create-hooks

Event-driven hooks trigger custom actions in response to specific events like tool usage, user prompts, or session start. Configure hooks with JSON schemas, matcher patterns, and transformations to automate validation, logging, and command execution across Bash, Write/Edit, MCP, and other tool categories without manual intervention.

Yes. PreToolUse hooks validate commands before execution by matching patterns against tool inputs and returning permission decisions. Block dangerous commands like `git push --force` or `rm -rf` by defining matcher patterns and setting deny permissions in your hook configuration.

Claude Code supports PreToolUse, PostToolUse, UserPromptSubmit, Stop, and SessionStart hooks. Each type triggers at different workflow stages, allowing you to inject context, log activity, validate inputs, enforce policies, or automate responses to user prompts and tool execution events.

Create a PostToolUse hook targeting Bash tools that captures command inputs and writes them to a log file. Configure the hook with a matcher pattern, define output transformation logic, and specify the destination file path to maintain an audit trail of all executed commands.

Yes. Hooks support environment variables for dynamic configuration and system messages for injecting context or instructions. Use these features to customize hook behavior per project, pass sensitive values securely, and modify Claude's responses based on hook-triggered events.

Hooks use JSON-based input/output schemas defining the structure of hook payloads and responses. Schemas specify matcher patterns for tool selection, command and prompt hook types, transformation fields (updatedInput, permission decisions), and system message injection for safe, validated automation.