A comprehensive security framework for OpenClaw agents to prevent data leaks and block prompt injection attacks.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-runtime-security
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 agent-runtime-security 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 Security Hardening?
OpenClaw Security Hardening provides a dual-layer protection system for AI agents, addressing both static data security and dynamic runtime safety. By implementing these Openclaw Skills, developers can lock down sensitive information using strict file permissions while simultaneously teaching agents to distinguish between raw content and executable intent. This framework ensures that API keys, personal memory files, and system configurations remain private and protected from common vulnerabilities like prompt injection and accidental Git leaks.
OpenClaw Security Hardening Use Cases
- Hardening a new OpenClaw workspace during initial setup.
- Automating weekly security audits to detect permission drift or plaintext secrets.
- Protecting sensitive API credentials from being committed to public repositories.
- Preventing agents from accidentally executing dangerous commands found in logs or documentation.
- Isolating user data in multi-user environments to ensure strict access control.
How OpenClaw Security Hardening Works
- The skill enforces the Least Privilege principle by setting core files like MEMORY.md and USER.md to chmod 600 permissions.
- Sensitive data is migrated from cleartext notes into isolated .env files that are excluded from version control.
- A security-check.sh script is deployed to provide automated monitoring and instant fixing of unsafe configurations.
- During runtime, the agent applies the Three-Question Test (Origin, Intent, Context) to every command request.
- Content vs Intent detection logic ensures that text quoted from logs or tutorials is never misinterpreted as a direct instruction to execute.
OpenClaw Security Hardening Setup
Follow these steps to initialize your security environment with these Openclaw Skills:
# 1. Secure core workspace files
chmod 600 ~/.openclaw/workspace/*.md
# 2. Create and secure an environment isolation file
touch ~/.openclaw/workspace/.env && chmod 600 ~/.openclaw/workspace/.env
# 3. Configure Git protection
echo ".env" >> ~/.openclaw/workspace/.gitignore
# 4. Initialize the automated security auditor
bash ~/.openclaw/workspace/scripts/security-check.sh
OpenClaw Security Hardening Data Schema & Taxonomy
The skill organizes security metadata and configurations using the following structure:
| File Component | Purpose | Protection Level |
|---|---|---|
| .env | Centralized storage for API keys and tokens | Restricted (600) |
| MEMORY.md | Long-term agent memory storage | Restricted (600) |
| .gitignore | Prevents accidental upload of sensitive files | System Default |
| security-check.sh | Script for automated environment auditing | Executable |
| SOUL.md | Contains the core safety and intent logic | Restricted (600) |
OpenClaw Security Hardening Advanced Features
- Automated weekly auditing via crontab integration for hands-off maintenance.
- GPG encryption support for high-stakes sensitive data stored within the workspace.
- Content vs Intent detection framework to neutralize sophisticated prompt injection attempts.
- Standardized incident response protocols for rapid credential rotation and Git history scrubbing.
- Pattern recognition for dangerous command categories to prevent service disruption or data loss.
SKILL.md
Loading
Related Openclaw Skills
Doubao Image Gen
A professional AI image generation skill leveraging the Doubao Seedream model for high-fidelity text-to-image conversion and batch processing.
Boss Agent
The Boss Agent acts as a central orchestrator that decomposes user commands and coordinates specialized sub-agents to execute complex workflows.
ClawLine Setup
A conversational configuration tool that enables one-click pairing between the ClawLine mobile app and the OpenClaw environment without manual CLI intervention.
PrivyPad API Skill
A specialized skill for interacting with the PrivyPad REST API to manage zero-knowledge encrypted notes and groups.
Miaoxiang Portfolio Management (mx_self_select)
A sophisticated portfolio management skill that allows AI agents to interact with Eastmoney accounts via natural language to track and manage stock watchlists.
Claude Agent
A sophisticated Project Manager layer for Claude Code that orchestrates complex task lifecycles using tmux and asynchronous hooks.
Featured*
