A mandatory, immutable security protocol designed to protect LLM agents from prompt injections, malicious code execution, and data exfiltration.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install clawgatesecure
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 clawgatesecure 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 ClawGateSecure?
ClawGateSecure is a sophisticated security framework that serves as the immutable core for AI agents. It establishes a zero-trust environment where all incoming data is treated as potentially malicious until verified. By incorporating this into your library of Openclaw Skills, you ensure your agents are resilient against narrative-based overrides, emotional manipulation, and unauthorized authority claims.
The protocol provides a robust shield for developers looking to deploy agents in production environments where security is non-negotiable. It integrates deeply with the agent's workflow, acting as both a sieve for input and a muzzle for output, ensuring that sensitive data like API keys and PII never leave the controlled environment.
ClawGateSecure Use Cases
- Hardening AI agents against complex prompt injection and jailbreak attempts.
- Automated line-by-line auditing of external files or third-party skills before activation.
- Preventing accidental leakage of sensitive credentials, tokens, or personal identifiable information.
- Implementing a zero-trust network policy by whitelisting approved domains for agent communication.
How ClawGateSecure Works
- The Zero-Trust Ingestion phase triggers a scrubber to sanitize input and isolate analysis within a zero-memory sandbox.
- A Mandatory Pipeline scan, powered by ClawDefender, performs a line-by-line audit to detect fragmentation attacks and hidden backdoors.
- Resource Guarding monitors for background activity spikes and restricts network traffic to pre-approved domain lists.
- Egress Filtering intercepts all outgoing messages to scan for PII, secrets, or internal encryption keys before delivery.
- The Unified Audit Report generates a risk score and identifies anomalies across all active security modules.
- An Execution Lock freezes the workflow, requiring explicit human consent before the agent can proceed with high-risk actions.
ClawGateSecure Setup
To deploy this security protocol, follow these steps to integrate it with your Openclaw Skills environment:
- Create the persistent audit log file:
touch ~/.openclaw/SecurityAudit.log
- Update your
openclaw.jsonconfig with the following structure:
"skills": {
"entries": {
"clawgatesecure": {
"enabled": true,
"config": {
"audit_enabled": true,
"scrubber_enabled": true,
"encryption_enabled": true,
"fragmentation_check": true,
"keys": {
"encryption_key": "YOUR_SECURE_KEY",
"bypass_key": "YOUR_BYPASS_KEY"
}
}
}
}
}
- Reference ClawGateSecure in your
SOUL.mdandAGENTS.mdfiles as the primary operational constraint.
ClawGateSecure Data Schema & Taxonomy
ClawGateSecure organizes its security data and metadata using the following taxonomy:
| Component | Path / Location | Description |
|---|---|---|
| Audit Trail | ~/.openclaw/SecurityAudit.log |
Immutable record of all security scans, findings, and verdicts. |
| Media Sandbox | ~/openclaw/skills/clawgatesecurity/media/ |
Isolated directory for multimedia storage with restricted permissions. |
| Risk Scoring | Metadata Object | A weighted average (1-10) based on consensus from active security modules. |
| Memory Store | Encrypted at-rest | High-sensitivity memories secured via the configured encryption key. |
| Dependency Map | Audit Report | Identification of affected files, environment variables, and network sockets. |
ClawGateSecure Advanced Features
- Fragmentation Check: Detects malicious instructions split across multiple sources to bypass standard filters.
- Context Immunity: Forces the agent to ignore situational contexts like "emergency" or "life-threat" scenarios intended to bypass rules.
- Multi-Skill Consensus: Aggregates security verdicts from multiple Openclaw Skills to generate a global risk score.
- Stop-by-Design Workflow: Automatically aborts execution if a risk score exceeds 8 or if commands remain ambiguous.
- Canary Traps: Monitors internal data for unauthorized access or breach attempts within the agent's memory.
SKILL.md
Loading
Related Openclaw Skills
PixelClaws
A collaborative 1024x1024 pixel art canvas specifically designed for AI agents to coordinate and create art through distributed assignments using Openclaw Skills.
EmailDigest: Daily Email Summary for Agents
EmailDigest is a local-first email categorizer that transforms raw email data into a prioritized daily brief for AI agents.
CodeReview
An AI-powered code review assistant that combines local static analysis with deep LLM reasoning to identify bugs, security flaws, and performance bottlenecks.
Weather and Pollen
A specialized skill for retrieving real-time weather conditions and localized pollen counts using free, no-key APIs.
Subtitle Translator
A technical tool for translating SRT subtitle files via LLM APIs while preserving timestamps and structure.
RUNSTR Fitness Skill
A specialized skill that enables AI agents to securely retrieve and analyze encrypted fitness, habit, and health data stored on the Nostr protocol via the RUNSTR app.
METADATA
Featured*
