vibe-testing v0.4.5 Run your first check

Release review / browser evidence / memory

Your coding agent shipped it. Now test what changed.

vibe-testing reads the routes and forms in your codebase, tests them in Playwright, and remembers the result. The next run tells you which fix held and which route regressed.

npx vibe-testing@latest init
Open on npm Read the source
13
testing tools
0
internal LLM calls
3
ways to run
QA-048 / EXAMPLE REPORT

Release verdict

BLOCKED
1 regression
project atlas-web commit 31da2e7 duration 42.8s
Changes since last runsnapshot_diff
FIXED /login redirect and session verified
REGRESSION /checkout payment request returned 500
NEW /account/security route found, coverage missing
Blocking evidence01

POST /api/checkout

Expected 200, observed 500

consolePaymentError: price_id missing
ScreenshotSTEP 04
/checkout
Complete orderWe could not process this payment.Try again or use another method.
OUTCOME VERIFIED 6 / 8 API ERRORS 1 CONSOLE ERRORS 1

The report is the product

A release decision with receipts.

Each run leaves a self-contained HTML report with route coverage, browser screenshots, observed API calls, console errors, and the steps needed to reproduce a failure.

Blocking issue

Checkout fails before order creation

P0
Route/checkout
Action

Submit a valid card

Expected

Order confirmation appears

Observed

Payment API returned 500

  1. 01

    Open /pricing and choose the Pro plan.

    PASS
  2. 02

    Enter the account and billing details found in the form schema.

    PASS
  3. 03

    Submit payment and wait for the order request.

    FAIL
Coverage gap

The new /account/security route has no completed scenario.

What the browser remembers

Your agent gets continuity between runs.

A

It reads the app before clicking.

Routes, forms, field names, existing tests, and coverage gaps come from the codebase. The agent starts with the product you built.

scan_codebase, then get_context
B

It keeps a regression ledger.

Run snapshots preserve pass and fail state by route. The next run separates fixes, regressions, known failures, and new routes.

snapshot_diff + route_changes
C

It returns evidence to the agent.

Step logs, screenshots, network observations, and verification status let the coding agent repair the failure and rerun the smallest scenario.

execute_scenario, then generate_report

The runner makes zero internal LLM calls. Your editor's model decides what to test. vibe-testing handles the browser, deterministic checks, memory, and report.

Thirteen tools, one QA workflow

Use one check or run the whole release.

The MCP server exposes each stage, so an agent can inspect context, test one route, examine evidence, fix the code, and rerun the affected flow.

Understand

scan_codebaseMap routes, forms, tests, and changes.
get_contextRead the source behind one feature.
get_coverageSee tested routes and open gaps.
suggest_testsTurn gaps into runnable scenarios.

Exercise

loginEnter authenticated test state.
scan_page_elementsInventory interactive controls.
explore_pageProbe broad page behavior.
execute_scenarioRun exact steps with evidence.
take_screenshotCapture a visual checkpoint.

Decide

generate_reportBuild the reviewable HTML dossier.
run_full_testScan, execute, compare, and report.
run_convergeRepeat until the threshold is met.
cleanupClose browsers and free resources.

Pick your entry point

Give your agent a QA shift.

Node 20 or newer. MIT licensed. The first run installs the matching Playwright Chromium build.

FastestCLI + editor setup

Detect installed editors, register the MCP server, and create project guidance.

npx vibe-testing@latest init
DirectOne-off browser run

Run against a local or deployed app without an editor integration.

npx vibe-testing@latest run http://localhost:3000
PortableAgent Plugin

Load the repository to get the MCP server and the senior-QA release skill together.

claude plugin marketplace add AishwaryShrivastav/vibe-testing
Works withClaude CodeCursorWindsurfVS CodeRoo CodeCline

Founding QA runs

Bring one app you are about to ship.

I am working with the first ten teams directly. We will install vibe-testing, run a release check, and use the report to improve the product. Thirty minutes.

Request a founding QA run