hook-development

Event-driven hooks automate actions triggered by Claude Code events like PreToolUse, PostToolUse, and SessionStart. Define hooks in hooks.json or settings.json with event type, matcher, command or prompt, and timeout. Use prompt-based hooks for complex decisions and command hooks for deterministic checks, then validate with validate-hook-schema.sh.

Claude Code supports PreToolUse, PostToolUse, Stop, SubagentStop, SessionStart, SessionEnd, and UserPromptSubmit events. Each event enables different automation points: PreToolUse for input validation, PostToolUse for output processing, SessionStart for context loading, and others for state management and cross-tool orchestration.

Validate hook configurations using validate-hook-schema.sh to check JSON structure, test-hook.sh to run hooks against sample inputs, and hook-linter.sh to lint for compliance. These tools catch schema errors, timeout issues, and policy violations before deployment.

Yes, hooks support environment persistence via CLAUDE_ENV_FILE. Store session state, context, and configuration data in the environment file, then reference it across PreToolUse, PostToolUse, and other events for stateful automation and cross-tool orchestration.

Prompt-based hooks output hookSpecificOutput for custom data, permissionDecision for access control, updatedInput to modify tool inputs, and systemMessage to inject instructions. JSON output schemas validate hook responses and ensure policy compliance across all hook types.

Yes, hook configurations support both wrapper format (hooks array nested under event) and direct format (flat event-to-hook mapping). Both formats use the same JSON schema with event, matcher, type, timeout, and command or prompt fields for flexibility across plugin architectures.