create-hooks

Event-driven hooks automate Claude's behavior by executing shell commands or LLM prompts in response to specific events like tool usage, session lifecycle, or user interactions. Configure JSON-based hooks with matcher patterns and choose between command hooks for deterministic tasks or prompt hooks for context-aware decisions.

Yes, you can implement custom validation logic through hooks that intercept PreToolUse events. Create command or prompt hooks with matcher patterns to block or flag specific git commands before execution, enforcing safety standards without modifying core code.

Claude Code supports two hook types: command hooks for deterministic shell command execution and logging, and prompt hooks for complex decision-making using an LLM. Both apply across PreToolUse, PostToolUse, UserPromptSubmit, SessionStart, SessionEnd, and related lifecycle events.

Create a command hook triggered on PreToolUse events with a matcher pattern targeting bash commands. Configure it to execute a shell command that appends command details to a log file, automating audit trails without manual intervention.

Yes, use prompt hooks triggered on SessionStart events to inject project context automatically. Prompt hooks use LLM reasoning to understand your project structure and inject relevant information at session initialization, eliminating manual context setup.

Hook configuration includes JSON schema validation for input/output schemas, matcher pattern validation, and safety checks for command and prompt hook types. Built-in debugging and documentation workflows help identify configuration errors before hooks execute.