AgentShield: security tool for Hermes Agent

AgentShield

aiconnai/agentshield

Offline Rust scanner for risky behavior in MCP servers and agent tools, with a Hermes Agent adapter

In short

AgentShield is an offline Rust security scanner for tool-enabled AI agents, with native adapters for MCP servers, OpenClaw skills and Hermes Agent configs. It flags risks such as command injection, credential exfiltration and SSRF before an agent can call the tools.

What AgentShield does

AgentShield normalizes 11 framework and client families into one intermediate representation: MCP, OpenClaw, Hermes Agent, CrewAI, LangChain and LangGraph, GPT Actions, Cursor Rules, Vercel AI SDK, AutoGen, LlamaIndex and Semantic Kernel. It applies 37 built-in contextual rules plus a declarative YAML rule engine read from .agentshield/rules, and tracks untrusted input across functions in Python and TypeScript through to execution sinks.

It runs as a CLI, GitHub Action, VS Code extension or Rust library, entirely offline, and emits console, JSON, SARIF and standalone HTML reports. agentshield fix applies automatic fixes for unsafe deserializers and unpinned dependencies. A runtime guard acts as a reverse proxy for MCP stdio and HTTP/SSE streams, inspecting tool calls and redacting leaked secrets. The README describes it as a complement to general SAST and secret scanning, not a replacement, and lists release 1.0.1 as General Availability.

Key features

Adapters for 11 framework families including MCP, OpenClaw and Hermes Agent
37 built-in rules plus custom YAML rules
Cross-function taint analysis for Python and TypeScript
agentshield fix for unsafe deserializers and unpinned dependencies
Runtime guard proxy for MCP stdio and HTTP/SSE streams with secret redaction
SARIF output for GitHub Code Scanning

When to use it

Scanning an MCP server before adding it to a client configuration
Checking a Hermes Agent config and tool code for risky patterns in CI
Reviewing community agent extensions for credential exfiltration or SSRF

Who it is for: Teams that ship or install tool-enabled agent extensions and want an offline pre-release security check.

How it fits with Hermes Agent

Hermes Agent configs are one of the natively supported scan targets, though the tool is a general scanner for many agent frameworks.

FAQ

What is AgentShield?

AgentShield is an offline, Rust-based security scanner for MCP servers and AI agent extensions. It finds command injection, credential theft, SSRF, unsafe file access and similar issues with static analysis.

Does AgentShield work with Hermes Agent?

Yes, Hermes Agent configs are one of its native adapters. They are checked with the same rules AgentShield applies to other frameworks, and the scan runs offline.

Is AgentShield free and open source?

The source is public, but GitHub does not report a standard license identifier. The README links LICENSE and LICENSE-MIT files, so confirm the terms there before reuse.