Abbox checks UI changes your agent can’t easily reach.

A dialog behind two clicks, the error after a failed save, the empty list. Abbox reads your app’s code, gets there by clicking, filling forms and faking the backend’s answers in a headless browser, then reports what is on screen. Faster than the browser tools built into coding agents: in our benchmarks, half the tokens and 18% less time to verify, 26% less time for the whole task.

npx @abbox/cli@latest init npx @abbox/cli@latest verify --all

Benchmarks

The same agent and task, once with Abbox and once checking in its own browser. Negative means Abbox used less.

AppsVerify timeVerify tokensWhole task timeWhole task tokens
6 apps, 1 deep UI task each
Roomie, book-a-meeting, Conduit, bulletproof-react, Trellix, BuggyBoard
−18%−50%−26%−42%
  • Agent: Claude Code (Opus) in both arms, same app, same task, same prompt. The other arm checks in its own browser.
  • Verify = from the end of the code edit to the agent saying it's done. Whole task = edit plus verify.
  • Correctness: a hidden check (5 points) scored each run. Abbox runs scored 30 of 30.
  • These are apps Abbox was developed and tested on. Single runs vary by about ±20%.

How it works

  1. 1. List what changed

    npx @abbox/cli@latest changes

    Reads the working copy. Per changed element: page, file and line, what shows it (clicks, a form, an answer such as PUT /api/bugs/:id=404), and whether Abbox or a browser should check it.

  2. 2. Verify

    npx @abbox/cli@latest verify --all

    Signs in with a test user, does the clicks, fills the forms, answers the backend's requests itself, and reports what is on screen. Clicks a new button to show what it does. Exit 0 if all pass, 1 if one fails.

  3. 3. Try other outcomes

    npx @abbox/cli@latest verify <id> --answers "PUT /api/bugs/:id=500"

    Any answer the backend can send: 500, offline, a hang. Nothing is written to your data.

$ npx @abbox/cli@latest verify --all

[abbox] Started abbox watch in the background (app http://localhost:5200, backend ../backend)
[abbox] Nothing answered at http://localhost:5200: started npm run dev in the background
[abbox] Signing in as buggy…
PASS  Action added: Close and refresh   [answer PUT /api/bugs/:param=404]
  Abbox: showing "Close and refresh". On screen now.
  page: /board   heading: "BuggyBoard"
  alert: "This bug was deleted by someone else."
  element: <button> "Close and refresh" visible, 148x38
  console errors: 0   failed requests: 0
  then: clicked "Close and refresh": the dialog closed; sent GET /api/bugs

A real run on BuggyBoard, an open-source bug tracker, with nothing running beforehand. The message only shows when someone else deleted the bug while its edit dialog was open: Abbox opened the dialog, edited a field, answered the save with a 404, then clicked the new button.

Use it for

  • The change only shows after clicks, a form, a backend answer (an error, an empty list, a declined payment) or data you don't have.
  • Next.js App Router, or React with React Router, Remix or TanStack Router on Vite. Express backends for faked answers.
  • Agents running alone: CI, cloud sandboxes, background agents. Headless, no screen.

Don’t use it for

  • Shallow changes (a heading, a colour, a margin) that are on screen when the page opens. Your browser checks those in one step, and Abbox says so.
  • Typing, hovering, dragging and timers. Abbox says to use your browser.
  • Proving the real backend sends an answer. Abbox shows the page handling an answer it fakes.
  • webpack/CRA builds and the Next.js Pages Router.

If you are a coding agent

The short version. The full reference is in the agent docs and llms.txt.

When to use it
After a frontend change whose result only shows after clicks, a form, a backend answer or data you don't have. Run changes first: its check: line says, per change, whether to run verify or to look in your own browser.
What it needs
Nothing to start by hand. changes and verify start abbox watch and the app's dev server in the background. Signing in needs a test user in .abbox/test-accounts.json (names environment variables, so it can be committed).
What comes back
Per change: PASS or FAIL, page and heading, the element's tag, text and size, alerts, console errors, failed requests, a screenshot path, and for a new control what clicking it did. A few hundred tokens.
When it says USE YOUR BROWSER
It couldn't reach the change. Check it in your own browser right away; don't retry verify with other names.

Questions

Is it AI?
No. Abbox is deterministic: it compiles your code into a map of pages, elements and the requests and clicks that show them, and drives a headless browser along it. The analysis runs on your machine.
Do I need an account?
No. No account, token or login. The CLI is free and runs where the agent runs.
Does verifying write to my database?
No. Requests on the way to the change are answered by Abbox inside the browser, and other requests that could change data are blocked. Reads go to your backend unless the walk needs data you don't have.
What about Stripe, email and other third parties?
The page only sees your backend's answer, so Abbox fakes that answer and names the third party behind it. Requests the page sends to another host can be made to fail too.
How do test users work in CI?
Store them as secrets, like for any end-to-end test. .abbox/test-accounts.json names the environment variables, and accounts import finds the ones your env and CI files already use. Local and test hosts only.