Hooks in Claude Code

Hooks let you automatically run your own commands at specific points in the Claude Code workflow – e.g. before every tool call.

What a hook is

A hook is a shell command you define, which Claude Code automatically runs at a specific point in the workflow – not optionally decided by the model, but deterministically triggered by the system. This lets you enforce rules that shouldn't depend on a model's "discretion."

Important hook events

PreToolUse: runs before a tool is executed – can even block the execution
PostToolUse: runs after a tool has been executed
UserPromptSubmit: runs when you submit a message
SessionStart / Stop: runs at the start or end of a session

What hooks are used for

Guardrails: automatically block dangerous commands (e.g. rm -rf) before they run
Automatic formatting or linting after every file change
Logging and notifications for certain actions
Custom validation before an agent is allowed to continue

Configuration

Hooks are configured in Claude Code's settings.json – per project or globally – making them reproducible independent of the current prompt, because they run outside the model.

EXAMPLE

A PreToolUse hook checks every Bash command for the pattern 'rm -rf' and aborts execution with an error message before the command even runs.

QUICK QUIZ

What sets a hook apart from a normal instruction in the prompt?