What hooks are

A hook is a shell command the harness runs automatically when something happens in a session. You do not call it - Claude Code fires it for you on a lifecycle event. Hooks are how you bolt your own automation onto the agent: format a file the moment it is edited, block a risky command before it runs, or log every prompt.

The lifecycle events

Each hook is tied to an event. The common ones:

There are more, but these cover most needs.

Where hooks are configured

Hooks live in settings.json under a hooks key. Each event maps to a list of matchers, and each matcher carries the command to run. A PostToolUse hook that runs after any Edit looks like this:

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Edit",
        "hooks": [
          { "type": "command", "command": "prettier --write $CLAUDE_FILE_PATHS" }
        ]
      }
    ]
  }
}

Read it inside out: the event is PostToolUse, the matcher picks which tool triggers it (here Edit), and each entry under the inner hooks array is a command with "type": "command" and the shell command to run.

Typical uses

• Auto-format or lint a file right after Claude edits it (PostToolUse).
• Block a dangerous command before it runs (PreToolUse).
• Append every submitted prompt to an audit log (UserPromptSubmit).

Because a hook is just a shell command, it can do anything your shell can. Next, add a PostToolUse hook of your own to a project's settings.json.

Your lab setup

This VM comes with Claude Code preinstalled and signed in - just run claude to start.