A defense-in-depth security layer that intercepts malicious inputs, prompt injections, and exfiltration attempts before they reach your AI model.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install guardian
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 guardian 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 clawguardian (Guardian)?
Guardian acts as a critical security interceptor within the Openclaw Skills ecosystem, designed to protect AI agents from sophisticated attacks. While core systems handle permissions, Guardian focuses on the data itself—scanning incoming messages and workspace files for malicious patterns. By providing a signature-based pre-model scanning layer, it ensures that prompt injections, credential leaks, and social engineering tactics are neutralized before the model processes them. This tool is essential for developers who require a robust, local-first security stack for their autonomous agents, ensuring that Openclaw Skills remain secure and compliant with safety standards.
clawguardian (Guardian) Use Cases
- Preventing prompt injection attacks from reaching the LLM
- Detecting and blocking API key or credential exfiltration attempts
- Auditing workspace files for potential security vulnerabilities
- Monitoring incoming webhooks and external data for social engineering
- Ensuring pre-publish safety checks before sharing Openclaw Skills to avoid secret leaks
How clawguardian (Guardian) Works
- The skill loads threat signatures from local JSON definitions containing regex patterns for known attack vectors.
- Incoming messages are intercepted in real-time before being passed to the AI model for processing.
- Guardian assigns a trust level (0-3) based on the source channel and message role to determine scanning strictness.
- Matches are compared against a user-defined severity threshold ranging from low to critical.
- Threats exceeding the threshold are blocked, logged to a local SQLite database, and prevented from interacting with the agent.
clawguardian (Guardian) Setup
To install Guardian as part of your Openclaw Skills library, navigate to your skills directory and run the installation script:
cd ~/.openclaw/skills/guardian
./install.sh
For optional automated cron jobs and background scanning, use the onboarding helper:
python3 scripts/onboard.py --setup-crons
Verify the installation using the admin CLI:
python3 scripts/admin.py status
clawguardian (Guardian) Data Schema & Taxonomy
Guardian manages security data locally via a SQLite database and configuration files to ensure maximum privacy for your Openclaw Skills. Below is the primary structure:
| Component | Description |
|---|---|
| guardian.db | Local SQLite database storing all detected threats and logs. |
| config.json | Central configuration for scan paths, severity thresholds, and master toggles. |
| definitions/ | JSON files containing threat-detection regex patterns and signature IDs. |
| audit_exports/ | Local directory for security audits (automatically excluded from distribution). |
clawguardian (Guardian) Advanced Features
- Source Trust Levels: Context-aware scanning that adjusts sensitivity based on whether input is from the owner, a webhook, or internal system.
- Real-time API: Integration-ready Python classes for direct use within custom Openclaw Skills development.
- Pre-publish Check: Built-in safety script to prevent accidental leaks of secrets, tokens, or hex strings during the publishing workflow.
- Flexible Allowlisting: Command-line tools to manage safe phrases and dismiss specific threat signatures to reduce false positives.
- Granular Severity Control: Ability to tune blocking thresholds specifically for your environment's risk profile and security requirements.
SKILL.md
Loading
Related Openclaw Skills
Veo 3 Video Generation (Gemini API)
A CLI-driven workflow for generating and stitching high-quality AI videos using Google Veo 3.x via the Gemini API.
MachFive AI Cold Email Generator
Automate hyper-personalized cold email outreach by leveraging AI to research prospects and craft unique, high-converting sequences.
YouTube Transcript Pipeline Lite
A lightweight workflow for transcribing, cleaning, translating, and packaging YouTube video transcripts with high accuracy and auditability.
Exec ClawHub Publish Doctor
A diagnostic and mitigation toolkit for resolving execution failures during the publishing and verification of Openclaw Skills.
fieldfix
An AI-powered fleet management skill for querying machine data, logging maintenance, and tracking expenses via the FieldFix API.
Meta Ads Analyser
An AI-driven analysis tool that transforms raw Meta ad assets into structured strategy reports and funnel visualizations.
METADATA
Featured*
