A security-first execution layer that enforces governed permissions for shell commands, outbound messaging, and external API actions.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install claw-guardirails-skill
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-guardirails-skill 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 Guardrails Safe Tools?
Guardrails Safe Tools is a critical security framework designed for autonomous AI agents that need to interact with the physical or digital world. It replaces raw, high-risk execution methods with safe wrappers—specifically safe_exec, safe_send, and safe_action—to ensure every potentially destructive operation is vetted by a permission resolver. By integrating this into your Openclaw Skills library, you ensure that agents cannot bypass security protocols or accidentally execute harmful system changes.
The skill moves beyond simple prompt-based instructions by implementing hard-coded tool contracts. This means that even if an agent attempts a risky operation, the underlying architecture requires explicit validation against defined security policies. It is an essential component for any production-grade deployment where system integrity and data safety are non-negotiable.
Guardrails Safe Tools Use Cases
- Executing system-level shell commands within a restricted and governed environment.
- Sending automated notifications or posts to communication channels like Slack or Telegram without risking spam or unauthorized disclosure.
- Triggering external API mutations, such as deleting database records or modifying cloud resources, through a controlled interface.
- Implementing human-in-the-loop approval workflows for high-stakes tasks that require manual verification.
- Sanitizing input arguments to prevent shell injection and ensuring narrow working directory access.
How Guardrails Safe Tools Works
- The AI agent identifies an action that involves a side effect, such as running a script or messaging a user.
- Instead of using a direct tool, the agent invokes a safe variant from the Openclaw Skills set, like safe_exec.
- The tool request is intercepted by a permission resolver which evaluates the command, arguments, and context.
- The resolver returns one of three statuses: allow, deny, or require_approval.
- If allowed, the action executes; if denied, the agent receives a failure code; if approval is required, the system pauses until a human grants permission via a specific command or interface.
Guardrails Safe Tools Setup
To deploy Guardrails Safe Tools, you must register the tool definitions within your agent's execution environment. Ensure your resolver logic is configured to handle the standard tool contracts.
# Install the skill into your local environment
openclaw skills add guardrails-safe-tools
Configure your environment variables to point to your permission resolver endpoint to manage allow/deny logic dynamically.
Guardrails Safe Tools Data Schema & Taxonomy
The skill utilizes structured JSON schemas to maintain input hygiene and ensure explicit resource targeting.
| Tool | Primary Keys | Description |
|---|---|---|
| safe_exec | command, args, cwd | Executes shell commands with mandatory directory scoping. |
| safe_send | channel, message, channelType | Routes outbound messages through verified communication adapters. |
| safe_action | action, payload, resources | Handles generic API mutations with detailed resource metadata. |
Guardrails Safe Tools Advanced Features
- Multi-stage resolution: Supports complex logic where permissions depend on the specific resources being accessed.
- Approval ID tracking: Generates unique IDs for tasks requiring manual intervention, allowing for asynchronous workflows.
- Context-aware security: Evaluates the sender, channel, and agent identity to determine execution rights.
- Reversible operation preference: Encourages agents to suggest non-destructive or read-only alternatives before attempting mutable actions.
- Audit logging: Every call made via these Openclaw Skills is logged with its resolution status for security auditing.
SKILL.md
Loading
Related Openclaw Skills
Crypto Auto Progression
An automated execution engine that drives the crypto-hedge-backtest project through cron-based task management and reporting.
Feishu Multi-Agent Orchestration
A production-grade blueprint for orchestrating a team of AI agents through Feishu using a foreman-worker model and file-driven task queues.
Feishu App Setup Automation
An automated browser-based skill designed to handle the entire lifecycle of Feishu open platform application configuration for AI agents.
AI Companion Agent for Feishu
A comprehensive framework for deploying an autonomous AI companion on Feishu featuring long-term memory, voice messaging, and image generation.
Garmin Health Analysis
Transform your Garmin Connect data into actionable health insights and interactive dashboards using natural language commands.
Crypto Price & Chart
A professional-grade utility for AI agents to fetch live cryptocurrency pricing and generate visual candlestick charts via leading market APIs.
Featured*
