Automated scripts or commands that fire at specific points in an AI coding agent's workflow — for example, before a shell command runs, after a file is edited, or when the agent tries to stop working. Hooks act as deterministic guardrails: unlike prompt-based instructions that an AI might ignore, hooks execute every time, enforcing rules like "run the test suite before committing" or "block any command that deletes files."

The concept borrows from a well-established pattern in software development (Git hooks, CI/CD pipelines, webhooks), but applied specifically to AI agents. Anthropic's Claude Code, Cursor, and other coding tools now support hooks that intercept the agent's actions at defined lifecycle events — before and after tool calls, when permissions are requested, or when the agent finishes a task. Each hook can inspect what the agent is about to do, allow it, modify it, or block it entirely.

For data reporters, hooks offer a practical safety net when using AI agents to write code. A reporter using an agent to build a web scraper could set up hooks that automatically run a linter after every file edit, block commands that might accidentally overwrite a dataset, or require tests to pass before the agent declares a task complete. Hooks turn vibe coding from a leap of faith into something closer to a supervised workflow — the agent moves fast, but hard rules keep it on track. They pair naturally with agent skills, which teach an agent how to do a task, while hooks enforce what it's not allowed to do.