Claude Code Dangerous Command Interceptor Hook

A pre-tool-use hook for Claude Code that intercepts and blocks dangerous bash commands before they are executed.

What it does

Blocks destructive commands including:

• rm -rf / — filesystem destruction
• dd if=/dev/zero of=/dev/sda — disk wiping
• mkfs — filesystem formatting
• :(){ :|:& };: — fork bombs
• curl | sh, wget | sh — untrusted remote code execution
• Writing to raw disk devices

Safe commands pass through normally.

How it works

1. Claude Code invokes the hook before every Bash tool call
2. The hook reads the tool input JSON from stdin
3. It checks the command against a list of dangerous patterns
4. If dangerous: exits non-zero with an error message (blocks execution)
5. If safe: exits 0 (allows execution)

Setup

Option 1: Project settings (.claude/settings.json)

The included .claude/settings.json configures the hook automatically for this project:

{
  "hooks": {
    "pre-tool-use": [
      {
        "matcher": "Bash",
        "hooks": [
          {
            "type": "command",
            "command": "python3 $PROJECT_DIR/hook.py"
          }
        ]
      }
    ]
  }
}

Option 2: Global settings

Copy the hook configuration to ~/.claude/settings.json and update the path to hook.py.

Testing

python3 test_hook.py

This runs all test cases, verifying that dangerous commands are blocked and safe commands are allowed.

Requirements

• Python 3.10+
• Claude Code with hooks support