>_TraceKit
Run Test
TRACEKIT RUNNER/AUTONOMOUS AGENT

AI web testing that acts, verifies, and explains.

TraceKit turns a plain-language testing goal into real browser actions, deterministic verification, and evidence you can inspect.

Run New TestHow it works
TraceKit Agent Console
AGENT TELEMETRY
PERCEIVING DOM
> page.accessibility.snapshot()
• Parsed 14 interactive DOM nodes
• Found product: "Sauce Labs Backpack"
• Target identified:role="button" name="Add to cart"
Objective: verify cart increment
Locator priority: role/name > selectorTrace verified
https://www.saucedemo.com/inventory.html
Cart(0)
Sauce Labs Backpack
$29.99
Agent
DOM Observer ActiveCapturing trace...
Deterministic Verification: Standby (Armed)
LLM chose action • Playwright deterministically verified outcome
sauce_cart_verified.png
AGENT PIPELINE ARCHITECTURE

What the agent actually does.

One continuous loop: observe page state, dispatch real browser actions, verify with deterministic assertions, and explain outcomes.

TRACEKIT CORE THESIS

The Execution Loop.

The AI decides what to do. Deterministic browser assertions decide whether it actually worked.

AI Agent Reasoning
No LLM guessing in assertions100% Deterministic Outcome
LIVE TEST UNFOLDING

Watch a test execute.

From initial natural-language intent to deterministic verification and report evidence.

Execution Inspector•Step #01
TraceKit Engine v0.1
Executed Command:
tracekit run "Verify user can add an item to cart"
Natural-language test goal specified with target URL https://saucedemo.com
Execution State Metrics
Target URL:https://saucedemo.com
Goal Specification:Locate backpack, add to cart, verify cart counter
Session State:Clean session (guest checkout)
Step verified in deterministic runner
Latency: 142ms
THE TWO TESTING PARADIGMS

Two testing loops.

Comparing brittle hardcoded automation against autonomous perception paired with deterministic assertions.

TRADITIONAL SCRIPTED TESTS

Rigid Execution Loop
High Maintenance
1
Write static test script
Predefined hardcoded test paths
2
Rigid CSS/XPath selector
Tied to implementation details
3
Blind action dispatch
No dynamic DOM state perception
!
UI refactor occurs
Selector changed or DOM restructured
!
Test breaks (Flaky false-alarm)
Build failed without actual bug
!
Manual script maintenance
Engineers re-record selectors
Failure Mode: Selector drift & flakinessConstant maintenance

TRACEKIT AUTONOMOUS LOOP

Deterministic Browser Loop
AUTONOMOUS LOOPZero Selector Maintenance
1
Natural-language goal
Plain English intent: "Verify checkout"
2
AI observes live DOM state
Inspects accessibility tree & interactive elements
3
Autonomous locator synthesis
Prefers accessible role/name over brittle selectors
4
Playwright executes action
Typed CDP actions with element coordinates
5
Deterministic verification
Zero-LLM-hallucination Playwright assertions
6
Inspectable evidence + diagnosis
Screenshot, trace, and automated root cause
Resolution: Deterministic Playwright verificationSelf-adapting
DETERMINISTIC DIAGNOSTICS

A failed test shouldn't end with a mystery.

TraceKit explains exactly what changed, why the assertion failed, and attaches the visual evidence.

GENERIC TEST RUNNER OUTPUT
✕ TEST FAILED: Timeout 30000ms exceeded
> waiting for locator('#submit-order')
No explanation. Is it a bug, network lag, or a selector change?
The TraceKit Diagnostic Standard

TraceKit differentiates Application Behavior Mismatches (actual product regressions) from Environment Errors (infrastructure/network downtime).

Classification:Exact regression category
Root Cause:Plain-English summary
Visual Evidence:Deterministic screenshot
Deterministic Failure Diagnosis
Run #live-checkout-422
APPLICATION_BEHAVIOR_MISMATCHASSERTION_FAILED
Root Cause Summary: The checkout form halted because the application returned HTTP 422 Unprocessable Entity due to missing billing CVV validation. The "Place order" button remained disabled.
Expected:
button[name="Place order"] to be enabled
Observed:
aria-disabled="true" (422 response)
Evidence:
checkout-disabled.png
Network Trace:
POST /checkout → 422
Failed At:
Step #4 (Verification)
INSPECTABLE RUN ARTIFACTS

Every run produces evidence you can inspect.

Chronological traces, deterministic assertion results, and visual screenshot artifacts.

RUN #7F82•https://www.saucedemo.com
4 / 4 PASSED
EXECUTION STEPS TRACE
4 Steps
Click any step to inspect assertion state & screenshot evidence.
DETERMINISTIC VERIFICATION
Invariant Validated
Active Step:#4 VERIFY
expect(cartBadge).toHaveText('1')
Telemetry: Deterministic assertion passed: value matched invariant✓ PASSED
VISUAL EVIDENCE ARTIFACT
04_cart_badge_verified.png
saucedemo.com viewport snapshot
1280 × 800
Sauce Labs Backpack
Step #4 Invariant Confirmed
Hash: sha256:7f82b9a1Evidence verified
Inspect Full Screenshot
INITIALIZE AUTONOMOUS RUN

Describe what you want tested.TraceKit handles the browser.You inspect the result.

Run New TestDocumentation
Chromium • Playwright Engine • 100% Deterministic Assertions