An inference-based intrusion detection system for AI agents that provides real-time scanning, automated quarantine, and human-in-the-loop security management.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install hopeids
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 hopeids 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 hopeIDS Security Skill?
hopeIDS is a sophisticated security layer designed specifically for AI agents within the Openclaw Skills ecosystem. It focuses on inference-based intrusion detection to identify and neutralize threats like command injection, credential theft, and jailbreak attempts before they reach the agent core. By prioritizing security invariants, such as metadata-only storage and a strict block-equals-abort policy, hopeIDS ensures that malicious payloads are never persisted or processed.
This skill is essential for developers building robust, production-grade agentic workflows where security and human oversight are paramount. It integrates seamlessly with messaging platforms like Telegram to provide real-time alerts and manual override capabilities for blocked communications, ensuring your Openclaw Skills remain protected against evolving prompt injection techniques.
hopeIDS Security Skill Use Cases
- Preventing jailbreak attempts and instruction overrides in public-facing AI agents.
- Protecting sensitive API keys and credentials from extraction through prompt injection.
- Monitoring multi-agent environments for malicious cross-agent communication.
- Implementing human-in-the-loop review processes for high-risk agent interactions.
- Applying differentiated security postures across various Openclaw Skills based on their specific risk profiles.
How hopeIDS Security Skill Works
- The system intercepts every incoming message before it is processed by the agent or the recall system.
- hopeIDS performs an automated scan using pattern matching and inference to assign a risk score and identify specific threat categories.
- If the risk score exceeds the defined threshold in strict mode, the message is blocked, a metadata-only quarantine record is created, and an alert is sent via Telegram.
- In non-strict mode, a security alert is injected into the message context, allowing the agent to proceed with awareness of the potential risk.
- Authorized users can review alerts and use commands like approve or trust to manage future interactions within the Openclaw Skills framework.
hopeIDS Security Skill Setup
To integrate hopeIDS into your environment, run the following setup command:
npx hopeid setup
After installation, restart your instance. You can then configure your settings in the plugin entry section of your configuration file to enable autoScan and define your risk thresholds for different Openclaw Skills.
hopeIDS Security Skill Data Schema & Taxonomy
hopeIDS generates a QuarantineRecord for every blocked interaction to maintain security without storing malicious payloads. The records are organized as follows:
| Field | Type | Description |
|---|---|---|
| id | String | Unique identifier for the quarantine record (e.g., q-7f3a2b). |
| ts | String | ISO 8601 timestamp of the blocked event. |
| agent | String | The specific agent targeted by the message. |
| intent | String | The detected threat category (e.g., command_injection). |
| risk | Number | The calculated risk probability (0.0 to 1.0). |
| patterns | Array | List of matched regex or keyword patterns. |
| contentHash | String | A hash of the original message for identification without raw storage. |
| status | String | The current resolution status (pending, approved, or rejected). |
hopeIDS Security Skill Advanced Features
- Granular per-agent configuration allows for customized risk thresholds and security modes across different Openclaw Skills.
- Real-time Telegram integration provides an interactive human-in-the-loop interface for immediate threat response and manual review.
- Automated quarantine management ensures malicious content is discarded while retaining actionable security metadata for forensic analysis.
- Sender trust system enables whitelisting of legitimate users to reduce false positives in high-traffic production environments.
- Comprehensive threat categorization covering critical risks ranging from data exfiltration to administrative impersonation.
SKILL.md
Loading
Related Openclaw Skills
exo-installer
The centralized command-line manager for installing, updating, and health-checking the entire E.x.O. suite of AI agent tools.
Failover Gateway
Automate high availability for your AI agent with an active-passive failover system that promotes a standby node when your primary instance fails.
Graphiti Knowledge Graph
A robust interface for managing knowledge graph facts, episodes, and entity relationships via the Graphiti API.
AgentFin Virtual Card Agent
A powerful API-driven skill for issuing and managing USDT-funded virtual payment cards for automated purchasing workflows.
Jasper ConfigGuard
A robust CLI tool for safely updating OpenClaw configurations with automated rollbacks and health monitoring.
Vagus: Phone MCP Connection
A bridge connecting AI agents to Android devices via the Model Context Protocol for real-time sensor data and physical device interactions.
METADATA
Featured*
