A security integration for OpenClaw that provides automated screening against prompt injections, jailbreaks, and PII exfiltration.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install promptdome
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 promptdome 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 PromptDome?
PromptDome is a specialized security layer designed to protect AI agents from malicious inputs. By incorporating these Openclaw Skills into your environment, you establish a proactive defense mechanism that scans every incoming message before it reaches your LLM. It functions as both a transparent gateway hook and an explicit tool that agents can use to verify untrusted content from external sources.
The integration ensures that your AI remains within its operational boundaries by identifying and neutralizing threats like system prompt overrides, social engineering, and data leaks. Whether you are running a customer-facing bot or a private research assistant, these Openclaw Skills provide the necessary hardening to prevent common LLM vulnerabilities.
PromptDome Use Cases
- Automating the detection of prompt injection and jailbreak attempts in real-time.
- Protecting sensitive information by screening for PII and credential exfiltration.
- Hardening Openclaw Skills against social engineering and fake system events.
- Scanning web search results or uploaded files for hidden malicious instructions before processing.
How PromptDome Works
- The promptdome-gate hook intercepts every message received by the OpenClaw gateway.
- The content is sent to the PromptDome engine, which analyzes it across 32 distinct attack categories.
- A security score is generated; scores of 70 or higher trigger an immediate block warning in the conversation.
- Moderate risks (scores between 40 and 69) result in a soft warning being injected to alert the agent and user.
- The promptdome_scan tool allows agents to manually trigger scans on specific data buffers, such as file contents or API responses.
PromptDome Setup
Quick Setup
Run the following command to install the necessary hooks and plugins automatically:
bash skills/promptdome/scripts/setup.sh --api-key sk_shield_live_YOUR_KEY
Manual Installation
- Copy the hook files to
~/.openclaw/hooks/promptdome-gate/. - Copy the plugin files to
~/.openclaw/extensions/promptdome/. - Add your API key to the
openclaw.jsonenvironment block:
{
"env": {
"PROMPTDOME_API_KEY": "your_key_here"
}
}
- Enable the hook using
openclaw hooks enable promptdome-gateand restart the gateway.
PromptDome Data Schema & Taxonomy
The skill manages security configuration and logging through the following structures:
| Data Point | Location / Format | Description |
|---|---|---|
| API Key | PROMPTDOME_API_KEY |
Environment variable for authentication |
| Scan Logs | ~/.openclaw/logs/promptdome-gate.log |
Local record of all scan results and scores |
| Hook State | openclaw.json |
Configuration for active hooks and tools |
| Block Threshold | JSON Value (70) | The score at which a message is considered a high-risk injection |
PromptDome Advanced Features
- Self-hosted API support by overriding the
PROMPTDOME_API_URLenvironment variable. - Fail-open architecture ensures that agent availability is maintained even if the security API is unreachable.
- Multilingual evasion detection covering 18 different languages.
- Comprehensive coverage for 32 attack categories, including HTML/DOM injection and agentic chain poisoning.
SKILL.md
Loading
Related Openclaw Skills
Mission Claw Activity Logging
A command-line tool for logging AI agent activities, tracking token usage, and monitoring project progress through a local dashboard.
Warm Wake
A specialized grounding tool that restores an AI agent's sense of self and long-term memory before it resumes task execution.
Wake State
A persistence layer for AI agents to maintain state, tasks, and checkpoints across crashes and context window resets.
Trust Verifier
A security auditing tool that assesses the provenance, integrity, and publisher reputation of Openclaw Skills to ensure safe installation and execution.
Strategy Advisor
A strategic decision-making engine for Evvo Labs that leverages five industry-standard frameworks to provide actionable business advice.
GitHub Issue Manager
A powerful skill for automating GitHub issue lifecycle management and autonomous sub-agent orchestration.
Featured*
