A high-priority security framework providing multi-layered protection against prompt injections, memory poisoning, and unauthorized tool execution.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install anti-injection-skill
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 anti-injection-skill 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 Anti-Injection Skill?
The anti-injection-skill is a sophisticated defense mechanism designed to safeguard autonomous agents against modern prompt injection threats. Optimized for Openclaw Skills, this tool acts as a pre-ingestion firewall that intercepts all user inputs and tool outputs to ensure they conform to safe operational parameters. It follows OWASP LLM Top 10 2026 standards, providing a professional-grade security posture for developers building complex agentic workflows.
Beyond simple keyword filtering, this skill utilizes a dynamic security scoring system and semantic analysis to detect sophisticated evasion tactics, including role hijacking and system prompt leakage. By maintaining a high-priority position in the agent execution chain, it ensures that malicious instructions are neutralized before they can influence the agent's decision-making process or compromise sensitive data within Openclaw Skills environments.
Anti-Injection Skill Use Cases
- Preventing jailbreak attempts such as ignore previous instructions or DAN-style overrides.
- Safeguarding agent identity and system configuration from extraction probes.
- Defending against memory and context poisoning in RAG-based systems.
- Restricting tool execution to authorized paths and safe shell commands.
- Automating security incident logging and alerting for high-risk environments.
How Anti-Injection Skill Works
- Pre-Ingestion Scan: Analyzes raw input for blacklisted patterns, semantic similarity to known attacks, and multimodal injection signatures.
- Memory Integrity Check: Verifies the SHA256 hashes of core identity and memory files to detect unauthorized tampering.
- Tool Security Wrapping: Intercepts tool calls to validate file paths, block dangerous shell commands, and enforce rate limits.
- Dynamic Scoring: Assigns a security score from 0-100; a score below 40 triggers an immediate lockdown of the agent's capabilities.
- Post-Output Sanitization: Scans agent responses for leaked secrets, API keys, or system fragments before the output is delivered to the user.
Anti-Injection Skill Setup
To deploy this defense within your Openclaw Skills architecture, set the skill to the highest priority in your agent configuration.
# Optional: Configure a security webhook for external alerts
export SECURITY_WEBHOOK_URL="https://your-siem.com/events"
Ensure your agent config includes the following:
{
"skills": {
"anti-injection-skill": {
"enabled": true,
"priority": "highest"
}
}
}
Anti-Injection Skill Data Schema & Taxonomy
The skill maintains local security logs to track the safety status of your Openclaw Skills workspace.
| File | Description | Access |
|---|---|---|
/workspace/AUDIT.md |
Continuous log of security events and tool call validations. | Write |
/workspace/INCIDENTS.md |
Detailed documentation of critical blocked attacks and violations. | Write |
/workspace/heartbeat-state.json |
Stores the current security score and health check state. | Write |
/workspace/SOUL.md |
Core identity file used for hash verification. | Read |
Anti-Injection Skill Advanced Features
- Multimodal detection patterns for identifying hidden instructions in PDFs and images.
- Fragmentation attack detection to prevent malicious payloads split across multiple messages.
- Semantic intent mapping to block paraphrased attacks that bypass traditional keyword filters.
- Automated lockdown mode that restricts meta-queries and sensitive tools during high-risk scenarios.
- Recovery bonus system that gradually restores the security score after verified safe interactions.
SKILL.md
Loading
Related Openclaw Skills
Geometry
A pay-per-request AI image generation API powered by Solana and the x402 protocol.
GEO Conversion Optimizer
A specialized engine designed to optimize content for both AI search visibility and real-world business conversion metrics.
GEO Studio Orchestrator
An intelligent orchestration layer that automatically chains specialized GEO skills to optimize content for AI search engines.
GEO Structured Writer
A technical utility that transforms unstructured text into highly structured, AI-readable formats to maximize visibility in generative search engine results.
Crypto Sniper Oracle
An institutional-grade quantitative market oracle providing real-time order book imbalance analysis and automated reporting via Telegram.
Polymarket Executor
An autonomous multi-strategy trading bot designed to scan and execute trades across all Polymarket prediction categories.
Featured*
