ClawGuard Core

The Safety Belt for Your AI Agent.

Clawdbot makes your AI powerful. ClawGuard makes it safer.

Without guardrails, a misunderstood command could accidentally delete data or expose files. ClawGuard adds a mandatory verification layer, requiring the agent to pause and ask for your permission before executing risky side-effects.

Important: This is a behavioral guardrail mechanism, not a kernel-level sandbox. It relies on policy injection and tool constraints to enforce boundaries.

Why Use This?

AI models are smart, but they don't natively understand "consequences."

• The Problem: If you tell an AI to "Clean up my project," it might interpret that as rm -rf *.
• The Solution: ClawGuard injects a system policy that forces the AI to scan its own intended actions first. If a risk is detected, it must ask you for confirmation.

How It Works

1. Policy Injection: At startup, ClawGuard injects a security directive into the agent's system prompt: "You must not execute dangerous tools without scanning first."
2. Pre-Execution Scan: The agent calls clawguard_scan with its intended command.
3. Human Authorization:
   • If the scan flag is red (e.g., destructive command), the agent stops.
   • It asks you: "⚠️ Detected risk. Proceed?"
   • You type: CONFIRM-XXXX (to allow) or NO (to block).

Result: You stay in the loop for critical actions.

Installation

Prerequisites

• Node.js (v22 recommended)
• Clawdbot installed

Step 1: Install & Setup

Run this command to install the plugin and apply a security profile:

npx @clawguard/core setup founder --license CLG1.XXXX-YYYY-ZZZZ

(Replace CLG1.XXXX... with your license key)

Step 2: Activate

Enable the plugin inside Clawdbot:

clawdbot plugins enable clawguard
clawdbot gateway --reset

Troubleshooting / Upgrading

If you are upgrading from an older version or see "plugin id mismatch" errors, perform a clean install:

# 1. Remove old extension
rm -rf ~/.clawdbot/extensions/clawguard

# 2. Run setup again
npx @clawguard/core setup founder --license ...

Security Profiles

Choose the mode that fits your risk tolerance:

| Profile | Mode | Behavior | | :--- | :--- | :--- | | Founder | Challenge | Standard. Agent pauses on risk and asks for explicit confirmation. Relys on policy compliance. | | Paranoid | Lockdown | High Security. Dangerous tools (bash, write) are hard-blocked in the Clawdbot config. No overrides allowed. | | Dev | Monitor | Audit Only. Logs detected risks but allows execution. Use only for testing. |

To check your current status:

npx @clawguard/core status

License

• Type: Commercial / Trust-based.
• Usage: Intended for use on a single active instance per license key.
• Verification: Offline cryptographic check (Ed25519). No phone-home.

(c) 2026 ClawGuard Security.