Claude Code Hooks: Automate the Checks You Run by Hand

A hook is a command that Claude Code runs automatically when a specific thing happens - a tool is about to run, an edit just landed, a session starts. Instead of you remembering to format, test, and check for dangerous commands, you write the rule once and it fires every time. Hooks live in your settings and run locally.

• PreToolUse - runs before a tool executes. Use it to block dangerous commands before they can touch anything (return a non-zero exit to stop the action).
• PostToolUse - runs after a tool succeeds. The classic use is auto-formatting every file the moment it is edited, so your code is always clean.
• UserPromptSubmit - runs when you send a message. Use it to inject context automatically - the current branch, an open ticket, project rules.
• Stop - runs when Claude finishes responding. A good place to run your test suite so you know immediately if something broke.
• SessionStart - runs when a session opens. Use it to brief Claude on the state of things - what is in progress, what changed since last time.

Add hooks to your settings file. The shape below shows a formatter on every edit and a guard before Bash runs. Exact field names can vary by version, so run /help or check the current docs to confirm the schema for your install - then adapt this.

json{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Edit|Write",
        "hooks": [
          { "type": "command", "command": "prettier --write $CLAUDE_FILE_PATHS" }
        ]
      }
    ],
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          { "type": "command", "command": "./scripts/guard-bash.sh" }
        ]
      }
    ]
  }
}

The guard script reads the command about to run and exits non-zero if it matches something you never want to run unattended. That single rule turns 'please be careful' into an actual wall.

1. 1Start with PostToolUse formatting - it is safe, instantly useful, and shows you how hooks fire.
2. 2Add a PreToolUse guard for the commands you never want run without you looking.
3. 3Add a Stop hook that runs your tests once you have a test suite worth running.
4. 4Only then reach for UserPromptSubmit and SessionStart - they are powerful but easier to get wrong.