hook-development

You can create Claude Code hooks by configuring event-driven validation in hooks.json to intercept PreToolUse events, allowing you to validate tool inputs and block unsafe actions before they execute. This skill generates the correct plugin-specific structure and matchers for intercepting tool calls.

Prompt-based hooks use LLM evaluation for context-aware decisions like approving or denying actions at PreToolUse and Stop events, while command-based hooks run bash scripts for deterministic validation, schema checks, and security gating. This skill implements both types with correct JSON handling and exit behavior.

Configuring hooks.json requires structuring plugin-specific hook events with the correct matchers and supporting either prompt-based or command-based hook types. This skill generates the right configuration format for hooks.json and settings.json to handle input/output JSON and standardized exit behavior.

Yes, you can run bash scripts as command-based hooks for fast checks, schema validation, security path safety, and performance-critical gating. This skill develops command hooks that execute deterministic validation around tool execution and session lifecycle events.

You can enforce safe file writes by implementing PreToolUse hooks that intercept Write and Edit tool calls, validate paths and sensitive content, and block dangerous operations before they run. This skill designs hooks that apply validation rules to tool inputs at lifecycle events.

Use prompt-based hooks when you need LLM evaluation for context-aware decisions like requesting clarification at Stop events, and command-based hooks when you need deterministic bash validation for security and performance gating. This skill implements both approaches for safe event-driven control.