What are test hooks in AI-native development?

Test hooks connect test and lint commands to lifecycle events in AI coding agents like Claude Code and Cursor. When an event fires (e.g., after a file edit or at session end), the registered command runs automatically and blocks the agent's action if it fails. This creates deterministic local validation that doesn't rely on prompts or agent memory. The post explains the three components of a test hook (event, command, blocking behavior), best practices for layering checks across the agent lifecycle, how hooks complement CI/CD rather than replace it, and how CircleCI's open-source Chunk CLI scaffolds the full setup including an AI code review layer driven by your team's actual PR history.