Claude Code Rm Guard

claude-code-rm-guard

**Important to note:** Claude Code's sandboxing feature (https://code.claude.com/docs/en/sandboxing) seems like a much better way to solve this problem, or otherwise using devcontainers. If that isn't what you want, then maybe this hook will help!

🛡️ Security hook for [Claude Code](https://docs.anthropic.com/en/docs/claude-code) that prevents destructive file operations outside your working directory.

Stop Claude from accidentally running `rm -rf ~/` or other catastrophic commands.

Quick Install

curl -fsSL https://raw.githubusercontent.com/elertan/claude-code-rm-guard/main/install.py | python3

The installer will:

1. Download the hook to ~/.claude/hooks/validate-rm.py
2. Add the hook configuration to ~/.claude/settings.json
3. Optionally add ask permissions for rm/unlink/rmdir (recommended)

**Uninstall:**

curl -fsSL https://raw.githubusercontent.com/elertan/claude-code-rm-guard/main/install.py | python3 - --uninstall

Features

• Path Validation: Blocks rm, unlink, rmdir, and shred commands targeting paths outside your project directory
• Shell-Aware Parsing: Uses Python's shlex with punctuation_chars=True for proper handling of pipes, chains, and quoted arguments
• Recursive Analysis: Detects dangerous commands hidden in sudo, xargs, find -exec, bash -c, and other wrappers
• Safety-First Blocking: Commands with unresolvable paths (variables, globs, command substitution) are blocked by default
• Zero Dependencies: Pure Python 3, no external packages required

How It Works

When Claude Code attempts to run a Bash command:

1. Hook Receives Input: Claude Code sends JSON via stdin with tool_name, tool_input.command, and cwd
2. Command Parsing: The script uses shlex to tokenize the command, splitting on ;, |, &&, ||
3. Recursive Analysis: Each subcommand is analyzed, unwrapping sudo, bash -c, etc.
4. Path Resolution: Target paths