Shellward
Protects AI agents from prompt injection, data exfiltration, and dangerous command execution by acting as an LLM security middleware and AI agent firewall.
Shellward is an AI Agent Security Middleware designed to safeguard AI agents with full access to tools like the shell, email, HTTP, or file system. It intercepts tool calls at runtime to enforce agent guardrails, preventing critical threats such as prompt injection, sensitive data exfiltration, and malicious command execution. With an 8-layer defense-in-depth approach and DLP-style data flow control, Shellward ensures that data remains internal, providing robust protection without complex configurations or external dependencies, available as a standalone SDK or OpenClaw plugin.
Key Features
01DLP model that allows full data access internally but blocks outbound sends if PII was recently accessed.
028-layer defense including prompt guard, tool blocker, and data flow guard for comprehensive protection.
03PII detection for SSN, credit cards, API keys, JWT, passwords, and Chinese ID/phone/bank cards.
0448 GitHub stars
05Zero dependencies and zero configuration, offering simple integration as an SDK or MCP server/OpenClaw plugin.
0632 prompt injection rules (18 Chinese + 14 English) with risk scoring and mixed-language detection.
Use Cases
01Protecting against data exfiltration chains and dangerous command execution in AI agent environments.
02Enforcing runtime security for AI agents in autonomous coding assistants, chatbots, and internal automation.
03Intercepting and filtering tool calls to add agent guardrails to any LLM-powered workflow.