Agent Action Gateway

Govern proposed agent actions before execution

Neura Relay turns proposed agent actions into Decision Receipts before execution.Developer-owned runtimes keep execution control after reading the receipt.

Run the Gateway proof ladder
git clone https://github.com/neurarelay/relay-action-card.git
cd relay-action-card
npm ci
npm run proof:agent-action-firewall -- --dry-run --json
npm run verify:decision-receipt-standard
npm run proof:mcp-risk-gate -- --dry-run --json
npm run proof:commerceops-fire-drill -- --dry-run --json

Gateway Path

Agent intent becomes a receipt before action

01

Agent wants to act

Runtime prepares the proposed actionbefore anything downstream changes

02

Action Card

Intent, actor, target, and policyevidence and risk are made explicit

03

Agent Action Gateway

Relay classifies the proposed actionand records the authority decision

04

Decision Receipt

System reads the receipt outcomethen routes execution downstream

Proof Ladder

Four local proofs, one Relay mechanism

Boundaries

The proof is deliberately narrow

  • Neura Relay remains the product; Agent Action Gateway is the proof package
  • Gateway input is an Action Card; Gateway output is a Decision Receipt
  • Developer-owned runtimes still execute, revise, escalate, or stop actions
  • Synthetic proofs avoid real stores, payments, customer records, and systems
  • No provider, certification, marketplace, partnership, or endorsement claim
  • No public tokens, private payload exposure, Registry auto-approval, or Neura execution

What to measure

Measure route visits, proof-doc clicks, command copies, local dry-runs, live receipt refs, and qualified replies using Agent Action Gateway language

route visitproof runreceipt ref