claude-code-hooks

Event-driven automation in Claude Code uses bash or TypeScript scripts that execute at nine key events—PreToolUse, PostToolUse, Notification, UserPromptSubmit, Stop, SubagentStop, PreCompact, SessionStart, and SessionEnd. Scripts receive stdin JSON input and output JSON with fields like continue, stopReason, or suppressOutput to control Claude's behavior.

Yes. Implement a PreToolUse hook to inspect and block commands before execution, or use exit code 2 to signal blocking errors. This enables security checks, coding standard enforcement, and custom validation without manual intervention.

Use a PostToolUse hook to detect file-write events and apply formatting. The hook receives tool execution details via stdin and can trigger style checks or reformatting based on your project's guidelines before returning control to Claude.

Hooks receive stdin as JSON containing event context and details. They output JSON with hook-specific fields (continue, stopReason, suppressOutput, decision, hookSpecificOutput). Exit code 0 signals success; code 2 indicates blocking errors.

The claude-hooks-sdk is recommended for TypeScript integration to streamline JSON parsing and event handling, but you can write hooks directly as bash or TypeScript scripts that follow the stdin/stdout JSON contract.

SessionStart runs once at Claude Code startup for initialization or context injection; UserPromptSubmit runs before Claude processes each user input. Choose SessionStart for one-time setup and UserPromptSubmit for per-prompt validation or formatting.