PRODUCTION/
New Test
GETTING STARTED / DEVELOPER ONBOARDING
BROWSER-NATIVE TESTING

Autonomous Web Testing in TraceKit

TraceKit translates natural-language test objectives into typed browser actions, validates application state through deterministic assertions, and produces inspectable step-by-step evidence with automated failure diagnosis.

Launch Test StudioView RunsDocumentation

Autonomous Execution Loop

4-STAGE VERIFICATION PIPELINE
01

OBSERVE

DOM & Accessibility Context
Core Capabilities
  • Accessibility tree extraction
  • Interactive element positioning
  • Captures interactive DOM node coordinates
  • 0-based disambiguation for duplicate locators
02

ACT

Typed Browser Actions
Core Capabilities
  • Click / Fill / Select / Navigate
  • Adaptive failure recovery
  • Executes typed actions via Patchright/Chromium
  • Keyboard shortcuts, options, and scroll offsets
03

VERIFY

Deterministic Assertions
Core Capabilities
  • Visible / Text / Count / URL / Title
  • Native Playwright assertion engine
  • Validates state directly against page APIs
  • Checkbox states, disabled buttons, and values
04

EXPLAIN

Failure Diagnosis & Evidence
Core Capabilities
  • Root-cause failure classification
  • Step screenshots & CDP diagnostics
  • Isolates App Bug vs Automation Failure
  • Console errors, HTTP 4xx/5xx, and trace exports

First Run

DEVELOPER ONBOARDING FLOW
STEP 1Choose Goal & Target

Provide a target URL and plain-English objective. Pick an action blueprint below or author a custom flow in Test Studio.

STEP 2Autonomous Execution

Chromium runs in headed or headless mode, explores the live DOM tree, dispatches typed interactions, and evaluates deterministic checks.

STEP 3Inspect Evidence

Open Run Details to review chronological step screenshots, CDP console diagnostics, and automated root-cause failure classifications.

Quick Launch Blueprints

3 STARTER JOURNEYS
Also selectable via 1-click “Starter Templates” inside Test Studio
BP-01Auth Flow

Auth & session redirect

Any web application with login forms and authenticated routes
Agent Objective

“Navigate to /dashboard without authentication, verify redirect to /login, fill in user credentials, submit the form, and verify successful redirect to the dashboard.”

Verification: Playwright url_contains + DOM visibility assertions
Launch in Test Studio
BP-02E-Commerce

Cart checkout flow

E-commerce storefront catalog with search, filter, and cart
Agent Objective

“Search for "Technical Shell Jacket", apply the sizing filter "XL", add product to cart, proceed to checkout page, and ensure the price calculation includes zero shipping fees.”

Verification: Element has_text, input has_value, and checkout button enabled
Launch in Test Studio
BP-03Site Audit

Global links & 404s

Production, staging, or preview documentation and landing pages
Agent Objective

“Inspect header navigation links, verify primary landing pages load with HTTP 200, and ensure no broken links or critical console errors are reported.”

Verification: HTTP 200 status assertions + CDP console & page error diagnostics
Launch in Test Studio

What TraceKit Trusts

ENGINE TRUST BOUNDARY
The agent interacts; deterministic assertions determine the test verdict.
REASONING

AGENT DECISION LOOP

The reasoning provider plans step-by-step navigation and resolves dynamic locators, but is never trusted to evaluate test success or failure.

Reasoning Responsibilities
  • Determines what action to attempt next (click, fill, select, scroll)
  • Extracts interactive DOM & accessibility tree structure
  • Resolves locator ambiguity with 0-based index disambiguation
  • Produces decision rationale and adapts to transient element states
VERIFICATION

DETERMINISTIC PLAYWRIGHT CHECKS

Assertions are executed directly in browser engine context. A test passes only if all deterministic checks succeed, eliminating hallucinated passes.

Native Assertion Protocols
visible
has_text
not_visible
has_count
url_contains
title_contains
Core Trust Invariant:LLMs choose the interaction path; native browser assertions prove the verdict.
Zero Hallucinated Passes

Workflow Deep-Dive

5-PHASE RUNBOOK
01

Define your test

Phase 01

Give TraceKit a URL and describe what you want verified.

  • Enter any web URL reachable from your environment (local or production).
  • Write your test goal in plain English (e.g. "Log in as standard_user, add the backpack to the cart, and verify the checkout button is enabled").
  • TraceKit translates your objective into an autonomous Observe → Reason → Act loop.
02

Configure your environment

Phase 02

Choose browser settings, authentication state, AI provider, and model.

  • Select Headless or Headed browser execution with Chromium.
  • Inject authenticated session state (storage_state JSON) to test post-login user flows without re-authenticating every run.
  • Pick your reasoning provider: Auto (Fallback), Google Gemini, Groq Cloud, or local Ollama.
  • Optionally specify custom model overrides or set maximum execution step budgets.
03

Run

Phase 03

TraceKit observes the website and performs browser actions.

  • The agent captures DOM accessibility snapshots and interactive element positions.
  • Actions executed include click, type, keypress, option select, scroll, and hover.
  • Multi-element disambiguation with 0-based indices prevents accidental clicks on duplicate locators.
  • Resilient failure recovery keeps tests moving forward if initial interactions require adjustment.
04

Verify

Phase 04

Deterministic assertions confirm whether the expected state was reached.

  • Deterministic assertions verify conditions directly against Playwright / Chromium APIs.
  • Supported checks include element visible, hidden, text content, input value, page URL, and page title.
  • Rich state assertions check whether checkboxes/toggles are checked or unchecked, buttons are enabled or disabled, and match element counts.
05

Review

Phase 05

Inspect actions, screenshots, diagnosis, and the generated report.

  • Examine the complete chronological step-by-step decision trace.
  • View visual screenshot evidence captured at each interaction point.
  • If a run fails, deterministic failure diagnosis isolates the root cause (e.g. ASSERTION_FAILED, LOCATOR_NOT_FOUND, BUDGET_EXCEEDED, PROVIDER_ERROR).
  • Download or export complete structured JSON and Markdown test reports.
START TESTING

Ready to execute your first autonomous test?

Launch an autonomous Chromium session and verify your application state in seconds.

Open Test StudioView Execution Runs