code was tested on macOS.