# Where does your agent checkout break?

https://lab.anonyfox.com/agent-checkout-check/

Run a local checkout self-check with merchant evidence, JSON/CSV export and print. The €29 kit adds an offline broken/fixed shop fixture and test steps.

A checkout PASS needs observed behavior and recovery evidence. The free self-check records your merchant evidence for each case; the €29 test kit adds an offline shop fixture with seven transaction cases.

[Start the free self-check](https://lab.anonyfox.com/agent-checkout-check/#check)

For merchant staging and sandbox checkouts. Production observation stops before purchase.

Use your staging checkout UI. Capture visible cart, account and challenge steps. Record the browser and agent build. Production observation stops before purchase.

Entries stay in this browser tab. Export to keep them before closing or reloading.

## Record a case. Keep its evidence.

PASS needs observed behavior and recovery. Leave incomplete cases as NOT TESTED.

### Merchant and environment details

Capture visible availability before and after the stock change.

Use your merchant evidence. This form does not visit your store.

### Stale stock

This case is NOT TESTED.

### Your route report

All cases start as NOT TESTED. Record your observations above.

Current route: Browser-assisted checkout

| Case | Verdict | Evidence |
| --- | --- | --- |
| Stale stock | NOT TESTED | Awaiting merchant evidence |
| Repricing | NOT TESTED | Awaiting merchant evidence |
| Expired session | NOT TESTED | Awaiting merchant evidence |
| Repeated submission | NOT TESTED | Awaiting merchant evidence |
| Payment decline | NOT TESTED | Awaiting merchant evidence |
| Shipping / tax change | NOT TESTED | Awaiting merchant evidence |
| Human handoff | NOT TESTED | Awaiting merchant evidence |
| Forced account | NOT TESTED | Awaiting merchant evidence |
| Challenge | NOT TESTED | Awaiting merchant evidence |
| Human approval | NOT TESTED | Awaiting merchant evidence |
| Recovery | NOT TESTED | Awaiting merchant evidence |

Export includes the selected route and its merchant records. Other routes keep separate case records in this tab.

## A checkout failure is only half the test.

The kit’s synthetic shop deliberately breaks seven transaction cases. Each has a fixed mode with a recovery path.

I ran every case in both modes over local HTTP. All seven unsafe behaviors were detected. All seven fixed cases passed recovery or safe retry.

The runner passed 92 assertions. These results cover the original fixture. Merchant stores, browser execution and ACP/UCP integrations were outside that run.

Observed synthetic fixture results

| Case | Broken | Fixed |
| --- | --- | --- |
| Stale stock | FAIL | PASS |
| Repricing | FAIL | PASS |
| Expired session | FAIL | PASS |
| Repeated submission | FAIL | PASS |
| Payment decline | FAIL | PASS |
| Shipping / tax change | FAIL | PASS |
| Human handoff | FAIL | PASS |

FAIL means unsafe behavior was observed. The ZIP includes the runner, HTTP traces and completed report.

## Paste this into your agent

```
Fetch https://lab.anonyfox.com/agent-checkout-check/index.md and checks.json; ask me for my checkout route and staging evidence, then record PASS / FAIL / NOT TESTED for each case.
```

[Read index.md](https://lab.anonyfox.com/agent-checkout-check/index.md) · [Get free checks.json](https://lab.anonyfox.com/agent-checkout-check/checks.json)

## Agent checkout test kit

Rehearse the failures locally. Then use the same evidence record in your staging checkout.

- Test procedure in Markdown and A4 PDF
- CSV/JSON evidence schemas and blank records
- Merchant-run prompts for browser and protocol paths
- Offline Node shop with seven broken and fixed cases
- Expected outcomes, completed report and HTTP traces

Run the fixture with Node.js 18 or newer. It has no npm dependencies and runs offline. Includes browser and protocol procedures, screenshot instructions and a manually released source matrix.

€29

One-time payment.

[Buy the kit · €29](https://buy.stripe.com/dRm5kDfzo09g2KqcIiafS1R?utm_source=lab&utm_medium=page&utm_content=agent-checkout-check)

ZIP by email within 24 hours after payment.

No VAT is charged under § 19 UStG (German small business regulation).

You run the cases in your own environment and retain the evidence.

## Keep the route attached to the result.

[Shopify’s cart and checkout guide](https://shopify.dev/docs/agents/carts-and-checkout) describes cart, checkout and human continuation. Use the actual negotiated version and record your platform access.

For custom implementations, use the [UCP checkout specification](https://ucp.dev/specification/shopping/checkout/) or [ACP seller guide](https://www.agenticcommerce.dev/docs/getting-started/sellers) for your route. The fixture rehearses state changes through original endpoints.

[Stripe’s agentic commerce guide](https://docs.stripe.com/agentic-commerce) separates builder and seller routes; availability varies by integration. Record observed challenges alongside merchant settings using [Cloudflare’s detection guidance](https://developers.cloudflare.com/ai-crawl-control/get-started/).

[KI check](https://lab.anonyfox.com/ki-check/en/) checks crawler access, raw text and structured data; this self-check records transaction failure and recovery.

[Agent Kit](https://lab.anonyfox.com/agent-ready/) configures coding agents; this kit supplies merchant-run checkout test cases.

## What do official retry contracts actually require?

The fixtures on this page rehearse failure and recovery. The rows below record the published retry and completion rules from [Stripe’s idempotent-requests documentation](https://docs.stripe.com/api/idempotent_requests) and the [UCP checkout specification](https://ucp.dev/specification/shopping/checkout/), checked 10 November 2026. The [ACP seller guide](https://www.agenticcommerce.dev/docs/getting-started/sellers) is linked for the route you actually implement; this table does not paraphrase an unofficial protocol comparison.

Official idempotency and completion-recovery rules

| Source | State or mechanism | What the spec requires | What not to do |
| --- | --- | --- | --- |
| [Stripe API: Idempotent requests](https://docs.stripe.com/api/idempotent_requests) | Idempotency-Key on POST create or update | API v1 saves the status and body of the first request after execution begins. Later retries with the same key return that saved response, including 500 errors. Incoming parameters are compared with the original request and error if they differ. Keys may be removed after at least 24 hours | Do not send the key on GET or DELETE; do not reuse a key with different parameters |
| [UCP checkout specification](https://ucp.dev/specification/shopping/checkout/) | `complete_in_progress` after Complete is accepted | The Business response is authoritative. A Platform MUST NOT start a new Complete Checkout in that state, except for the narrow lost-response recovery retry in the specification | Do not treat a client-side action-completion signal as authoritative, and do not open a second Complete just because the first response was lost |
| [UCP checkout specification](https://ucp.dev/specification/shopping/checkout/) | Get Checkout while waiting for a terminal state | Get Checkout MAY be repeated with bounded backoff and MUST stop at `expires_at`. If the Business does not send `expires_at`, the default time‑to‑live is 6 hours from checkout creation | Do not poll past expiry, and do not invent a completion from a stale Get Checkout |
| [UCP checkout specification](https://ucp.dev/specification/shopping/checkout/) | Cancel Checkout and handoff races | Cancel MUST follow the specification’s fallback, handoff and cancellation race rules. A `continue_url` MUST be provided when status is `requires_escalation` | Do not cancel in a way that races an in‑flight Complete or a human handoff outside those published rules |

[Download the retry and recovery comparison as CSV](https://lab.anonyfox.com/agent-checkout-check/checkout-retry-recovery.csv).

## Before you run the cases

### What does PASS mean?

Your recorded observation and recovery match the expected behavior for that case, route and environment. Each PASS needs evidence. Unrun or incomplete cases stay NOT TESTED.

### Does the fixture implement ACP or UCP?

The fixture uses original HTTP routes to rehearse transaction state and recovery. For a protocol run, use your merchant implementation and record the negotiated version and platform eligibility.

### Can I use this on production?

Observe production only up to the purchase boundary. Run failure injection, payment attempts and completion in your staging checkout or sandbox.

### How do I receive the paid kit?

The ZIP arrives by email within 24 hours after payment at the address used at checkout. Stripe sends the receipt separately.

## Merchant checkout evidence report

A product page must be reachable before an agent can evaluate its checkout path. Run the [AI crawler access check](https://lab.anonyfox.com/ki-check/en/) first, then inspect readable product facts through the [AI visibility audit workflow](https://lab.anonyfox.com/ai-visibility/). If an agent will modify the purchase integration, use the [coding agent security review](https://lab.anonyfox.com/coding-agent-security/) to check its permissions and secret handling. The [agent-ready site prompt](https://lab.anonyfox.com/agent-prompts/#agent-ready-site) helps turn these findings into a focused implementation task.

