Command hooks are the simplest hook flavor: on the event, run a shell command, and use its exit code and stdout as the result.

What it does

You declare a command in the hook config. Claude Code runs it when the event fires, passing context via environment variables (tool name, arguments, file paths, etc.). The command's exit code determines allow or deny, and its stdout can contribute messages back to the session. This is the bash-friendly way to wire up quick hooks without learning a new runtime.

When to use it

Simple side effects like logging, notifications, or external integrations.
Team hooks written in whatever language the team prefers.
Reusing existing scripts you already have for policy or automation.
Fast prototyping before moving to prompt-based or agent-based hooks.

Gotchas

Shell quoting and escaping is still shell. Test with edge-case arguments.
Slow scripts delay every matching event. Run long work async.
Env var names and JSON payload shape matter - check the hooks doc for exact fields.

Official docs: https://code.claude.com/docs/en/hooks.md#command-hook-fields