Shadow Execution Kit

Run new agent logic in shadow mode alongside production without affecting real outcomes.

Configuration

Shadow Execution Dashboard

Shadow Runs

0/6

Divergences

50%

Improvements

6%

Refund request #4821
Prod: ApproveShadow: ApproveMatch
Account suspension #9102
Prod: WarnShadow: SuspendDiverge
Priority escalation #3341
Prod: LowShadow: HighDiverge
Spam detection #7720
Prod: BlockShadow: BlockMatch
Credit limit #5512
Prod: $500Shadow: $750Diverge
Fraud alert #8834
Prod: FlagShadow: FlagMatch

Integration Code

import { createShadowExecution } from 'agent-tools-kit/experimentation'

const shadow = createShadowExecution({ mode: 'full-mirror' })
shadow.attach(productionAgent, experimentalAgent)

// Run shadow comparison on live traffic
const results = await shadow.compare({
  sampleSize: 1000,
  onDivergence: (prod, exp) => {
    console.log('Divergence:', prod.decision, '→', exp.decision)
  },
})

console.log('Match rate:', results.matchRate)
console.log('Improvement potential:', results.improvementRate)