Destructive Command Guard — Block Dangerous Shell Commands from AI Agents

Introduction

Destructive Command Guard (DCG) is a lightweight Rust tool that acts as a safety layer between AI coding agents and your shell. It intercepts commands before execution, checks them against a configurable blocklist of destructive patterns, and prevents accidental data loss from force pushes, hard resets, and file deletions.

What DCG Does

• Intercepts shell commands before execution and checks for destructive patterns
• Blocks dangerous git operations like force push, reset --hard, and clean -fd
• Prevents accidental rm -rf and other file system destructive commands
• Provides configurable allow/deny lists for fine-grained control
• Logs blocked commands for audit and review

Architecture Overview

DCG is a single static Rust binary with zero dependencies. It operates as a command wrapper or shell hook, parsing the command line before passing it to the underlying shell. Pattern matching uses a compiled rule set that evaluates in microseconds, adding negligible latency to command execution.

Self-Hosting & Configuration

• Install via cargo: cargo install destructive-command-guard
• Or download prebuilt binaries from GitHub releases
• Configure rules in ~/.config/dcg/rules.toml
• Integrate as a git hook or shell alias
• Set environment-specific overrides for CI/CD pipelines

Key Features

• Zero-dependency single binary written in Rust
• Sub-millisecond command evaluation overhead
• Configurable rules with allow and deny patterns
• Works with any AI coding agent that executes shell commands
• Audit logging of all blocked commands

Comparison with Similar Tools

• Git hooks — Per-repo only; DCG protects globally across all repos
• Shell aliases — Easy to bypass; DCG wraps the actual binary
• Husky — JS-specific git hooks; DCG is language-agnostic
• pre-commit — Runs checks on staged files; DCG blocks dangerous commands themselves

FAQ

Q: Does DCG add latency to commands? A: Negligible. Rule evaluation takes microseconds since patterns are compiled at startup.

Q: Can I allow specific destructive commands? A: Yes, configure allow rules in rules.toml to permit specific patterns when needed.

Q: Does it work with Claude Code, Codex, etc.? A: Yes, DCG works with any tool that executes shell commands, including all major AI coding agents.

Q: What happens when a command is blocked? A: DCG prints a warning explaining why the command was blocked and exits with a non-zero status code.

Sources

• https://github.com/Dicklesworthstone/destructive_command_guard