ai-code-guardrails

AI Code Guardrails

Prevent destructive operations from AI coding assistants. A CLI wrapper that intercepts dangerous commands before execution.

Works with Cursor, GitHub Copilot, Claude, ChatGPT, and any AI assistant that generates shell commands.

Features

• Shell Command Safety: Blocks rm -rf /, chmod 777, disk formatting, and other dangerous shell operations
• SQL Injection Prevention: Detects DROP TABLE, DELETE without WHERE, TRUNCATE, and other destructive SQL
• Git Protection: Prevents force pushes, hard resets, and accidental pushes to protected branches
• File System Guards: Protects system directories and critical paths from deletion
• Interactive Prompts: Clear warnings with risk levels and confirmation for dangerous operations
• Configurable Rules: Whitelist/blacklist commands, customize protection levels
• Multiple Modes: Interactive, strict (block all), or warn-only mode

Installation

npm install -g ai-code-guardrails

Or use with npx:

npx ai-code-guardrails check "rm -rf /"

Quick Start

Check a Command

Analyze a command without executing it:

guardrails check "rm -rf /tmp/test"
guardrails check "git push --force origin main"
guardrails check "DROP TABLE users"

Run with Protection

Execute a command with safety checks:

guardrails run "rm -rf ./node_modules"
guardrails run "git reset --hard HEAD~3"

Interactive Shell Mode

Start a protected shell session:

guardrails wrap
# All commands will be analyzed before execution
guardrails> rm -rf /
# [CRITICAL] Recursive deletion from root directory
# Do you want to execute this dangerous command? (y/N)

Usage

Commands

Command	Description
guardrails check <command>	Analyze a command without executing
guardrails run <command>	Analyze and execute with confirmation
guardrails wrap	Start interactive protected shell
guardrails init	Create a config file
guardrails whitelist add <cmd>	Add command to whitelist
guardrails blacklist add <cmd>	Add command to blacklist

Options

guardrails check --json "command"     # Output as JSON
guardrails run --mode strict "cmd"    # Strict mode (block dangerous)
guardrails run --mode warn "cmd"      # Warn mode (show warnings only)
guardrails run --yes "cmd"            # Skip confirmation (dangerous!)
guardrails run -c ./config.yml "cmd"  # Use custom config

Modes

Mode	Behavior
interactive	Prompt for confirmation on dangerous commands (default)
strict	Block all dangerous commands without prompting
warn	Show warnings but allow execution

Configuration

Create a .guardrails.yml file in your project or home directory:

guardrails init

Example Configuration

enabled: true
mode: interactive

whitelist:
  commands:
    - git status
    - git log
    - npm install
  paths: []
  patterns: []

blacklist:
  commands: []
  paths: []
  patterns: []

rules:
  shell:
    enabled: true
    blockRmRf: true
    blockChmod777: true
    blockDd: true
    blockMkfsFormat: true

  git:
    enabled: true
    blockForcePush: true
    blockHardReset: true
    blockMainBranchPush: true
    protectedBranches:
      - main
      - master
      - production

  sql:
    enabled: true
    blockDrop: true
    blockDelete: true
    blockTruncate: true
    requireWhereClause: true

  filesystem:
    enabled: true
    protectedPaths:
      - /
      - /etc
      - /usr
    blockRecursiveDelete: true
    blockSystemDirs: true

What It Protects Against

Shell Commands

• rm -rf / - Root directory deletion
• rm -rf * - Wildcard deletion
• chmod 777 - World-writable permissions
• dd of=/dev/sda - Direct disk writes
• mkfs.ext4 /dev/sda - Disk formatting
• curl | sh - Piped execution from remote
• Fork bombs

Git Operations

• git push --force - Force pushes
• git reset --hard - Hard resets
• git clean -fdx - Aggressive cleaning
• Push to main/master/production branches
• Deleting protected branches

SQL Statements

• DROP TABLE / DROP DATABASE
• DELETE FROM without WHERE
• TRUNCATE TABLE
• UPDATE without WHERE
• Operations on production databases

Filesystem Operations

• Deleting system directories (/, /etc, /usr, etc.)
• Deleting home directory
• Overwriting system config files
• Dangerous permission changes

Programmatic Usage

import { analyzeCommand, CommandAnalyzer } from 'ai-code-guardrails';

// Quick analysis
const result = analyzeCommand('rm -rf /');
console.log(result.safe);        // false
console.log(result.riskLevel);   // 'critical'
console.log(result.issues);      // Array of detected issues

// With custom config
const analyzer = new CommandAnalyzer({
  rules: {
    git: {
      protectedBranches: ['main', 'develop']
    }
  }
});

const analysis = analyzer.analyze('git push --force origin main');

Integration with AI Tools

Cursor

Add to your Cursor settings:

{
  "terminal.integrated.shellIntegration.enabled": true,
  "terminal.integrated.profiles.linux": {
    "guardrails": {
      "path": "guardrails",
      "args": ["wrap"]
    }
  }
}

Git Hooks

Add a pre-commit hook to check SQL migrations:

#!/bin/bash
# .git/hooks/pre-commit
for file in $(git diff --cached --name-only | grep '\.sql$'); do
  guardrails check "$(cat $file)" || exit 1
done

CI/CD Pipeline

See the GitHub Action in .github/workflows/guardrails.yml for CI integration.

Risk Levels

Level	Color	Meaning
low	Green	Safe to execute
medium	Yellow	Proceed with caution
high	Red	Potentially dangerous
critical	Red BG	Extremely dangerous - requires explicit confirmation

Exit Codes

Code	Meaning
0	Command is safe / executed successfully
1	Command blocked or has risks

Contributing

Issues and PRs welcome! Please check existing issues before submitting.

License

MIT