The diff of what your code did.

Run your tests after each agent turn. See what each test did differently.

$ npx diffew

~/shop/agent-a

# agent: refactor loadProducts

diffew run shop -- pnpm test:e2e

diffew run 20261019T093304Z: shop/desktop

service shop node :3000 recording

✓ cart.e2e.test.ts (5 tests) 341ms

✓ checkout.e2e.test.ts (9 tests) 574ms

Tests 14 passed (14)

diffew compare shop

shop agent/checkout b71d03f → 0c5e8a1 paired with the previous run

14 tests 1 changed "checkout > places an order"

1 server call changed loadProducts postgres 1 → 12 statements

http://localhost:4400/w/3f9a1c07e2/api/shop

diffew1 test changed

checkout > places an order

52 → 136 ms

POST /orders201

createOrder

select carts

loadProducts

~select products×12

insert orders

insert order_items

publish order.created

Don’t read the code diff. Read what it did.

claude/checkout-refactor

src/checkout/PaymentStep.tsx+20 −16

@@ -1,34 +1,38 @@

1-import { useState } from "react"

2-import { Button } from "@/components/ui/button"

3-import { PromoCodeInput } from "./PromoCodeInput"

4-import { formatPrice } from "@/lib/format"

1+import { Button } from "@/components/ui/button"

2+import { ExpressCheckoutButton } from "./ExpressCheckoutButton"

3+import { formatMoney } from "@/lib/money"

4+import { useT } from "@/i18n"

55 import type { Order } from "@/types"

7-type Props = { order: Order; onSubmit: () => void; loading?: boolean }

7+interface PaymentStepProps {

8+ order: Order

9+ onPlaceOrder: () => Promise<void>

10+ isSubmitting?: boolean

11+}

812

9-export function PaymentStep({ order, onSubmit, loading }: Props) {

10- const [promo, setPromo] = useState("")

13+export function PaymentStep({ order, onPlaceOrder, isSubmitting = false }: PaymentStepProps) {

14+ const t = useT()

1115 return (

12- <section className="payment-step flex flex-col gap-4">

13- <h2 className="text-lg font-semibold">Order summary</h2>

16+ <section className={cn("payment-step", "grid gap-3")}>

17+ <h2 className="payment-step__title">{t("checkout.summary")}</h2>

1418 <ul className="payment-step__lines">

1519 {order.lines.map((line) => (

16- <li key={line.id} className="flex justify-between">

20+ <li key={line.sku} className="payment-step__line">

1721 <span>{line.title}</span>

18- <span>{formatPrice(line.amount)}</span>

22+ <span>{formatMoney(line.amount, order.currency)}</span>

1923 </li>

2024 ))}

2125 </ul>

22- <p className="flex justify-between font-semibold">

26+ <p className="payment-step__total">

2327 <span>{t("checkout.total")}</span>

24- <span>{formatPrice(order.total)}</span>

28+ <span>{formatMoney(order.total, order.currency)}</span>

2529 </p>

26- <PromoCodeInput value={promo} onChange={setPromo} />

2730 <div className="payment-step__actions">

31+ <ExpressCheckoutButton amount={order.total} onApprove={onPlaceOrder} />

28- <Button className="w-full" disabled={loading} onClick={onSubmit}>

32+ <Button size="lg" disabled={isSubmitting} onClick={onPlaceOrder}>

29- Pay now

33+ Place order

3034 </Button>

3135 </div>

3236 </section>

3337 )

3438 }

src/checkout/payment-step.css+9 −3

@@ -8,8 +8,14 @@

88 .payment-step__lines {

9- margin: 0;

9+ margin-block: var(--space-2);

1010 padding: 0;

1111 }

1212

13-.payment-step__actions { margin-top: 16px; }

13+.payment-step__actions {

14+ display: grid;

15+ gap: var(--space-2);

16+ margin-top: var(--space-4);

17+}

1418

15-.payment-step__total { font-weight: 600; }

19+.payment-step__total {

20+ font-weight: 600;

21+}

src/checkout/PaymentStep.tsx+20 −16

@@ -1,34 +1,38 @@

1-import { useState } from "react"

2-import { Button } from "@/components/ui/button"

3-import { PromoCodeInput } from "./PromoCodeInput"

4-import { formatPrice } from "@/lib/format"

1+import { Button } from "@/components/ui/button"

2+import { ExpressCheckoutButton } from "./ExpressCheckoutButton"

3+import { formatMoney } from "@/lib/money"

4+import { useT } from "@/i18n"

55 import type { Order } from "@/types"

7-type Props = { order: Order; onSubmit: () => void; loading?: boolean }

7+interface PaymentStepProps {

8+ order: Order

9+ onPlaceOrder: () => Promise<void>

10+ isSubmitting?: boolean

11+}

812

9-export function PaymentStep({ order, onSubmit, loading }: Props) {

10- const [promo, setPromo] = useState("")

13+export function PaymentStep({ order, onPlaceOrder, isSubmitting = false }: PaymentStepProps) {

14+ const t = useT()

1115 return (

12- <section className="payment-step flex flex-col gap-4">

13- <h2 className="text-lg font-semibold">Order summary</h2>

16+ <section className={cn("payment-step", "grid gap-3")}>

17+ <h2 className="payment-step__title">{t("checkout.summary")}</h2>

1418 <ul className="payment-step__lines">

1519 {order.lines.map((line) => (

16- <li key={line.id} className="flex justify-between">

20+ <li key={line.sku} className="payment-step__line">

1721 <span>{line.title}</span>

18- <span>{formatPrice(line.amount)}</span>

22+ <span>{formatMoney(line.amount, order.currency)}</span>

1923 </li>

2024 ))}

2125 </ul>

22- <p className="flex justify-between font-semibold">

26+ <p className="payment-step__total">

2327 <span>{t("checkout.total")}</span>

24- <span>{formatPrice(order.total)}</span>

28+ <span>{formatMoney(order.total, order.currency)}</span>

2529 </p>

26- <PromoCodeInput value={promo} onChange={setPromo} />

2730 <div className="payment-step__actions">

31+ <ExpressCheckoutButton amount={order.total} onApprove={onPlaceOrder} />

28- <Button className="w-full" disabled={loading} onClick={onSubmit}>

32+ <Button size="lg" disabled={isSubmitting} onClick={onPlaceOrder}>

29- Pay now

33+ Place order

3034 </Button>

3135 </div>

3236 </section>

3337 )

3438 }

src/checkout/payment-step.css+9 −3

@@ -8,8 +8,14 @@

88 .payment-step__lines {

9- margin: 0;

9+ margin-block: var(--space-2);

1010 padding: 0;

1111 }

1212

13-.payment-step__actions { margin-top: 16px; }

13+.payment-step__actions {

14+ display: grid;

15+ gap: var(--space-2);

16+ margin-top: var(--space-4);

17+}

1418

15-.payment-step__total { font-weight: 600; }

19+.payment-step__total {

20+ font-weight: 600;

21+}

ScreenFunctions

shopCartAccount

Order summary

Desk lamp€49.00

Bulb, warm white ×2€12.00

Total€61.00

Promo code

Express checkoutPay nowPlace order

~ Changedbutton “Place order”text “Pay now” → “Place order”

− Removedinput[textbox] “Promo code”

+ Addedbutton “Express checkout”

Test time171→186msslower

~PaymentStep
−PromoCodeInput
+ExpressCheckoutButton
+createExpressSession
~onPlaceOrder
createOrder
POST /orders201

What it flags

An N+1 query

postgres

checkout › places an order

Test time52→136msslower

createOrder
loadProducts
~select products×1 → ×12
insert orders

A new external call

http

checkout › places an order

Test time136→256msslower

createOrder
insert orders
+chargeCard
+POST pay.dev/charge

A slower query

postgres

orders › lists orders

Test time18→352msslower

listOrders
select orders
toOrderRows

A message that is gone

kafka

checkout › places an order

Test time136→128msfaster

createOrder
insert orders
insert order_items
−Kafka produce order.created

A screen changed

checkout › paymentswipe

shopCartAccount

Order summary

Desk lamp€49.00

Bulb, warm white ×2€12.00

Total€61.00

Promo code

Pay now

shopCartAccount

Order summary

Desk lamp€49.00

Bulb, warm white ×2€12.00

Total€61.00

Promo code

Express checkoutPlace order

~ Changedbutton “Place order”text “Pay now” → “Place order”

− Removedinput[textbox] “Promo code”

+ Addedbutton “Express checkout”

A comment on each pull request

What each test did differently, right on the pull request. Cloud

github.com/acme/shop/pull/482

Refactor checkout into createOrder, add card payments #482

Openclaude wants to merge 3 commits into main

claude commentedAuthor
Summary
Moves order creation into createOrder and adds card payments.
- loadProducts takes the cart items.
- New chargeCard client for the payments API.
diffew bot commented
Behaviour changed in 2 of 14 tests against main
Test What changed
checkout › places an order ~12 sequential select products (was 1) · +84 ms
checkout › pays by card +1 HTTP call POST /charge · +120 ms
Suggestion
loadProducts now loads the products one by one inside the loop. Load them in one where id in (…) query.
View the diff →

Test	What changed
`checkout › places an order`	~12 sequential `select products` (was 1) · +84 ms
`checkout › pays by card`	+1 HTTP call `POST /charge` · +120 ms

Agents check their own work.

The same facts, as text. The agent fixes what it did not mean to change.

claude — ~/shop

> add card payments to checkout

⏺Update(src/orders/checkout.ts)

⎿Updated src/orders/checkout.ts with 24 additions and 19 removals

⏺Bash(npx diffew run tests --section api --label "turn 2" -- pnpm test && npx diffew compare tests)

⎿tests claude/checkout-refactor b71d03f → 0c5e8a1 (dirty) paired with the previous run

14 tests 2 changed "checkout > places an order" "checkout > pays by card"

2 server calls changed

loadProducts 4 → 41 ms postgres 1 → 12 statements

chargeCard added

http://localhost:4400/w/3f9a1c07e2/api/tests

⏺The diff shows loadProducts is 10× slower, 4 → 41 ms. It now runs one query per item, 12 instead of 1: an N+1. I'll load the products in one query.

⏺Update(src/catalog/products.ts)

⎿Updated src/catalog/products.ts with 1 addition and 2 removals

12- for (const item of items) rows.push(await db.select().from(products).where(eq(products.id, item.productId)))

13- return rows

12+ return db.select().from(products).where(inArray(products.id, items.map((i) => i.productId)))

⏺Bash(npx diffew run tests --section api --label "turn 3" -- pnpm test && npx diffew compare tests --against main)

⎿tests claude/checkout-refactor 4e1a2c9 → 0c5e8a1 (dirty) paired against the base branch

14 tests 1 changed "checkout > pays by card"

1 server call changed chargeCard added

http://localhost:4400/w/3f9a1c07e2/api/tests

⏺Verified: the only new behaviour is the card charge.

Open source

Runs on your machine

CLI and desktop app. Apache-2.0.

$ npx diffew

Star on GitHubmacOS · soon

Hosted by you

Freein CI, on every run
4runners: Vitest, Jest, node:test, Playwright
Localruns stay in your repository
Per turneach worktree, each agent

Cloud soon

Hosted by us

For teams and pull requests.

Everything in open source, plus

Hosted runs, baselines from main
A comment and a check on each pull request
Diffs you share with a link
Trends per test over time

The diff of what your code did.

Don’t read the code diff. Read what it did.

What it flags

A comment on each pull request

Summary

Agents check their own work.

Open source and cloud

Runs on your machine

Hosted by us