mcpbeat

Creating Claude Hooks

agentworkforce/creating-claude-hooks

Use when creating or publishing Claude Code hooks - covers executable format, event types, JSON I/O, exit codes, security requirements, and PRPM package structure

3k tokens
context cost
the whole folder, loaded on every use
1
files
instructions only
0
copies elsewhere
how many repositories repackaged it
784
stars on the repo
on the repository, not the skill itself

Install

one command, takes just this skill from the repository
npx skills add https://github.com/AgentWorkforce/relay --skill creating-claude-hooks

The instruction itself

36 sections, as written by the author

Creating Claude Code Hooks

Use this skill when creating, improving, or publishing Claude Code hooks. Provides essential guidance on hook format, event handling, I/O conventions, and package structure.

When to Use This Skill

Activate this skill when:

  • User asks to create a new Claude Code hook
  • User wants to publish a hook as a PRPM package
  • User needs to understand hook format or events
  • User is troubleshooting hook execution
  • User asks about hook vs skill vs command differences

Quick Reference

Hook File Format

| Aspect | Requirement |

| --------------- | ------------------------------------------------- |

| Location | .claude/hooks/<event-name> |

| Format | Executable file (shell, TypeScript, Python, etc.) |

| Permissions | Must be executable (chmod +x) |

| Shebang | Required (#!/bin/bash or #!/usr/bin/env node) |

| Input | JSON via stdin |

| Output | Text via stdout (shown to user) |

| Exit Codes | 0 = success, 2 = block, other = error |

Available Events

| Event | When It Fires | Common Use Cases |

| -------------------- | --------------------------- | ---------------------------------------- |

| session-start | New session begins | Environment setup, logging, checks |

| user-prompt-submit | Before user input processes | Validation, enhancement, filtering |

| tool-call | Before tool execution | Permission checks, logging, modification |

| assistant-response | After assistant responds | Formatting, logging, cleanup |

Hook Format Requirements

File Location

Project hooks:

.claude/hooks/session-start
.claude/hooks/user-prompt-submit

User-global hooks:

~/.claude/hooks/session-start
~/.claude/hooks/tool-call

Executable Requirements

Every hook MUST:

  • Have a shebang line:
#!/bin/bash
# or
#!/usr/bin/env node
# or
#!/usr/bin/env python3
  • Be executable:
chmod +x .claude/hooks/session-start
  • Handle JSON input from stdin:
#!/bin/bash
INPUT=$(cat)
FILE=$(echo "$INPUT" | jq -r '.input.file_path // empty')
  • Exit with appropriate code:
exit 0  # Success
exit 2  # Block operation
exit 1  # Error (logs but continues)

Input/Output Format

JSON Input Structure

Hooks receive JSON via stdin with event-specific data:

{
  "event": "tool-call",
  "timestamp": "2025-01-15T10:30:00Z",
  "session_id": "abc123",
  "current_dir": "/path/to/project",
  "input": {
    "file_path": "/path/to/file.ts",
    "command": "npm test",
    "old_string": "...",
    "new_string": "..."
  }
}

Stdout Output

  • Normal output shows in transcript
  • Empty output runs silently
  • Use stderr (>&2) for errors

Exit Codes

| Code | Meaning | Behavior |

| ------------ | ------- | -------------------------- |

| 0 | Success | Continue normally |

| 2 | Block | Stop operation, show error |

| 1 or other | Error | Log error, continue |

Schema Validation

Hooks should validate against the JSON schema:

Schema URL: https://github.com/pr-pm/prpm/blob/main/packages/converters/schemas/claude-hook.schema.json

Required frontmatter fields:

  • name - Hook identifier (lowercase, hyphens only)
  • description - What the hook does
  • event - Event type (optional, inferred from filename)
  • language - bash, typescript, javascript, python, binary (optional)
  • hookType: "hook" - For round-trip conversion

Common Mistakes

| Mistake | Problem | Solution |

| ---------------------- | ------------------------- | ------------------------------------ |

| Not quoting variables | Breaks on spaces | Always use "$VAR" |

| Missing shebang | Won't execute | Add #!/bin/bash |

| Not executable | Permission denied | Run chmod +x hook-file |

| Logging to stdout | Clutters transcript | Use stderr: echo "log" >&2 |

| Wrong exit code | Doesn't block when needed | Use exit 2 to block |

| No input validation | Security risk | Always validate JSON fields |

| Slow operations | Blocks Claude | Run in background or use PostToolUse |

| Absolute paths missing | Can't find scripts | Use $CLAUDE_PLUGIN_ROOT |

Basic Hook Examples

Shell Script Hook

#!/bin/bash
# .claude/hooks/session-start

# Log session start
echo "Session started at $(date)" >> ~/.claude/session.log

# Check environment
if ! command -v node &> /dev/null; then
  echo "Warning: Node.js not installed" >&2
fi

# Output to user
echo "Development environment ready"
exit 0

TypeScript Hook

#!/usr/bin/env node
// .claude/hooks/user-prompt-submit

import { readFileSync } from 'fs';

// Read JSON from stdin
const input = readFileSync(0, 'utf-8');
const data = JSON.parse(input);

// Validate prompt
if (data.prompt.includes('API_KEY')) {
  console.error('Warning: Prompt may contain secrets');
  process.exit(2); // Block
}

console.log('Prompt validated');
process.exit(0);

Best Practices

1. Keep Hooks Fast

Target < 100ms for PreToolUse hooks:

  • Cache results where possible
  • Run heavy operations in background
  • Use specific matchers, not wildcards

2. Handle Errors Gracefully

# Check dependencies exist
if ! command -v jq &> /dev/null; then
  echo "jq not installed, skipping" >&2
  exit 0
fi

# Validate input
FILE=$(echo "$INPUT" | jq -r '.input.file_path // empty')
if [[ -z "$FILE" ]]; then
  echo "No file path provided" >&2
  exit 1
fi

3. Use Shebangs

Always start with shebang:

#!/bin/bash
#!/usr/bin/env node
#!/usr/bin/env python3

4. Secure Sensitive Files

BLOCKED=(".env" ".env.*" "*.pem" "*.key")
for pattern in "${BLOCKED[@]}"; do
  case "$FILE" in
    $pattern)
      echo "Blocked: $FILE is sensitive" >&2
      exit 2
      ;;
  esac
done

5. Quote All Variables

# WRONG - breaks on spaces
prettier --write $FILE

# RIGHT - handles spaces
prettier --write "$FILE"

6. Log for Debugging

LOG_FILE=~/.claude-hooks/debug.log

# Log to file
echo "[$(date)] Processing $FILE" >> "$LOG_FILE"

# Log to stderr (shows in transcript)
echo "Hook running..." >&2

Publishing as PRPM Package

Package Structure

my-hook/
├── prpm.json          # Package manifest
├── HOOK.md            # Hook documentation
└── hook-script.sh     # Hook executable

prpm.json

{
  "name": "@username/hook-name",
  "version": "1.0.0",
  "description": "Brief description shown in search",
  "author": "Your Name",
  "format": "claude",
  "subtype": "hook",
  "tags": ["automation", "security", "formatting"],
  "main": "HOOK.md"
}

HOOK.md Format

---
name: session-logger
description: Logs session start/end times for tracking
event: SessionStart
language: bash
hookType: hook
---

# Session Logger Hook

Logs Claude Code session activity for tracking and debugging.

## Installation

This hook will be installed to `.claude/hooks/session-start`.

## Behavior

- Logs session start time to `~/.claude/session.log`
- Displays environment status
- Runs silent dependency checks

## Requirements

- bash 4.0+
- write access to `~/.claude/`

## Source Code

\`\`\`bash
#!/bin/bash
echo "Session started at $(date)" >> ~/.claude/session.log
echo "Environment ready"
exit 0
\`\`\`

Publishing Process

# Test locally first
prpm test

# Publish to registry
prpm publish

# Version bumps
prpm publish patch  # 1.0.0 -> 1.0.1
prpm publish minor  # 1.0.0 -> 1.1.0
prpm publish major  # 1.0.0 -> 2.0.0

Security Requirements

Input Validation

# Parse JSON safely
INPUT=$(cat)
if ! FILE=$(echo "$INPUT" | jq -r '.input.file_path // empty' 2>&1); then
  echo "JSON parse failed" >&2
  exit 1
fi

# Validate field exists
[[ -n "$FILE" ]] || exit 1

Path Sanitization

# Prevent directory traversal
if [[ "$FILE" == *".."* ]]; then
  echo "Path traversal detected" >&2
  exit 2
fi

# Keep in project directory
if [[ "$FILE" != "$CLAUDE_PROJECT_DIR"* ]]; then
  echo "File outside project" >&2
  exit 2
fi

User Confirmation

Claude Code automatically:

  • Requires confirmation before installing hooks
  • Shows hook source code to user
  • Warns about hook execution
  • Displays hook output in transcript

Hooks vs Skills vs Commands

| Feature | Hooks | Skills | Commands |

| ------------ | ---------------------- | ---------------------- | --------------------- |

| Format | Executable code | Markdown | Markdown |

| Trigger | Automatic (events) | Automatic (context) | Manual (/command) |

| Language | Any executable | N/A | N/A |

| Use Case | Automation, validation | Reference, patterns | Quick tasks |

| Security | Requires confirmation | No special permissions | Inherits from session |

Examples:

  • Hook: Auto-format files on save
  • Skill: Reference guide for testing patterns
  • Command: /review-pr quick code review
  • claude-hook-writer skill - Detailed hook development guidance
  • typescript-hook-writer skill - TypeScript-specific hook development
  • Claude Code Docs
  • Schema

Checklist for New Hooks

Before publishing:

  • [ ] Shebang line included
  • [ ] File is executable (chmod +x)
  • [ ] Validates all stdin input
  • [ ] Quotes all variables
  • [ ] Handles missing dependencies gracefully
  • [ ] Uses appropriate exit codes
  • [ ] Logs errors to stderr or file
  • [ ] Tests with edge cases (spaces, Unicode, missing fields)
  • [ ] Documents dependencies in HOOK.md
  • [ ] Includes installation instructions
  • [ ] Source code included in documentation
  • [ ] Clear description and tags in prpm.json
  • [ ] Version number is semantic

How to use it

Copy the folder

Take agentworkforce/creating-claude-hooks from the repository into ~/.claude/skills for personal use, or into .claude/skills inside a project.

Check the name does not clash

The agent identifies a skill by the name field in its header. Two skills with the same name cannot sit side by side — one of them will be ignored.