claude-hook-writer（Claude hook 写作）

Claude Hook Writer

Use this skill when creating or improving Claude Code hooks. This skill ensures hooks are secure, reliable, performant, and follow best practices.

When to Use This Skill

• Designing a new Claude Code hook
• Reviewing existing hook code
• Debugging hook failures
• Optimizing slow hooks
• Securing hooks that handle sensitive data
• Publishing hooks as PRPM packages

Core Principles

1. Security is Non-Negotiable

Hooks execute automatically with user permissions. They can read, modify, or delete any file the user can access.

ALWAYS validate and sanitize all input. Hooks receive JSON via stdin—never trust it blindly.

2. Reliability Over Features

A hook that works 99% of the time is a broken hook. Edge cases (Unicode filenames, spaces in paths, missing tools) will happen.

Test with edge cases before deploying.

3. Performance Matters

Hooks block operations. A 5-second hook means Claude waits 5 seconds before continuing.

Keep hooks fast. Run heavy operations in background.

4. Fail Gracefully

Missing dependencies, malformed input, and disk errors will occur.

Handle errors explicitly. Log failures. Return meaningful exit codes.

Hook Design Checklist

Before writing code, answer these questions:

What Event Does This Hook Target?

• PreToolUse - Before tool execution (modify input, validate, block)
• PostToolUse - After tool completes (format, log, cleanup)
• UserPromptSubmit - Before user input processes (validate, enhance)
• SessionStart - When Claude Code starts (setup, env check)
• SessionEnd - When Claude Code exits (cleanup, persist state)
• Notification - During alerts (desktop notifications, logging)
• Stop / SubagentStop - When responses finish (cleanup, summary)
• PreCompact - Before context compaction (save important context)

Common mistake: Using PostToolUse for validation (too late—tool already ran). Use PreToolUse to block operations.

Which Tools Should Trigger This Hook?

Be specific. matcher: "*" runs on every tool call.

Good matchers:

• "Write" - Only file writes
• "Edit|Write" - File modifications
• "Bash" - Shell commands
• "mcp__github__*" - All GitHub MCP tools

Bad matchers:

• "*" - Everything (use only for logging/metrics)

What Input Does This Hook Need?

Different tools provide different input. Check what's available:

# PreToolUse / PostToolUse
{
  "input": {
    "file_path": "/path/to/file.ts",     // Read, Write, Edit
    "command": "npm test",                // Bash
    "old_string": "...",                  // Edit
    "new_string": "..."                   // Edit
  }
}

Validate fields exist before using them:

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

Should This Be a Command Hook or Prompt Hook?

Command hooks (type: "command"):

• Fast (milliseconds)
• Deterministic
• Good for: formatting, logging, file checks

Prompt hooks (type: "prompt"):

• Slow (2-10 seconds)
• Context-aware (uses LLM)
• Good for: complex validation, security analysis, intent detection

Rule of thumb: Use command hooks unless you need LLM reasoning.

What Exit Code Communicates Success/Failure?

• exit 0 - Success (continue operation)
• exit 2 - Block operation (show error to Claude)
• exit 1 or other - Non-blocking error (log but continue)

For PreToolUse hooks:

• Exit 2 blocks the tool from running
• Exit 0 allows it (optionally with modified input)

For PostToolUse hooks:

• Exit codes don't block (tool already ran)
• Use exit 0 for success, 1 for logging errors

Security Requirements

MUST-HAVE Security Checks

Every hook must implement these:

1. Input Validation

#!/bin/bash
set -euo pipefail  # Exit on errors, undefined vars

INPUT=$(cat)

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

# Validate field exists
if [[ -z "$FILE" ]]; then
  echo "No file path in input" >&2
  exit 1
fi

2. Path Sanitization

# Validate file is in project
if [[ "$FILE" != "$CLAUDE_PROJECT_DIR"* ]]; then
  echo "File outside project: $FILE" >&2
  exit 2  # Block operation
fi

# Validate no directory traversal
if [[ "$FILE" == *".."* ]]; then
  echo "Path traversal detected: $FILE" >&2
  exit 2
fi

3. Sensitive File Protection

# Block list (extend as needed)
BLOCKED_PATTERNS=(
  ".env"
  ".env.*"
  "*.pem"
  "*.key"
  "*credentials*"
  ".git/*"
  ".ssh/*"
)

for pattern in "${BLOCKED_PATTERNS[@]}"; do
  if [[ "$FILE" == $pattern ]]; then
    echo "Blocked: $FILE matches sensitive pattern $pattern" >&2
    exit 2
  fi
done

4. Quote All Variables

Spaces and special characters in paths break unquoted variables:

# WRONG
cat $FILE              # Breaks on "my file.txt"
prettier --write $FILE # Fails with spaces

# RIGHT
cat "$FILE"              # Handles spaces
prettier --write "$FILE" # Safe

5. Use Absolute Paths for Scripts

# WRONG - relative path might not resolve
./my-script.sh

# RIGHT - explicit path
"${CLAUDE_PLUGIN_ROOT}/scripts/my-script.sh"

# ALSO RIGHT - use full path
/Users/username/.claude/scripts/my-script.sh

Reliability Requirements

Handle Missing Dependencies

# Check tool exists
if ! command -v prettier &> /dev/null; then
  echo "prettier not installed, skipping" >&2
  exit 0  # Success exit (just skip)
fi

# Check file exists
if [[ ! -f "$FILE" ]]; then
  echo "File not found: $FILE" >&2
  exit 1
fi

Set Timeouts

Default is 60 seconds. For slow operations, set explicit timeout:

{
  "hooks": [{
    "type": "command",
    "command": "./slow-operation.sh",
    "timeout": 10000  // 10 seconds
  }]
}

Or run in background:

# Don't block Claude
(heavy_operation "$FILE" &)
exit 0

Log Errors Properly

LOG_FILE=~/.claude-hooks/my-hook.log

# Log to stderr (shown in transcript)
echo "Hook failed: some reason" >&2

# Or log to file (for debugging)
echo "[$(date)] Error: some reason" >> "$LOG_FILE"

Don't log to stdout unless you want output in Claude's transcript.

Test With Edge Cases

Test files:

• "file with spaces.txt"
• "文件.txt" (Unicode)
• "src/deep/nested/path/file.tsx" (deep paths)
• "/absolute/path.txt" (absolute paths)
• "../../../etc/passwd" (traversal attempts)

Test input:

• Malformed JSON
• Missing fields
• Empty strings
• null values

Performance Requirements

Keep Hooks Fast

Target < 100ms for PreToolUse hooks. Longer hooks block Claude visibly.

Slow operations:

• Running tests: Run in background or use PostToolUse
• Type checking: Cache results by file hash
• Network calls: Avoid in hooks (use subagents instead)
• Heavy linting: Only lint changed file, not entire project

Use Specific Matchers

// BAD - runs on everything
{"matcher": "*", ...}

// GOOD - only file writes
{"matcher": "Write", ...}

// BETTER - only TypeScript writes
// (check file extension in hook)
{"matcher": "Write", ...}

Dedupe Expensive Operations

If multiple hooks match, they run in parallel. Dedupe with locks:

LOCK_FILE="/tmp/claude-hook-${SESSION_ID}-${HOOK_NAME}.lock"

if [[ -f "$LOCK_FILE" ]]; then
  exit 0  # Already running
fi

touch "$LOCK_FILE"
trap "rm -f '$LOCK_FILE'" EXIT  # Clean up on exit

# Do work here
expensive_operation

Code Templates

Template: Format On Save Hook

#!/bin/bash
set -euo pipefail

# Parse input
INPUT=$(cat)
FILE=$(echo "$INPUT" | jq -r '.input.file_path // empty')

# Validate
[[ -n "$FILE" ]] || exit 0
[[ -f "$FILE" ]] || exit 0
[[ "$FILE" == "$CLAUDE_PROJECT_DIR"* ]] || exit 0

# Check formatter installed
if ! command -v prettier &> /dev/null; then
  exit 0
fi

# Format by extension
case "$FILE" in
  *.ts|*.tsx|*.js|*.jsx)
    prettier --write "$FILE" 2>/dev/null || exit 0
    ;;
  *.py)
    black "$FILE" 2>/dev/null || exit 0
    ;;
  *.go)
    gofmt -w "$FILE" 2>/dev/null || exit 0
    ;;
esac

JSON config:

{
  "hooks": {
    "PostToolUse": [{
      "matcher": "Edit|Write",
      "hooks": [{
        "type": "command",
        "command": "/path/to/format-on-save.sh",
        "timeout": 5000
      }]
    }]
  }
}

Template: Block Sensitive Files Hook

#!/bin/bash
set -euo pipefail

INPUT=$(cat)
FILE=$(echo "$INPUT" | jq -r '.input.file_path // empty')

[[ -n "$FILE" ]] || exit 0

# Sensitive patterns
BLOCKED=(
  ".env"
  ".env.*"
  "*.pem"
  "*.key"
  "*secret*"
  "*credential*"
  ".git/*"
)

for pattern in "${BLOCKED[@]}"; do
  # Use case for glob matching
  case "$FILE" in
    $pattern)
      echo "🚫 Blocked: $FILE is a sensitive file" >&2
      echo "   Pattern: $pattern" >&2
      exit 2  # Block operation
      ;;
  esac
done

exit 0  # Allow

JSON config:

{
  "hooks": {
    "PreToolUse": [{
      "matcher": "Edit|Write",
      "hooks": [{
        "type": "command",
        "command": "/path/to/block-sensitive.sh"
      }]
    }]
  }
}

Template: Command Logger Hook

#!/bin/bash
set -euo pipefail

INPUT=$(cat)
COMMAND=$(echo "$INPUT" | jq -r '.input.command // empty')

[[ -n "$COMMAND" ]] || exit 0

LOG_FILE=~/claude-commands.log
mkdir -p "$(dirname "$LOG_FILE")"

# Log with timestamp and context
{
  echo "---"
  echo "Time: $(date '+%Y-%m-%d %H:%M:%S')"
  echo "Directory: $CLAUDE_CURRENT_DIR"
  echo "Command: $COMMAND"
} >> "$LOG_FILE"

exit 0

JSON config:

{
  "hooks": {
    "PreToolUse": [{
      "matcher": "Bash",
      "hooks": [{
        "type": "command",
        "command": "/path/to/command-logger.sh"
      }]
    }]
  }
}

Template: Prompt-Based Security Hook

{
  "hooks": {
    "PreToolUse": [{
      "matcher": "Write",
      "hooks": [{
        "type": "prompt",
        "prompt": "Analyze the file content being written to ${input.file_path}. Check if it contains: hardcoded API keys, AWS credentials, private keys, passwords, or secrets. Return {\"decision\": \"block\", \"reason\": \"<specific issue>\"} if found, otherwise {\"decision\": \"allow\"}.",
        "schema": {
          "type": "object",
          "properties": {
            "decision": {"enum": ["allow", "block"]},
            "reason": {"type": "string"}
          },
          "required": ["decision"]
        }
      }]
    }]
  }
}

Use sparingly: Prompt hooks take 2-10 seconds. Only use for critical security checks.

Testing Hooks

Manual Testing

Create test input:

# Test with sample JSON
echo '{
  "session_id": "test",
  "input": {
    "file_path": "/tmp/test.ts"
  }
}' | ./my-hook.sh

# Check exit code
echo $?  # 0 = success, 2 = blocked, 1 = error

Edge Case Testing

#!/bin/bash
# test-hook.sh

HOOK=./my-hook.sh

test_case() {
  local description="$1"
  local input="$2"
  local expected_exit="$3"

  echo "Testing: $description"

  echo "$input" | $HOOK
  actual_exit=$?

  if [[ $actual_exit -eq $expected_exit ]]; then
    echo "  ✓ PASS"
  else
    echo "  ✗ FAIL (expected exit $expected_exit, got $actual_exit)"
    return 1
  fi
}

# Test cases
test_case "Normal file" \
  '{"input":{"file_path":"/tmp/test.ts"}}' \
  0

test_case "Sensitive .env file" \
  '{"input":{"file_path":".env"}}' \
  2

test_case "File with spaces" \
  '{"input":{"file_path":"/tmp/my file.ts"}}' \
  0

test_case "Missing file_path" \
  '{"input":{}}' \
  1

test_case "Malformed JSON" \
  'not json' \
  1

echo "All tests passed"

Integration Testing

1. Register hook in Claude Code
2. Trigger the event (write file, run command)
3. Check transcript (Ctrl-R) for hook output
4. Verify expected behavior

Publishing Hooks as PRPM Packages

Package Structure

my-hook/
├── prpm.json          # Package manifest
├── hook.json          # Hook configuration
├── scripts/
│   └── my-hook.sh     # Hook script
└── README.md          # Documentation

prpm.json

{
  "name": "@yourname/my-hook",
  "version": "1.0.0",
  "description": "Brief description of what hook does (shown in search)",
  "author": "Your Name",
  "format": "claude",
  "subtype": "hook",
  "tags": [
    "formatting",
    "security",
    "automation"
  ],
  "main": "hook.json",
  "scripts": {
    "test": "./test-hook.sh"
  }
}

hook.json

{
  "hooks": {
    "PostToolUse": [{
      "matcher": "Edit|Write",
      "hooks": [{
        "type": "command",
        "command": "${CLAUDE_PLUGIN_ROOT}/scripts/my-hook.sh",
        "timeout": 5000
      }]
    }]
  }
}

Use ${CLAUDE_PLUGIN_ROOT} to reference scripts—expands to hook installation directory.

Advanced Hook Configuration

All hook types support optional fields for controlling execution behavior:

{
  "hooks": {
    "PreToolUse": [{
      "matcher": "Write",
      "hooks": [{
        "type": "command",
        "command": "./my-hook.sh",
        "timeout": 5000,
        "continue": true,              // Whether Claude continues after hook (default: true)
        "stopReason": "string",        // Message shown when continue is false
        "suppressOutput": false,       // Hide stdout from transcript (default: false)
        "systemMessage": "string"      // Warning message shown to user
      }]
    }]
  }
}

continue (boolean, default: true)

Controls whether Claude continues after hook execution.

When to use false:

• Security hooks that must block operations
• Validation hooks that found critical errors
• Hooks that require user intervention

{
  "type": "command",
  "command": "./validate-security.sh",
  "continue": false,
  "stopReason": "Security validation failed. Please review the detected issues before proceeding."
}

Exit code interaction:

• If hook exits with code 2 (block): continue is ignored, operation is blocked
• If hook exits with code 0 or 1: continue field determines behavior

stopReason (string)

Message displayed to user when continue: false. Should explain why execution stopped and what action is needed.

{
  "continue": false,
  "stopReason": "Pre-commit checks failed. Fix linting errors and try again."
}

suppressOutput (boolean, default: false)

Hides hook stdout from transcript mode (Ctrl-R). Stderr is always shown.

When to use true:

• Hooks that produce verbose output
• Debugging logs not useful to users
• Noisy background operations

{
  "type": "command",
  "command": "./sync-to-cloud.sh",
  "suppressOutput": true  // Don't show sync progress in transcript
}

Note: Always show critical errors via stderr, as stderr is never suppressed.

systemMessage (string)

Warning or info message shown to user when hook executes. Useful for non-blocking warnings.

{
  "type": "command",
  "command": "./check-dependencies.sh",
  "systemMessage": "⚠️  Some dependencies are outdated. Consider running 'npm update'."
}

Difference from stopReason:

• systemMessage: Informational, Claude continues
• stopReason: Critical, requires continue: false

README.md

# My Hook

Brief description.

## What It Does

- Clear, specific bullet points
- Mention which events it triggers on
- Mention which tools it matches

## Installation

prpm install @yourname/my-hook

## Requirements

- prettier (install: `npm install -g prettier`)
- jq (install: `brew install jq`)

## Configuration

Optional: How to customize behavior.

## Examples

Show example output or behavior.

## Troubleshooting

Common issues and fixes.

Publishing

# Test locally first
prpm test

# Publish
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

Common Pitfalls

❌ Pitfall 1: Not Quoting Variables

# BREAKS on spaces
prettier --write $FILE

# SAFE
prettier --write "$FILE"

❌ Pitfall 2: Trusting Input

# DANGEROUS - no validation
FILE=$(jq -r '.input.file_path')
rm "$FILE"

# SAFE - validate first
FILE=$(jq -r '.input.file_path // empty')
[[ "$FILE" == "$CLAUDE_PROJECT_DIR"* ]] || exit 2
[[ "$FILE" != *".env"* ]] || exit 2
rm "$FILE"

❌ Pitfall 3: Blocking Operations Too Long

# BLOCKS Claude for 30 seconds
npm test

# RUN IN BACKGROUND
(npm test &)
exit 0

❌ Pitfall 4: Wrong Exit Code

# PreToolUse hook that should block

if [[ $FILE == ".env" ]]; then
  echo "Don't edit .env" >&2
  exit 1  # WRONG - doesn't block, just logs error
fi

# RIGHT
if [[ $FILE == ".env" ]]; then
  echo "Blocked: .env is protected" >&2
  exit 2  # Blocks operation
fi

❌ Pitfall 5: Logging to stdout

# WRONG - appears in transcript
echo "Hook running..."

# RIGHT - stderr or file
echo "Hook running..." >&2
# or
echo "Hook running..." >> ~/.claude-hooks/debug.log

❌ Pitfall 6: Assuming Tools Exist

# BREAKS if prettier not installed
prettier --write "$FILE"

# SAFE
if command -v prettier &>/dev/null; then
  prettier --write "$FILE"
fi

Debugging Hooks

Enable Verbose Logging

#!/bin/bash
set -x  # Print commands as they execute

Check Transcript

Run Claude Code with Ctrl-R (transcript mode) to see hook execution:

PreToolUse hook: ./my-hook.sh
  stdout: Formatted file.ts
  stderr:
  exit: 0
  duration: 47ms

Test JSON Parsing

# Debug what jq extracts
INPUT=$(cat)
echo "$INPUT" | jq '.' >&2  # Show full JSON
echo "$INPUT" | jq -r '.input.file_path' >&2  # Show field

Check Environment Variables

echo "PROJECT_DIR: $CLAUDE_PROJECT_DIR" >&2
echo "CURRENT_DIR: $CLAUDE_CURRENT_DIR" >&2
echo "SESSION_ID: $SESSION_ID" >&2
echo "PLUGIN_ROOT: $CLAUDE_PLUGIN_ROOT" >&2

Quick Reference

Exit Codes

• 0 = Success (continue)
• 2 = Block operation (PreToolUse only)
• 1 or other = Non-blocking error

Hook Configuration Fields

Required:

• type - "command" or "prompt"
• command or prompt - Script path or prompt text

Optional:

• timeout - Max execution time in ms (default: 60000)
• continue - Continue after hook? (default: true)
• stopReason - Message when continue=false
• suppressOutput - Hide stdout from transcript (default: false)
• systemMessage - Warning message to user

Environment Variables

• $CLAUDE_PROJECT_DIR - Project root
• $CLAUDE_CURRENT_DIR - Current directory
• $SESSION_ID - Session identifier
• $CLAUDE_PLUGIN_ROOT - Hook installation directory
• $CLAUDE_ENV_FILE - File for persisting vars

JSON Input Structure

{
  "session_id": "...",
  "transcript_path": "...",
  "current_dir": "...",
  "input": {
    // Tool-specific fields
  }
}

Common jq Patterns

# Extract with default
$(jq -r '.input.file_path // empty')

# Extract array
$(jq -r '.input.files[]')

# Check field exists
if jq -e '.input.file_path' >/dev/null; then

# Parse entire object
INPUT_OBJ=$(jq '.input')

Final Checklist

Before publishing:

• Validates all stdin input
• Quotes all variables
• Uses absolute paths for scripts
• Blocks sensitive files
• Handles missing tools gracefully
• Sets reasonable timeout
• Logs errors to stderr or file
• Tests with edge cases
• Tests in real Claude session
• Documents dependencies
• README includes examples
• Semantic version number
• Clear description and tags

Resources

• Claude Code Hooks Docs
• Claude Hooks Best Practices Blog
• PRPM Hook Packages
• Hook Examples Repo