n8n - 💡(How to fix) Fix Community Node: AgentShield — security check node for AI Agent workflows [2 comments, 3 participants]

Summary

We've built a community node that adds runtime security checks to n8n workflows, specifically designed for AI Agent tool call scenarios.

npm: n8n-nodes-agentshield (pending publish) Source: github.com/hidearmoon/agentshield/packages/integrations/n8n

How it works

The node acts as a security gate with two outputs: Allowed and Blocked.

[Trigger] → [AgentShield] → Allowed → [Send Email]
                           → Blocked → [Notify Admin]

For AI Agent workflows, place it between the Agent node and tool nodes to intercept tool calls before execution.

Each item is checked against the AgentShield core engine, which evaluates:

• Trust level of the data source
• Intent consistency (does this tool call match what the agent is supposed to do?)
• 22 built-in security rules (injection defense, data exfiltration, privilege escalation)

Node features

• Dual output routing (Allowed / Blocked)
• Credential management via n8n's credential system
• Fail-open mode (doesn't break workflows if security service is down)
• Decision metadata attached to each item (_agentshield.decision, _agentshield.reason, _agentshield.trace_id)

Use case

n8n's AI Agent node is increasingly used to give agents access to real tools (HTTP requests, database queries, code execution). A single prompt injection in input data could trick the agent into misusing these tools. This node provides a security checkpoint.

Would love feedback from the n8n team. Happy to adjust the node to match n8n's community node standards.