Claude Code hooks, explained

Hooks are shell commands Claude Code runs automatically on lifecycle events. They turn conventions you keep repeating into deterministic automation — the machine enforces them, not a markdown reminder.

The events

PreToolUse

Runs before a tool call. Can block it — e.g. refuse a destructive git command or an edit outside the project.

PostToolUse

Runs after a tool succeeds. Great for auto-formatting a file Claude just wrote, or running the type checker.

Stop

Runs when Claude finishes a turn. Use it to run tests, or to display a summary.

Notification / UserPromptSubmit

React to prompts and notifications — log commands, inject context, or gate actions.

Common uses

Auto-format / lint the file Claude just edited (PostToolUse)
Block risky shell commands like rm -rf or force-push (PreToolUse)
Run the test suite when a turn ends (Stop)
Log every command for an audit trail (PreToolUse)

Where they live

Hooks are configured in your settings (.claude/settings.json). A hook can block an action, inject context, or just run a command. Because they are deterministic, they are the reliable way to enforce guardrails.

Ready-made hooks

Browse Claude Code hooks and other configs on ConfigDeck and install into .claude/ in one click.

Browse the marketplace →

FAQ

How does a hook block a tool call?

If the hook script exits with code 2, the tool call is refused before it runs.

Where are hooks configured?

In settings.json (project or ~/.claude/), under hooks.<EventName>.

What event types exist?

PreToolUse, PostToolUse, Stop, SessionStart, SubagentStop and UserPromptSubmit, among others.

Are hooks enough for safety on their own?

No — pair them with permissions (allow/deny) for defense in depth.