Creating Hooks

Creating Hooks

Build event-driven automation for Claude Code using hooks - scripts that execute at specific workflow points.

Quick Reference

Configuration Locations

Hooks are configured in settings files (in order of precedence):

Hook Structure

{
  "hooks": {
    "EventName": [
      {
        "matcher": "ToolPattern",
        "hooks": [
          {
            "type": "command",
            "command": "your-command-here",
            "timeout": 30
          }
        ]
      }
    ]
  }
}

Matcher Syntax

Note: Matchers are case-sensitive and only apply to PreToolUse, PostToolUse, and PermissionRequest.

Hook Types

Hook Options

Note: As of 2.1.3, the maximum hook timeout was increased from 60 seconds to 10 minutes (600s).

Exit Codes

Exit Code 2 Behavior by Event

Environment Variables

Hook Input (stdin)

All hooks receive JSON via stdin with common fields:

{
  "session_id": "abc123",
  "transcript_path": "/path/to/transcript.jsonl",
  "cwd": "/current/directory",
  "permission_mode": "default",
  "hook_event_name": "EventName"
}

Permission modes: default, plan, acceptEdits, dontAsk, bypassPermissions

Decision Guide: Which Hook Do I Need?

Before Tool Execution

Use PreToolUse to:

• Validate tool inputs before execution
• Auto-approve safe operations (e.g., reading docs)
• Block dangerous commands
• Modify tool inputs

After Tool Execution

Use PostToolUse to:

• Auto-format code after Write/Edit
• Run linters after file changes
• Log file modifications
• Provide feedback to Claude

Permission Automation

Use PermissionRequest to:

• Auto-allow trusted operations
• Auto-deny blocked patterns
• Enforce security policies

Prompt Processing

Use UserPromptSubmit to:

• Inject context (current time, git status)
• Validate prompts for secrets
• Block sensitive requests

Session Lifecycle

Use SessionStart to:

• Load development context
• Set environment variables
• Install dependencies

Use SessionEnd to:

• Clean up resources
• Log session statistics

Agent Completion

Use Stop / SubagentStop to:

• Verify task completion
• Force Claude to continue working
• Add completion checks

Context Management

Use PreCompact to:

• Customize compaction behavior
• Add pre-compaction context

Alerts

Use Notification to:

• Custom notification routing
• Third-party integrations (Slack, Discord)

Workflow: Creating a Hook

Prerequisites

• Identify which event to hook into
• Decide: command (bash) or prompt (LLM) type
• Plan exit code behavior

Steps

1. Create hook script

   • Write executable script (bash, python, etc.)
   • Read JSON from stdin
   • Output JSON to stdout (if needed)
   • Use appropriate exit code

2. Configure in settings

   • Add to appropriate settings file
   • Set matcher pattern (if applicable)
   • Set timeout if needed (default: 60s)

3. Test

   • Run claude --debug to see hook execution
   • Check /hooks menu for registration
   • Verify exit codes work as expected

Validation

• Script is executable (chmod +x)
• JSON input/output is valid
• Exit codes are correct
• Matcher pattern works

Tool-Specific Hooks

Common patterns for hooks targeting specific tools.

Bash Tool Hooks

Validate commands before execution, log sensitive operations, or block dangerous commands.

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          {
            "type": "command",
            "command": "\"$CLAUDE_PROJECT_DIR\"/.claude/hooks/validate-bash.sh"
          }
        ]
      }
    ]
  }
}

Common validations:

• Block rm -rf / patterns
• Require approval for sudo commands
• Log all commands to audit file
• Block network commands in certain contexts

Write Tool Hooks

Validate file paths, enforce naming conventions, or auto-format after write.

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Write",
        "hooks": [
          {
            "type": "command",
            "command": "\"$CLAUDE_PROJECT_DIR\"/.claude/hooks/after-write.sh"
          }
        ]
      }
    ]
  }
}

Common patterns:

• Auto-format with Prettier/Black
• Validate file encoding (UTF-8)
• Check for accidental credential writes
• Run type-checking after TypeScript writes

Edit Tool Hooks

Validate edits, prevent changes to critical files, or run linting after edits.

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Edit",
        "hooks": [
          {
            "type": "command",
            "command": "\"$CLAUDE_PROJECT_DIR\"/.claude/hooks/validate-edit.sh"
          }
        ]
      }
    ]
  }
}

Common patterns:

• Block edits to lock files (package-lock.json)
• Prevent edits to generated files
• Run linter after file edits
• Validate imports/exports after module changes

Read Tool Hooks

Log file access, validate read permissions, or inject context based on files read.

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Read",
        "hooks": [
          {
            "type": "command",
            "command": "\"$CLAUDE_PROJECT_DIR\"/.claude/hooks/validate-read.sh"
          }
        ]
      }
    ]
  }
}

Common patterns:

• Block reading sensitive files (.env, credentials)
• Log file access for auditing
• Auto-approve reading documentation
• Inject related context when reading specific files

Common Patterns

Auto-Format on File Write

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Write|Edit",
        "hooks": [
          {
            "type": "command",
            "command": "\"$CLAUDE_PROJECT_DIR\"/.claude/hooks/format.sh"
          }
        ]
      }
    ]
  }
}

Inject Context on Session Start

{
  "hooks": {
    "SessionStart": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "echo \"Git branch: $(git branch --show-current)\""
          }
        ]
      }
    ]
  }
}

Auto-Approve Documentation Reads

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Read",
        "hooks": [
          {
            "type": "command",
            "command": "\"$CLAUDE_PROJECT_DIR\"/.claude/hooks/approve-docs.py"
          }
        ]
      }
    ]
  }
}

Hook Framework (YAML Configuration)

For projects with multiple hooks, the Hook Framework provides a YAML-based configuration with built-in handlers and environment variable injection.

Installation

bun add claude-code-sdk

YAML Configuration

Create hooks.yaml in your project root:

version: 1

settings:
  debug: false
  parallelExecution: true
  defaultTimeoutMs: 30000

builtins:
  # Human-friendly session names (e.g., "brave-elephant")
  session-naming:
    enabled: true
    options:
      format: adjective-animal

  # Track turns between Stop events
  turn-tracker:
    enabled: true

  # Block dangerous Bash commands
  dangerous-command-guard:
    enabled: true
    options:
      blockedPatterns:
        - "rm -rf /"
        - "rm -rf ~"

  # Inject session context
  context-injection:
    enabled: true
    options:
      template: "Session: ${sessionName} | Turn: ${turnId}"

  # Log tool usage
  tool-logger:
    enabled: true
    options:
      outputPath: ~/.claude/logs/tools.log

handlers:
  # Custom command handlers
  my-validator:
    events: [PreToolUse]
    matcher: "Bash"
    command: ./scripts/validate-command.sh
    timeoutMs: 5000

Built-in Handlers

Environment Variables for Custom Handlers

Custom command handlers receive these environment variables:

TypeScript Framework

import { createFramework, handler, blockResult } from 'claude-code-sdk/hooks/framework';

const framework = createFramework({ debug: true });

// Block dangerous commands
framework.onPreToolUse(
  handler()
    .id('danger-guard')
    .forTools('Bash')
    .handle(ctx => {
      const input = ctx.event.tool_input as { command?: string };
      if (input.command?.includes('rm -rf /')) {
        return blockResult('Dangerous command blocked');
      }
      return { success: true };
    })
);

// Access turn/session context
framework.onPostToolUse(
  handler()
    .id('context-logger')
    .handle(ctx => {
      const turnId = ctx.results.get('turn-tracker')?.data?.turnId;
      const sessionName = ctx.results.get('session-naming')?.data?.sessionName;
      console.error(`[${sessionName}] Turn ${turnId}: ${ctx.event.tool_name}`);
      return { success: true };
    })
);

await framework.run();

Using with settings.json

Point your settings.json to the framework entry point:

{
  "hooks": {
    "PreToolUse": [{ "matcher": "*", "hooks": [{ "type": "command", "command": "bun run hooks-framework" }] }],
    "PostToolUse": [{ "matcher": "*", "hooks": [{ "type": "command", "command": "bun run hooks-framework" }] }],
    "SessionStart": [{ "matcher": "*", "hooks": [{ "type": "command", "command": "bun run hooks-framework" }] }],
    "Stop": [{ "matcher": "*", "hooks": [{ "type": "command", "command": "bun run hooks-framework" }] }]
  }
}

Debugging

Run with debug mode:

claude --debug

Security Considerations

• Validate and sanitize all inputs
• Quote shell variables ("$VAR" not $VAR)
• Check for path traversal (..)
• Use absolute paths for scripts
• Skip sensitive files (.env, keys)

Frontmatter Hooks

Hooks can also be defined directly in YAML frontmatter of Skills, Agents, and Slash Commands. These hooks are:

• Lifecycle-scoped - Only active while the component executes
• Auto-cleanup - Removed when the component finishes
• Portable - Packaged with the component for distribution

Supported events: PreToolUse, PostToolUse, Stop

Quick Example (in a Skill)

---
name: my-skill
description: A skill with lifecycle hooks
hooks:
  PreToolUse:
    - matcher: "Bash"
      hooks:
        - type: command
          command: "./validate.sh"
          once: true
---

Key Differences from Settings Hooks

See FRONTMATTER-HOOKS.md for complete documentation.

Reference Files