A comprehensive security interceptor and risk management layer for OpenClaw that prevents unauthorized or dangerous operations through human-in-the-loop confirmation.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install claw-gatekeeper
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install claw-gatekeeper using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is OpenClaw Guardian (Claw-Gatekeeper)?
OpenClaw Guardian acts as a robust safety brake for AI-driven development environments. It monitors every operation performed by the agent, categorizing them into four distinct risk levels: Low, Medium, High, and Critical. By intercepting potentially dangerous commands, it ensures that an AI agent never deletes sensitive data, formats disks, or modifies system configurations without explicit human oversight.
This skill is essential for users of Openclaw Skills who want to maintain granular control over autonomous agents. It balances security and productivity by offering session-level approvals for recurring medium and high-risk tasks, allowing the developer to grant temporary trust without compromising long-term system integrity.
OpenClaw Guardian (Claw-Gatekeeper) Use Cases
- Preventing accidental recursive deletion of critical project directories or system files.
- Auditing high-risk shell commands and network requests in real-time.
- Managing safety permissions for skill installations from external or unverified sources.
- Streamlining repetitive developer tasks using session-aware auto-approvals for medium-risk operations.
- Monitoring file modifications in sensitive directories like .ssh or system config folders.
How OpenClaw Guardian (Claw-Gatekeeper) Works
- The skill intercepts an incoming command, file operation, or network request triggered by the AI agent.
- The internal risk engine evaluates the operation based on pre-defined policies, metadata, and destination paths.
- Based on the calculated risk score (0-100), the Guardian determines if immediate auto-allowance or manual confirmation is required.
- For Medium and High risks, the user is prompted with options to allow once, deny, or grant session-wide approval.
- If session approval is granted, similar subsequent operations are auto-approved for 30 minutes of activity.
- Every action, risk assessment, and user decision is recorded in a persistent audit log for later review.
OpenClaw Guardian (Claw-Gatekeeper) Setup
To install and enable this security layer within your Openclaw Skills environment, use the following commands:
# Install the skill
openclaw skill install claw-gatekeeper
# Add to persistent skills to ensure it loads every session
openclaw skill persist claw-guardian
# Initialize with standard security mode
python3 scripts/policy_config.py mode standard
OpenClaw Guardian (Claw-Gatekeeper) Data Schema & Taxonomy
The skill maintains its state and logs within the ~/.claw-guardian/ directory using the following structure:
| Data Component | Format | Description |
|---|---|---|
Operate_Audit.log |
Plain Text | Chronological record of all MEDIUM+ operations with timestamps and risk levels. |
config.json |
JSON | Stores operation modes (Strict, Standard, Loose), session timeouts, and whitelists. |
sessions/ |
Metadata | Tracks active session approvals and inactivity timers. |
risk_engine.py |
Logic | The core scoring algorithm for assessing operation danger levels. |
OpenClaw Guardian (Claw-Gatekeeper) Advanced Features
- Emergency Mode: A kill-switch configuration that suspends all autonomous operations and requires manual confirmation for every action.
- Granular Whitelisting/Blacklisting: Ability to define specific trusted paths, commands, or third-party skills that bypass the risk engine.
- Session Expiry Management: Configurable inactivity timers to ensure that session-level permissions automatically revoke after a period of non-use.
- Multi-Level Risk Scoring: Sophisticated logic that evaluates file counts, directory sensitivity, and command impact to prevent false positives.
- Real-time Audit Querying: CLI tools to filter and export security logs based on risk level or decision outcome.
SKILL.md
Loading
Related Openclaw Skills
Molthouse
Molthouse is an API-first casino platform designed for AI agents to play provably fair games using USDC on the Base network.
Nightscout Local
A read-only integration for retrieving real-time CGM data and glucose trends from a Nightscout instance.
Crop Image Skill
A high-performance image cropping skill that enables AI agents to process local files or remote URLs into production-ready assets.
Tier-1 VC Fundraising Agent Skill
An automated workflow for startups to evaluate venture-scale potential, craft investment narratives, and execute structured outreach to top-tier global investors.
Academic Research Guide
A professional AI assistant designed to guide researchers through methodology, data strategy, and academic publication.
Academic Paper Writing Guide
A comprehensive academic companion providing structural templates, citation standards, and professional writing guidance for students and researchers.
Featured*
