~4 min to deploy

LangGraph Plan → Execute Gate

Drop a gate node between plan and execute so the graph's conditional edge branches on the verdict instead of a guess.

For: JS / Node agent builders running multi-step LangGraph workflows

SOC 2 CC7.2ISO 27001 A.8.9Agent action evidence

What it enforces

Step blast-radius gate

Escalates a planned step whose blast radius exceeds the encoded ceiling.

Out-of-scope step block

Blocks a step that falls outside the agent's declared scope.

Human-in-the-loop hold

Escalates when the plan itself flags that a human approval is required.

Deploy in Shadow Mode

langgraph-plan-execute-gate.yaml

# LangGraph Plan → Execute Gate
# Fork: place decionisGateNode between plan and execute; the conditional edge
# branches on state.decionis.outcome (allowed / blocked / errored).
apiVersion: decionis.dev/v1
kind: PolicyPack
metadata:
  name: langgraph-plan-execute-gate
  surface: langchain
  workflow_key: agent_plan_execution
  standards: [SOC2-CC7.2, ISO27001-A.8.9]
defaults:
  mode: shadow
  emit_dossier: true
rules:
  - name: step_blast_radius_gate
    when: "node == 'execute'"
    decision: |
      ALLOW IF step.risk_score < 60
      ESCALATE IF step.risk_score < 85
      BLOCK OTHERWISE
    reason_code: step_risk_over_threshold
  - name: out_of_scope_step_block
    when: "node == 'execute'"
    decision: |
      BLOCK IF outside_scope == true
      ALLOW OTHERWISE
    reason_code: step_outside_agent_scope
  - name: human_in_the_loop_hold
    when: "node == 'execute'"
    decision: |
      ESCALATE IF requires_human == true
      ALLOW OTHERWISE
    reason_code: plan_requires_human_approval

Fork it, change the thresholds to match your environment, and deploy in shadow mode first — it defaults to listen-only so nothing in your live pipeline changes.

How to install it

Same wrap-and-forget pattern as the Python version, for Node-land agents.

LangChain.js — shadow mode

// npm i @decionis/langchain @decionis/sdk-node
import { createDecionisNodeSdk } from "@decionis/sdk-node";
import { DecionisGateTool } from "@decionis/langchain";

const client = createDecionisNodeSdk({
  baseUrl: "https://api.decionis.com",
  apiKey:  process.env.DECIONIS_API_KEY!,
});

const gatedRefund = DecionisGateTool.wrap({
  innerTool: sendRefund,
  client,
  orgId: process.env.DECIONIS_ORG_ID!,
  decisionType: "refund_execution",
  shadowMode: true,         // ← every verdict recorded; inner tool always runs
  siteBaseUrl: "https://decionis.com",
});
await agent.bindTools([gatedRefund]);

Install @decionis/langchainShips in shadow mode — every verdict is recorded, nothing is blocked.

Loading…

Open-Policy Exchange