Public White Paper v1.0

Agent Action Firewall

A runtime security architecture for protected external action release in agentic systems

No consequential agent action may execute unless its supporting memory, retrieval, policy, workflow state, evidence, authority, and receipts satisfy the integrity requirements of the action class.

Reference architecture Safe-to-act check before external mutations — live demo on Developers →