Hooks in Claude Code

What Are Hooks?

• Deterministic execution: Hooks always run, unlike prompt instructions which may not
• Run commands at specific points in Claude's lifecycle

Hook Events

• User prompt submit: Runs when you submit a prompt (before Claude processes it)
• Pre-tool use: Runs before a tool call executes
• Post-tool use: Runs after a tool call completes
• Notification: Runs when Claude sends a notification
• Stop: Runs when Claude finishes responding

Common Use Cases

• Auto-formatting files after edits
• Logging executed commands for compliance
• Blocking dangerous operations (e. g. , modifying production files)
• Sending notifications when tasks complete

Configuration

• Configured in settings. json file
• Specify: event type, optional matcher (which tools apply), and command to run
• Project-level hooks can be checked into your repo for team-wide consistency

Pre-Tool Use Hooks (Blocking Operations)

• Receive tool name and input as JSON on stdin
• Exit code 0 = proceed
• Exit code 2 = block the action
• Error message sent to Claude as feedback explaining why it was blocked

Best Practices

• Use post-tool use for auto-formatting and logging
• Use pre-tool use to enforce hard rules and block dangerous operations
• Use CLAUDE_PROJECT_DIR environment variable in commands to reference project scripts
• If something must happen every time without fail, use a hook instead of a prompt

Takeaways

• Hooks guarantee execution at specific lifecycle points, unlike prompt instructions which may be skipped—use them when something must always happen

• Pre-tool use hooks can block operations by returning exit code 2, making them ideal for enforcing safety rules (e. g. , preventing production file modifications)

• Post-tool use hooks are for automation: auto-formatting, logging, and notifications that should run after tool execution completes

• **Configure hooks in settings. json** with event type, optional tool matcher, and command; project-level hooks can be version-controlled for team consistency

• Use exit codes strategically: exit 0 to proceed, exit 2 to block, with error messages sent to Claude explaining the rejection