Claude Code Hooks System

You can automate custom scripts during the Claude Code lifecycle by defining shell commands, HTTP requests, or LLM prompts in settings.json. These hooks execute at specific events like UserPromptSubmit or PreToolUse to control agent behavior.

Yes, you can block specific tool calls like rm -rf by configuring a PreToolUse hook. The hook intercepts the tool execution, validates the structured JSON input, and blocks dangerous commands before the Bash tool runs them.

To trigger desktop notifications when Claude sends a message, configure a Notification hook. This lifecycle event executes your custom shell command or script automatically whenever Claude outputs a message to the user.

Yes, the Claude Code hooks system supports synchronous execution and structured JSON input/output. This allows you to manage control flow, validate tool calls, and automate responses reliably at specific lifecycle events.

Available lifecycle event types for hook customization include UserPromptSubmit, PreToolUse, and Notification. These events allow you to intercept and modify Claude's behavior at various points by executing custom logic via shell commands or HTTP requests.

To approve all tool calls by default, configure a PreToolUse hook in your settings.json file. This setup automatically validates and approves tool execution requests, streamlining the agent's workflow without manual intervention.